diff --git a/.gitattributes b/.gitattributes index 29b6488c631dadc6649bb3df8d119b5bd280729a..b19354579d2915c271864dc1a2acc98a5a8d6b49 100644 --- a/.gitattributes +++ b/.gitattributes @@ -5117,3 +5117,9 @@ lambda0.007/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample32-layer4 lambda0.007/hyperprior-featurecoding-8bit-individual/fc_arc_challenge/sample43-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample495-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.007/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample44-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample412-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample855-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample86-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample969-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample17-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text diff --git a/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/dtufc_elic-featurecoding_qwen_individual.log b/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/dtufc_elic-featurecoding_qwen_individual.log index d5f0cb5f54c9930cae82a60dbe06853555029247..b53a4fd7c523424db05c5e74bfa638d70768f997 100644 --- a/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/dtufc_elic-featurecoding_qwen_individual.log +++ b/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/dtufc_elic-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_elic-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/dtufc_elic-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: elic-featurecoding - handler: qwen - checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar - transform_type: kmeans - transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json - bit_depth: 8 - device: cuda:0 -Loading checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 506 -Loaded elic-featurecoding (1-channel) on cuda:0 -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json -Loaded per-key mappings: model=qwen - Keys: ['layer.0.k_cache', 'layer.0.v_cache', 'layer.1.k_cache', 'layer.1.v_cache', 'layer.2.k_cache', 'layer.2.v_cache', 'layer.3.k_cache', 'layer.3.v_cache', 'layer.4.k_cache', 'layer.4.v_cache', 'layer.4.output'] ----------------- ------------------------------------------------------------------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding -Checkpoint codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Transform type kmeans -Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json -Input ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge -Output output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge ----------------- ------------------------------------------------------------------------------------------------------------------- -Files found: 100 ----------------------------------------------------------------------- - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample0-layer4-item1.zst (1/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample0-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 243, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 243, 128) -Output shape: (1, 243, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.0.v_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.1.k_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.1.v_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.2.k_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.2.v_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.3.k_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.3.v_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.4.k_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.4.v_cache: torch.Size([1, 8, 243, 128]) -> torch.Size([1, 1, 243, 1024]) - layer.4.output: torch.Size([1, 243, 4096]) -> torch.Size([1, 1, 243, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 12,916B, BPFP=0.4153 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 66,436B, BPFP=2.1359 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,848B, BPFP=1.5383 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,236B, BPFP=2.2581 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 60,088B, BPFP=1.9318 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,356B, BPFP=2.2941 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,172B, BPFP=1.9345 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,752B, BPFP=2.2747 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,880B, BPFP=1.6358 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 72,132B, BPFP=2.3191 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 201,284B, BPFP=1.6178 -⌛️ [2/4] FRONTEND: Frontend time: 2.841s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 243, 128]) - layer.0.v_cache: torch.Size([1, 8, 243, 128]) - layer.1.k_cache: torch.Size([1, 8, 243, 128]) - layer.1.v_cache: torch.Size([1, 8, 243, 128]) - layer.2.k_cache: torch.Size([1, 8, 243, 128]) - layer.2.v_cache: torch.Size([1, 8, 243, 128]) - layer.3.k_cache: torch.Size([1, 8, 243, 128]) - layer.3.v_cache: torch.Size([1, 8, 243, 128]) - layer.4.k_cache: torch.Size([1, 8, 243, 128]) - layer.4.v_cache: torch.Size([1, 8, 243, 128]) - layer.4.output: torch.Size([1, 243, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.924s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 243, 128]) - layer.0.v_cache: torch.Size([1, 8, 243, 128]) - layer.1.k_cache: torch.Size([1, 8, 243, 128]) - layer.1.v_cache: torch.Size([1, 8, 243, 128]) - layer.2.k_cache: torch.Size([1, 8, 243, 128]) - layer.2.v_cache: torch.Size([1, 8, 243, 128]) - layer.3.k_cache: torch.Size([1, 8, 243, 128]) - layer.3.v_cache: torch.Size([1, 8, 243, 128]) - layer.4.k_cache: torch.Size([1, 8, 243, 128]) - layer.4.v_cache: torch.Size([1, 8, 243, 128]) - layer.4.output: torch.Size([1, 243, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02602252 7.07570921 - layer.0.v_cache 0.00000026 0.00013915 - layer.1.k_cache 0.00298528 0.74894030 - layer.1.v_cache 0.00000081 0.00050512 - layer.2.k_cache 0.00118445 0.37142100 - layer.2.v_cache 0.00000112 0.00072005 - layer.3.k_cache 0.00134199 0.43721756 - layer.3.v_cache 0.00000210 0.00116123 - layer.4.k_cache 0.00355146 0.79448157 - layer.4.v_cache 0.00000307 0.00202025 - layer.4.output 0.00016770 0.06840519 - ------------------------------------------------------------------------------------- - TOTAL 0.00255456 0.69328116 - (elements=3,483,648) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3483648 -Total Bytes 784100 -BPFP 1.8006 bits/point -EBPFP 3.6013 equivalent bits/point -MSE 0.693281 ----------------------- -------------------------------------------------------- -Time: 4.776s Load: 0.012s, Pack+Encode: 2.841s, Decode+Unpack: 1.924s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 243, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 243, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6933 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample0-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample0-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample1-layer4-item1.zst (2/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample1-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 265, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.014s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 265, 128) -Output shape: (1, 265, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.0.v_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.1.k_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.1.v_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.2.k_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.2.v_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.3.k_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.3.v_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.4.k_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.4.v_cache: torch.Size([1, 8, 265, 128]) -> torch.Size([1, 1, 265, 1024]) - layer.4.output: torch.Size([1, 265, 4096]) -> torch.Size([1, 1, 265, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 14,688B, BPFP=0.4330 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 80,888B, BPFP=2.3847 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 52,644B, BPFP=1.5520 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 85,900B, BPFP=2.5324 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 66,712B, BPFP=1.9667 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 86,400B, BPFP=2.5472 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 70,608B, BPFP=2.0816 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 86,264B, BPFP=2.5432 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 53,920B, BPFP=1.5896 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 87,400B, BPFP=2.5767 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 223,212B, BPFP=1.6451 -⌛️ [2/4] FRONTEND: Frontend time: 2.596s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 265, 128]) - layer.0.v_cache: torch.Size([1, 8, 265, 128]) - layer.1.k_cache: torch.Size([1, 8, 265, 128]) - layer.1.v_cache: torch.Size([1, 8, 265, 128]) - layer.2.k_cache: torch.Size([1, 8, 265, 128]) - layer.2.v_cache: torch.Size([1, 8, 265, 128]) - layer.3.k_cache: torch.Size([1, 8, 265, 128]) - layer.3.v_cache: torch.Size([1, 8, 265, 128]) - layer.4.k_cache: torch.Size([1, 8, 265, 128]) - layer.4.v_cache: torch.Size([1, 8, 265, 128]) - layer.4.output: torch.Size([1, 265, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.061s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 265, 128]) - layer.0.v_cache: torch.Size([1, 8, 265, 128]) - layer.1.k_cache: torch.Size([1, 8, 265, 128]) - layer.1.v_cache: torch.Size([1, 8, 265, 128]) - layer.2.k_cache: torch.Size([1, 8, 265, 128]) - layer.2.v_cache: torch.Size([1, 8, 265, 128]) - layer.3.k_cache: torch.Size([1, 8, 265, 128]) - layer.3.v_cache: torch.Size([1, 8, 265, 128]) - layer.4.k_cache: torch.Size([1, 8, 265, 128]) - layer.4.v_cache: torch.Size([1, 8, 265, 128]) - layer.4.output: torch.Size([1, 265, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02444536 7.08293918 - layer.0.v_cache 0.00000028 0.00014410 - layer.1.k_cache 0.00298152 0.87255071 - layer.1.v_cache 0.00000087 0.00051499 - layer.2.k_cache 0.00117974 0.35589392 - layer.2.v_cache 0.00000116 0.00073737 - layer.3.k_cache 0.00134702 0.40649276 - layer.3.v_cache 0.00000215 0.00116827 - layer.4.k_cache 0.00352357 0.75944634 - layer.4.v_cache 0.00000310 0.00200273 - layer.4.output 0.00016704 0.06678874 - ------------------------------------------------------------------------------------- - TOTAL 0.00243950 0.69636038 - (elements=3,799,040) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3799040 -Total Bytes 908636 -BPFP 1.9134 bits/point -EBPFP 3.8268 equivalent bits/point -MSE 0.696360 ----------------------- -------------------------------------------------------- -Time: 4.671s Load: 0.014s, Pack+Encode: 2.596s, Decode+Unpack: 2.061s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 265, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 265, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6964 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample1-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample10-layer4-item1.zst (3/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample10-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 213, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 213, 128) -Output shape: (1, 213, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.output: torch.Size([1, 213, 4096]) -> torch.Size([1, 1, 213, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,680B, BPFP=0.4284 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,512B, BPFP=2.4029 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 43,212B, BPFP=1.5849 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,212B, BPFP=2.5386 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,520B, BPFP=2.0731 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,300B, BPFP=2.5785 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,820B, BPFP=2.1207 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,468B, BPFP=2.5480 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,604B, BPFP=1.6360 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 70,448B, BPFP=2.5839 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 200,920B, BPFP=1.8424 -⌛️ [2/4] FRONTEND: Frontend time: 2.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.output: torch.Size([1, 213, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.999s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.output: torch.Size([1, 213, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02614268 6.91579555 - layer.0.v_cache 0.00000026 0.00015023 - layer.1.k_cache 0.00299160 0.80968551 - layer.1.v_cache 0.00000086 0.00053648 - layer.2.k_cache 0.00117415 0.36937373 - layer.2.v_cache 0.00000138 0.00076539 - layer.3.k_cache 0.00130742 0.41664367 - layer.3.v_cache 0.00000224 0.00121945 - layer.4.k_cache 0.00347054 0.78505173 - layer.4.v_cache 0.00000320 0.00202932 - layer.4.output 0.00018153 0.05718652 - ------------------------------------------------------------------------------------- - TOTAL 0.00255861 0.68071408 - (elements=3,053,568) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3053568 -Total Bytes 759696 -BPFP 1.9903 bits/point -EBPFP 3.9806 equivalent bits/point -MSE 0.680714 ----------------------- -------------------------------------------------------- -Time: 4.417s Load: 0.011s, Pack+Encode: 2.407s, Decode+Unpack: 1.999s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 213, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6807 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample10-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample10-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample100-layer4-item1.zst (4/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample100-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.013s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,144B, BPFP=0.4581 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,564B, BPFP=2.2834 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,872B, BPFP=1.6199 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,396B, BPFP=2.4113 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,136B, BPFP=2.0383 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,948B, BPFP=2.4362 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,856B, BPFP=2.0257 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,352B, BPFP=2.4093 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,468B, BPFP=1.6469 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,344B, BPFP=2.4541 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 167,568B, BPFP=1.8918 -⌛️ [2/4] FRONTEND: Frontend time: 2.213s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.540s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02711869 7.17517866 - layer.0.v_cache 0.00000027 0.00015200 - layer.1.k_cache 0.00316613 0.83979586 - layer.1.v_cache 0.00000085 0.00053243 - layer.2.k_cache 0.00116533 0.37720415 - layer.2.v_cache 0.00000134 0.00073869 - layer.3.k_cache 0.00131265 0.41170131 - layer.3.v_cache 0.00000224 0.00116050 - layer.4.k_cache 0.00350563 0.81082250 - layer.4.v_cache 0.00000306 0.00196414 - layer.4.output 0.00018814 0.06502701 - ------------------------------------------------------------------------------------- - TOTAL 0.00264491 0.70566845 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 605648 -BPFP 1.9536 bits/point -EBPFP 3.9072 equivalent bits/point -MSE 0.705668 ----------------------- -------------------------------------------------------- -Time: 3.766s Load: 0.013s, Pack+Encode: 2.213s, Decode+Unpack: 1.540s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7057 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample100-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample101-layer4-item1.zst (5/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample101-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 162, 128) -Output shape: (1, 162, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.output: torch.Size([1, 162, 4096]) -> torch.Size([1, 1, 162, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,092B, BPFP=0.4385 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,456B, BPFP=2.3850 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,016B, BPFP=1.6404 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,232B, BPFP=2.5189 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,868B, BPFP=2.1155 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,380B, BPFP=2.5743 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,560B, BPFP=2.1489 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,496B, BPFP=2.5316 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,672B, BPFP=1.7203 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,516B, BPFP=2.5808 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 138,688B, BPFP=1.6721 -⌛️ [2/4] FRONTEND: Frontend time: 1.934s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.output: torch.Size([1, 162, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.576s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.output: torch.Size([1, 162, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02797630 7.14845106 - layer.0.v_cache 0.00000026 0.00015149 - layer.1.k_cache 0.00305816 0.80668348 - layer.1.v_cache 0.00000079 0.00053766 - layer.2.k_cache 0.00117271 0.38929883 - layer.2.v_cache 0.00000112 0.00077820 - layer.3.k_cache 0.00136252 0.41491134 - layer.3.v_cache 0.00000203 0.00118843 - layer.4.k_cache 0.00349277 0.83123836 - layer.4.v_cache 0.00000321 0.00209728 - layer.4.output 0.00018139 0.06623011 - ------------------------------------------------------------------------------------- - TOTAL 0.00269967 0.70430404 - (elements=2,322,432) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2322432 -Total Bytes 566976 -BPFP 1.9530 bits/point -EBPFP 3.9061 equivalent bits/point -MSE 0.704304 ----------------------- -------------------------------------------------------- -Time: 3.519s Load: 0.009s, Pack+Encode: 1.934s, Decode+Unpack: 1.576s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7043 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample101-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample101-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample102-layer4-item1.zst (6/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample102-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 156, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 156, 128) -Output shape: (1, 156, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.output: torch.Size([1, 156, 4096]) -> torch.Size([1, 1, 156, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,964B, BPFP=0.4489 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,008B, BPFP=2.5044 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,848B, BPFP=1.6951 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,096B, BPFP=2.6591 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 42,148B, BPFP=2.1108 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,520B, BPFP=2.6803 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,140B, BPFP=2.2105 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,052B, BPFP=2.6569 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,872B, BPFP=1.7464 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,064B, BPFP=2.7075 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 103,368B, BPFP=1.2942 -⌛️ [2/4] FRONTEND: Frontend time: 1.953s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.output: torch.Size([1, 156, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.528s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.output: torch.Size([1, 156, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02683717 7.16035579 - layer.0.v_cache 0.00000027 0.00015075 - layer.1.k_cache 0.00303383 0.78772853 - layer.1.v_cache 0.00000085 0.00054007 - layer.2.k_cache 0.00113250 0.35597769 - layer.2.v_cache 0.00000110 0.00072938 - layer.3.k_cache 0.00132873 0.39865479 - layer.3.v_cache 0.00000207 0.00115059 - layer.4.k_cache 0.00336583 0.73689554 - layer.4.v_cache 0.00000309 0.00210161 - layer.4.output 0.00014117 0.06851907 - ------------------------------------------------------------------------------------- - TOTAL 0.00259072 0.69416864 - (elements=2,236,416) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2236416 -Total Bytes 531080 -BPFP 1.8998 bits/point -EBPFP 3.7995 equivalent bits/point -MSE 0.694169 ----------------------- -------------------------------------------------------- -Time: 3.490s Load: 0.009s, Pack+Encode: 1.953s, Decode+Unpack: 1.528s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 156, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6942 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample102-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample102-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample103-layer4-item1.zst (7/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample103-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 158, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 158, 128) -Output shape: (1, 158, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.0.v_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.1.k_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.1.v_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.2.k_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.2.v_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.3.k_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.3.v_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.4.k_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.4.v_cache: torch.Size([1, 8, 158, 128]) -> torch.Size([1, 1, 158, 1024]) - layer.4.output: torch.Size([1, 158, 4096]) -> torch.Size([1, 1, 158, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,712B, BPFP=0.4308 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,152B, BPFP=2.4798 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,704B, BPFP=1.6665 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,812B, BPFP=2.6114 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,620B, BPFP=2.1568 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,544B, BPFP=2.6475 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,116B, BPFP=2.1814 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,928B, BPFP=2.6171 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 33,780B, BPFP=1.6703 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,940B, BPFP=2.6671 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 109,640B, BPFP=1.3553 -⌛️ [2/4] FRONTEND: Frontend time: 1.953s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 158, 128]) - layer.0.v_cache: torch.Size([1, 8, 158, 128]) - layer.1.k_cache: torch.Size([1, 8, 158, 128]) - layer.1.v_cache: torch.Size([1, 8, 158, 128]) - layer.2.k_cache: torch.Size([1, 8, 158, 128]) - layer.2.v_cache: torch.Size([1, 8, 158, 128]) - layer.3.k_cache: torch.Size([1, 8, 158, 128]) - layer.3.v_cache: torch.Size([1, 8, 158, 128]) - layer.4.k_cache: torch.Size([1, 8, 158, 128]) - layer.4.v_cache: torch.Size([1, 8, 158, 128]) - layer.4.output: torch.Size([1, 158, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.569s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 158, 128]) - layer.0.v_cache: torch.Size([1, 8, 158, 128]) - layer.1.k_cache: torch.Size([1, 8, 158, 128]) - layer.1.v_cache: torch.Size([1, 8, 158, 128]) - layer.2.k_cache: torch.Size([1, 8, 158, 128]) - layer.2.v_cache: torch.Size([1, 8, 158, 128]) - layer.3.k_cache: torch.Size([1, 8, 158, 128]) - layer.3.v_cache: torch.Size([1, 8, 158, 128]) - layer.4.k_cache: torch.Size([1, 8, 158, 128]) - layer.4.v_cache: torch.Size([1, 8, 158, 128]) - layer.4.output: torch.Size([1, 158, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02787131 7.00428019 - layer.0.v_cache 0.00000026 0.00014942 - layer.1.k_cache 0.00312961 0.78944817 - layer.1.v_cache 0.00000083 0.00053500 - layer.2.k_cache 0.00117781 0.37872254 - layer.2.v_cache 0.00000107 0.00074589 - layer.3.k_cache 0.00134038 0.40842274 - layer.3.v_cache 0.00000203 0.00117721 - layer.4.k_cache 0.00346136 0.81557397 - layer.4.v_cache 0.00000301 0.00207983 - layer.4.output 0.00016064 0.06760067 - ------------------------------------------------------------------------------------- - TOTAL 0.00268788 0.69082412 - (elements=2,265,088) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2265088 -Total Bytes 536948 -BPFP 1.8964 bits/point -EBPFP 3.7929 equivalent bits/point -MSE 0.690824 ----------------------- -------------------------------------------------------- -Time: 3.532s Load: 0.010s, Pack+Encode: 1.953s, Decode+Unpack: 1.569s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 158, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 158, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6908 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample103-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample103-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample104-layer4-item1.zst (8/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample104-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 182, 128) -Output shape: (1, 182, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.output: torch.Size([1, 182, 4096]) -> torch.Size([1, 1, 182, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,760B, BPFP=0.4190 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,612B, BPFP=2.1296 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,488B, BPFP=1.5663 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,672B, BPFP=2.2610 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,224B, BPFP=1.9413 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,684B, BPFP=2.3044 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,060B, BPFP=1.9342 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,912B, BPFP=2.2713 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,956B, BPFP=1.5864 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,824B, BPFP=2.3104 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 128,712B, BPFP=1.3813 -⌛️ [2/4] FRONTEND: Frontend time: 1.936s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.535s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02710123 6.85396366 - layer.0.v_cache 0.00000028 0.00014966 - layer.1.k_cache 0.00304590 0.79905307 - layer.1.v_cache 0.00000075 0.00049890 - layer.2.k_cache 0.00114207 0.36272623 - layer.2.v_cache 0.00000117 0.00071540 - layer.3.k_cache 0.00132402 0.42336638 - layer.3.v_cache 0.00000202 0.00113285 - layer.4.k_cache 0.00339800 0.83256355 - layer.4.v_cache 0.00000283 0.00187712 - layer.4.output 0.00018478 0.08129646 - ------------------------------------------------------------------------------------- - TOTAL 0.00262553 0.68580233 - (elements=2,609,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2609152 -Total Bytes 564904 -BPFP 1.7321 bits/point -EBPFP 3.4641 equivalent bits/point -MSE 0.685802 ----------------------- -------------------------------------------------------- -Time: 3.481s Load: 0.010s, Pack+Encode: 1.936s, Decode+Unpack: 1.535s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6858 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample104-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample104-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample105-layer4-item1.zst (9/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample105-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 178, 128) -Output shape: (1, 178, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,808B, BPFP=0.4305 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,852B, BPFP=2.1880 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,460B, BPFP=1.5564 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,620B, BPFP=2.3095 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,752B, BPFP=1.9642 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,328B, BPFP=2.3406 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,564B, BPFP=1.9559 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,988B, BPFP=2.3257 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,108B, BPFP=1.6287 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,836B, BPFP=2.3629 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 111,216B, BPFP=1.2203 -⌛️ [2/4] FRONTEND: Frontend time: 1.899s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.445s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02621927 6.53497795 - layer.0.v_cache 0.00000027 0.00014880 - layer.1.k_cache 0.00300007 0.74263806 - layer.1.v_cache 0.00000077 0.00052358 - layer.2.k_cache 0.00117703 0.36926617 - layer.2.v_cache 0.00000113 0.00072222 - layer.3.k_cache 0.00133116 0.41016144 - layer.3.v_cache 0.00000210 0.00118247 - layer.4.k_cache 0.00351662 0.81633073 - layer.4.v_cache 0.00000301 0.00199099 - layer.4.output 0.00016984 0.06709759 - ------------------------------------------------------------------------------------- - TOTAL 0.00256649 0.65330948 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 545532 -BPFP 1.7103 bits/point -EBPFP 3.4205 equivalent bits/point -MSE 0.653309 ----------------------- -------------------------------------------------------- -Time: 3.355s Load: 0.011s, Pack+Encode: 1.899s, Decode+Unpack: 1.445s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6533 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample105-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample105-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample106-layer4-item1.zst (10/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample106-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 201, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 201, 128) -Output shape: (1, 201, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.0.v_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.1.k_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.1.v_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.2.k_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.2.v_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.3.k_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.3.v_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.4.k_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.4.v_cache: torch.Size([1, 8, 201, 128]) -> torch.Size([1, 1, 201, 1024]) - layer.4.output: torch.Size([1, 201, 4096]) -> torch.Size([1, 1, 201, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,108B, BPFP=0.3929 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 63,876B, BPFP=2.4827 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 39,432B, BPFP=1.5326 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 67,068B, BPFP=2.6068 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 52,744B, BPFP=2.0501 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 67,404B, BPFP=2.6199 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,452B, BPFP=2.1942 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 67,480B, BPFP=2.6228 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 39,304B, BPFP=1.5277 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 68,424B, BPFP=2.6595 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 120,692B, BPFP=1.1728 -⌛️ [2/4] FRONTEND: Frontend time: 2.353s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 201, 128]) - layer.0.v_cache: torch.Size([1, 8, 201, 128]) - layer.1.k_cache: torch.Size([1, 8, 201, 128]) - layer.1.v_cache: torch.Size([1, 8, 201, 128]) - layer.2.k_cache: torch.Size([1, 8, 201, 128]) - layer.2.v_cache: torch.Size([1, 8, 201, 128]) - layer.3.k_cache: torch.Size([1, 8, 201, 128]) - layer.3.v_cache: torch.Size([1, 8, 201, 128]) - layer.4.k_cache: torch.Size([1, 8, 201, 128]) - layer.4.v_cache: torch.Size([1, 8, 201, 128]) - layer.4.output: torch.Size([1, 201, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.025s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 201, 128]) - layer.0.v_cache: torch.Size([1, 8, 201, 128]) - layer.1.k_cache: torch.Size([1, 8, 201, 128]) - layer.1.v_cache: torch.Size([1, 8, 201, 128]) - layer.2.k_cache: torch.Size([1, 8, 201, 128]) - layer.2.v_cache: torch.Size([1, 8, 201, 128]) - layer.3.k_cache: torch.Size([1, 8, 201, 128]) - layer.3.v_cache: torch.Size([1, 8, 201, 128]) - layer.4.k_cache: torch.Size([1, 8, 201, 128]) - layer.4.v_cache: torch.Size([1, 8, 201, 128]) - layer.4.output: torch.Size([1, 201, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02721510 6.42558448 - layer.0.v_cache 0.00000027 0.00015542 - layer.1.k_cache 0.00313268 0.83467428 - layer.1.v_cache 0.00000087 0.00052564 - layer.2.k_cache 0.00111785 0.36064342 - layer.2.v_cache 0.00000101 0.00065022 - layer.3.k_cache 0.00129873 0.40284110 - layer.3.v_cache 0.00000193 0.00104725 - layer.4.k_cache 0.00357359 0.84391868 - layer.4.v_cache 0.00000284 0.00173268 - layer.4.output 0.00012728 0.05851358 - ------------------------------------------------------------------------------------- - TOTAL 0.00263243 0.65041625 - (elements=2,881,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2881536 -Total Bytes 652984 -BPFP 1.8129 bits/point -EBPFP 3.6258 equivalent bits/point -MSE 0.650416 ----------------------- -------------------------------------------------------- -Time: 4.388s Load: 0.010s, Pack+Encode: 2.353s, Decode+Unpack: 2.025s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 201, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 201, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6504 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample106-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample106-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample107-layer4-item1.zst (11/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample107-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 163, 128) -Output shape: (1, 163, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.output: torch.Size([1, 163, 4096]) -> torch.Size([1, 1, 163, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,688B, BPFP=0.4164 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,184B, BPFP=2.3574 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,600B, BPFP=1.6584 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 51,420B, BPFP=2.4645 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,032B, BPFP=2.0625 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,528B, BPFP=2.5176 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,068B, BPFP=2.1122 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,708B, BPFP=2.4783 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,016B, BPFP=1.7262 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 52,848B, BPFP=2.5330 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 102,096B, BPFP=1.2234 -⌛️ [2/4] FRONTEND: Frontend time: 1.973s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.532s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02712815 7.24005764 - layer.0.v_cache 0.00000026 0.00014862 - layer.1.k_cache 0.00315731 0.80513450 - layer.1.v_cache 0.00000075 0.00051331 - layer.2.k_cache 0.00115274 0.38190453 - layer.2.v_cache 0.00000104 0.00070684 - layer.3.k_cache 0.00137819 0.42518148 - layer.3.v_cache 0.00000196 0.00113328 - layer.4.k_cache 0.00344312 0.82453076 - layer.4.v_cache 0.00000284 0.00195136 - layer.4.output 0.00020076 0.07391944 - ------------------------------------------------------------------------------------- - TOTAL 0.00264781 0.71263857 - (elements=2,336,768) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2336768 -Total Bytes 526188 -BPFP 1.8014 bits/point -EBPFP 3.6028 equivalent bits/point -MSE 0.712639 ----------------------- -------------------------------------------------------- -Time: 3.514s Load: 0.009s, Pack+Encode: 1.973s, Decode+Unpack: 1.532s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7126 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample107-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample107-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample108-layer4-item1.zst (12/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample108-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 163, 128) -Output shape: (1, 163, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.output: torch.Size([1, 163, 4096]) -> torch.Size([1, 1, 163, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,720B, BPFP=0.4179 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,336B, BPFP=2.3646 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,756B, BPFP=1.6658 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 51,740B, BPFP=2.4799 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,088B, BPFP=2.1131 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,644B, BPFP=2.5232 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,084B, BPFP=2.1129 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,836B, BPFP=2.4845 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,624B, BPFP=1.7074 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,040B, BPFP=2.5422 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 107,152B, BPFP=1.2839 -⌛️ [2/4] FRONTEND: Frontend time: 1.940s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.582s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02704066 7.13262528 - layer.0.v_cache 0.00000026 0.00015038 - layer.1.k_cache 0.00309474 0.76622243 - layer.1.v_cache 0.00000079 0.00053084 - layer.2.k_cache 0.00116731 0.36760522 - layer.2.v_cache 0.00000107 0.00071733 - layer.3.k_cache 0.00138539 0.42015179 - layer.3.v_cache 0.00000193 0.00114633 - layer.4.k_cache 0.00353297 0.81392225 - layer.4.v_cache 0.00000296 0.00199471 - layer.4.output 0.00020208 0.07552489 - ------------------------------------------------------------------------------------- - TOTAL 0.00264546 0.70051186 - (elements=2,336,768) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2336768 -Total Bytes 533020 -BPFP 1.8248 bits/point -EBPFP 3.6496 equivalent bits/point -MSE 0.700512 ----------------------- -------------------------------------------------------- -Time: 3.530s Load: 0.009s, Pack+Encode: 1.940s, Decode+Unpack: 1.582s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7005 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample108-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample109-layer4-item1.zst (13/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample109-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 178, 128) -Output shape: (1, 178, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,800B, BPFP=0.4301 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,028B, BPFP=2.1958 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,728B, BPFP=1.5681 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,744B, BPFP=2.3150 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,104B, BPFP=1.9796 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,676B, BPFP=2.3559 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,884B, BPFP=1.9700 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,060B, BPFP=2.3288 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,960B, BPFP=1.6222 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,980B, BPFP=2.3692 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 138,700B, BPFP=1.5219 -⌛️ [2/4] FRONTEND: Frontend time: 1.953s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.546s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02815543 6.99614998 - layer.0.v_cache 0.00000027 0.00014941 - layer.1.k_cache 0.00311575 0.81333975 - layer.1.v_cache 0.00000080 0.00054133 - layer.2.k_cache 0.00115654 0.38754290 - layer.2.v_cache 0.00000128 0.00075355 - layer.3.k_cache 0.00138975 0.42702158 - layer.3.v_cache 0.00000206 0.00117295 - layer.4.k_cache 0.00349998 0.85007554 - layer.4.v_cache 0.00000302 0.00201668 - layer.4.output 0.00017961 0.06949260 - ------------------------------------------------------------------------------------- - TOTAL 0.00271738 0.69690958 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 574664 -BPFP 1.8016 bits/point -EBPFP 3.6032 equivalent bits/point -MSE 0.696910 ----------------------- -------------------------------------------------------- -Time: 3.509s Load: 0.010s, Pack+Encode: 1.953s, Decode+Unpack: 1.546s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6969 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample109-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample109-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample11-layer4-item1.zst (14/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample11-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 190, 128) -Output shape: (1, 190, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,740B, BPFP=0.4416 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,848B, BPFP=2.0908 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,412B, BPFP=1.4972 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,752B, BPFP=2.2102 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,148B, BPFP=1.8564 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,276B, BPFP=2.2317 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,200B, BPFP=1.8586 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 54,020B, BPFP=2.2212 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 38,220B, BPFP=1.5715 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,508B, BPFP=2.2413 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 168,124B, BPFP=1.7282 -⌛️ [2/4] FRONTEND: Frontend time: 1.943s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.560s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02605901 6.89398450 - layer.0.v_cache 0.00000026 0.00014486 - layer.1.k_cache 0.00295386 0.94311732 - layer.1.v_cache 0.00000080 0.00052738 - layer.2.k_cache 0.00114499 0.38086865 - layer.2.v_cache 0.00000122 0.00072513 - layer.3.k_cache 0.00131775 0.45837639 - layer.3.v_cache 0.00000219 0.00119072 - layer.4.k_cache 0.00343004 0.85670383 - layer.4.v_cache 0.00000314 0.00200061 - layer.4.output 0.00018832 0.06313341 - ------------------------------------------------------------------------------------- - TOTAL 0.00254761 0.69929807 - (elements=2,723,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2723840 -Total Bytes 611248 -BPFP 1.7953 bits/point -EBPFP 3.5905 equivalent bits/point -MSE 0.699298 ----------------------- -------------------------------------------------------- -Time: 3.514s Load: 0.010s, Pack+Encode: 1.943s, Decode+Unpack: 1.560s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6993 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample11-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample11-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample111-layer4-item1.zst (15/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample111-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 167, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 167, 128) -Output shape: (1, 167, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.output: torch.Size([1, 167, 4096]) -> torch.Size([1, 1, 167, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,408B, BPFP=0.4401 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,788B, BPFP=2.3292 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,580B, BPFP=1.6177 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,188B, BPFP=2.4414 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,752B, BPFP=2.0468 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,056B, BPFP=2.4820 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,468B, BPFP=2.0803 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,268B, BPFP=2.4452 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,288B, BPFP=1.6508 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,340B, BPFP=2.4953 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 137,612B, BPFP=1.6094 -⌛️ [2/4] FRONTEND: Frontend time: 1.958s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.output: torch.Size([1, 167, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.554s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.output: torch.Size([1, 167, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02739838 7.03707356 - layer.0.v_cache 0.00000026 0.00014948 - layer.1.k_cache 0.00313824 0.77538642 - layer.1.v_cache 0.00000083 0.00053513 - layer.2.k_cache 0.00117185 0.36864241 - layer.2.v_cache 0.00000105 0.00071671 - layer.3.k_cache 0.00137006 0.42159938 - layer.3.v_cache 0.00000199 0.00114181 - layer.4.k_cache 0.00340675 0.84074356 - layer.4.v_cache 0.00000299 0.00201766 - layer.4.output 0.00021989 0.07095393 - ------------------------------------------------------------------------------------- - TOTAL 0.00266943 0.69513013 - (elements=2,394,112) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2394112 -Total Bytes 565748 -BPFP 1.8905 bits/point -EBPFP 3.7809 equivalent bits/point -MSE 0.695130 ----------------------- -------------------------------------------------------- -Time: 3.523s Load: 0.010s, Pack+Encode: 1.958s, Decode+Unpack: 1.554s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 167, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6951 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample111-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample111-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample112-layer4-item1.zst (16/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample112-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 164, 128) -Output shape: (1, 164, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,744B, BPFP=0.4165 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,208B, BPFP=2.3441 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,812B, BPFP=1.6583 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 51,544B, BPFP=2.4554 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,544B, BPFP=2.0743 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,320B, BPFP=2.4924 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,940B, BPFP=2.0932 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,840B, BPFP=2.4695 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,204B, BPFP=1.6770 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 52,960B, BPFP=2.5229 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 104,712B, BPFP=1.2470 -⌛️ [2/4] FRONTEND: Frontend time: 1.982s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.549s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02765335 7.00502647 - layer.0.v_cache 0.00000026 0.00015261 - layer.1.k_cache 0.00306024 0.79336101 - layer.1.v_cache 0.00000075 0.00051655 - layer.2.k_cache 0.00113567 0.37990258 - layer.2.v_cache 0.00000102 0.00070805 - layer.3.k_cache 0.00140214 0.42675186 - layer.3.v_cache 0.00000194 0.00112149 - layer.4.k_cache 0.00347389 0.85414375 - layer.4.v_cache 0.00000286 0.00198653 - layer.4.output 0.00016775 0.07210316 - ------------------------------------------------------------------------------------- - TOTAL 0.00267165 0.69657739 - (elements=2,351,104) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2351104 -Total Bytes 528828 -BPFP 1.7994 bits/point -EBPFP 3.5988 equivalent bits/point -MSE 0.696577 ----------------------- -------------------------------------------------------- -Time: 3.541s Load: 0.010s, Pack+Encode: 1.982s, Decode+Unpack: 1.549s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6966 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample112-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample112-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample12-layer4-item1.zst (17/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample12-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 204, 128) -Output shape: (1, 204, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.output: torch.Size([1, 204, 4096]) -> torch.Size([1, 1, 204, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,372B, BPFP=0.4355 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,260B, BPFP=2.4992 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,728B, BPFP=1.5980 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 66,632B, BPFP=2.5518 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,804B, BPFP=2.0605 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 68,592B, BPFP=2.6268 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,328B, BPFP=2.1572 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 68,364B, BPFP=2.6181 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 41,400B, BPFP=1.5855 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 69,444B, BPFP=2.6595 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 137,480B, BPFP=1.3163 -⌛️ [2/4] FRONTEND: Frontend time: 2.358s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.output: torch.Size([1, 204, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.008s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.output: torch.Size([1, 204, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02692792 6.65853942 - layer.0.v_cache 0.00000026 0.00014416 - layer.1.k_cache 0.00296634 0.86855623 - layer.1.v_cache 0.00000076 0.00051036 - layer.2.k_cache 0.00113428 0.39193939 - layer.2.v_cache 0.00000108 0.00070870 - layer.3.k_cache 0.00134079 0.42401194 - layer.3.v_cache 0.00000199 0.00112054 - layer.4.k_cache 0.00357197 0.78744627 - layer.4.v_cache 0.00000302 0.00196783 - layer.4.output 0.00017435 0.06556030 - ------------------------------------------------------------------------------------- - TOTAL 0.00261756 0.67122757 - (elements=2,924,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2924544 -Total Bytes 680404 -BPFP 1.8612 bits/point -EBPFP 3.7224 equivalent bits/point -MSE 0.671228 ----------------------- -------------------------------------------------------- -Time: 4.379s Load: 0.012s, Pack+Encode: 2.358s, Decode+Unpack: 2.008s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6712 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample12-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample12-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample13-layer4-item1.zst (18/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample13-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 207, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 207, 128) -Output shape: (1, 207, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.output: torch.Size([1, 207, 4096]) -> torch.Size([1, 1, 207, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,496B, BPFP=0.4339 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,112B, BPFP=2.4197 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,876B, BPFP=1.5805 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 67,564B, BPFP=2.5500 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,696B, BPFP=2.0643 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 67,432B, BPFP=2.5450 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,636B, BPFP=2.1375 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 68,404B, BPFP=2.5817 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 41,476B, BPFP=1.5654 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 69,000B, BPFP=2.6042 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 150,052B, BPFP=1.4158 -⌛️ [2/4] FRONTEND: Frontend time: 2.381s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.output: torch.Size([1, 207, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.002s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.output: torch.Size([1, 207, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02640885 7.12998188 - layer.0.v_cache 0.00000027 0.00014481 - layer.1.k_cache 0.00296321 0.85340211 - layer.1.v_cache 0.00000080 0.00052379 - layer.2.k_cache 0.00115200 0.37096350 - layer.2.v_cache 0.00000114 0.00072083 - layer.3.k_cache 0.00135178 0.40427303 - layer.3.v_cache 0.00000205 0.00117061 - layer.4.k_cache 0.00347904 0.77515496 - layer.4.v_cache 0.00000310 0.00202967 - layer.4.output 0.00016949 0.07055709 - ------------------------------------------------------------------------------------- - TOTAL 0.00257430 0.70147097 - (elements=2,967,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2967552 -Total Bytes 692744 -BPFP 1.8675 bits/point -EBPFP 3.7350 equivalent bits/point -MSE 0.701471 ----------------------- -------------------------------------------------------- -Time: 4.394s Load: 0.012s, Pack+Encode: 2.381s, Decode+Unpack: 2.002s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 207, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7015 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample13-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample14-layer4-item1.zst (19/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample14-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 192, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 192, 128) -Output shape: (1, 192, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.0.v_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.1.k_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.1.v_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.2.k_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.2.v_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.3.k_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.3.v_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.4.k_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.4.v_cache: torch.Size([1, 8, 192, 128]) -> torch.Size([1, 1, 192, 1024]) - layer.4.output: torch.Size([1, 192, 4096]) -> torch.Size([1, 1, 192, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,000B, BPFP=0.4069 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,104B, BPFP=1.9980 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,380B, BPFP=1.4396 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,092B, BPFP=2.1196 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,452B, BPFP=1.7681 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,668B, BPFP=2.1431 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,088B, BPFP=1.7533 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,216B, BPFP=2.1247 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,152B, BPFP=1.5117 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,236B, BPFP=2.1662 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 142,864B, BPFP=1.4533 -⌛️ [2/4] FRONTEND: Frontend time: 1.998s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 192, 128]) - layer.0.v_cache: torch.Size([1, 8, 192, 128]) - layer.1.k_cache: torch.Size([1, 8, 192, 128]) - layer.1.v_cache: torch.Size([1, 8, 192, 128]) - layer.2.k_cache: torch.Size([1, 8, 192, 128]) - layer.2.v_cache: torch.Size([1, 8, 192, 128]) - layer.3.k_cache: torch.Size([1, 8, 192, 128]) - layer.3.v_cache: torch.Size([1, 8, 192, 128]) - layer.4.k_cache: torch.Size([1, 8, 192, 128]) - layer.4.v_cache: torch.Size([1, 8, 192, 128]) - layer.4.output: torch.Size([1, 192, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.542s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 192, 128]) - layer.0.v_cache: torch.Size([1, 8, 192, 128]) - layer.1.k_cache: torch.Size([1, 8, 192, 128]) - layer.1.v_cache: torch.Size([1, 8, 192, 128]) - layer.2.k_cache: torch.Size([1, 8, 192, 128]) - layer.2.v_cache: torch.Size([1, 8, 192, 128]) - layer.3.k_cache: torch.Size([1, 8, 192, 128]) - layer.3.v_cache: torch.Size([1, 8, 192, 128]) - layer.4.k_cache: torch.Size([1, 8, 192, 128]) - layer.4.v_cache: torch.Size([1, 8, 192, 128]) - layer.4.output: torch.Size([1, 192, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02717149 6.83115641 - layer.0.v_cache 0.00000027 0.00013674 - layer.1.k_cache 0.00306874 0.78850969 - layer.1.v_cache 0.00000081 0.00049516 - layer.2.k_cache 0.00115207 0.35727111 - layer.2.v_cache 0.00000118 0.00067121 - layer.3.k_cache 0.00133924 0.38898619 - layer.3.v_cache 0.00000214 0.00107163 - layer.4.k_cache 0.00342495 0.73055347 - layer.4.v_cache 0.00000316 0.00187651 - layer.4.output 0.00018805 0.05754121 - ------------------------------------------------------------------------------------- - TOTAL 0.00263687 0.66649235 - (elements=2,752,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2752512 -Total Bytes 571252 -BPFP 1.6603 bits/point -EBPFP 3.3206 equivalent bits/point -MSE 0.666492 ----------------------- -------------------------------------------------------- -Time: 3.550s Load: 0.010s, Pack+Encode: 1.998s, Decode+Unpack: 1.542s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 192, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 192, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6665 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample14-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample14-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample16-layer4-item1.zst (20/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample16-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 207, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 207, 128) -Output shape: (1, 207, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) -> torch.Size([1, 1, 207, 1024]) - layer.4.output: torch.Size([1, 207, 4096]) -> torch.Size([1, 1, 207, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 12,100B, BPFP=0.4567 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 63,984B, BPFP=2.4149 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,744B, BPFP=1.5755 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 68,452B, BPFP=2.5835 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,428B, BPFP=2.0919 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 68,460B, BPFP=2.5838 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,364B, BPFP=2.1650 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,372B, BPFP=2.6182 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 42,684B, BPFP=1.6110 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 69,752B, BPFP=2.6325 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 200,760B, BPFP=1.8942 -⌛️ [2/4] FRONTEND: Frontend time: 2.348s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.output: torch.Size([1, 207, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.014s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 207, 128]) - layer.0.v_cache: torch.Size([1, 8, 207, 128]) - layer.1.k_cache: torch.Size([1, 8, 207, 128]) - layer.1.v_cache: torch.Size([1, 8, 207, 128]) - layer.2.k_cache: torch.Size([1, 8, 207, 128]) - layer.2.v_cache: torch.Size([1, 8, 207, 128]) - layer.3.k_cache: torch.Size([1, 8, 207, 128]) - layer.3.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.k_cache: torch.Size([1, 8, 207, 128]) - layer.4.v_cache: torch.Size([1, 8, 207, 128]) - layer.4.output: torch.Size([1, 207, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02737031 6.85184379 - layer.0.v_cache 0.00000026 0.00014809 - layer.1.k_cache 0.00295586 0.76526196 - layer.1.v_cache 0.00000087 0.00053954 - layer.2.k_cache 0.00115404 0.36688973 - layer.2.v_cache 0.00000118 0.00072774 - layer.3.k_cache 0.00130266 0.40537465 - layer.3.v_cache 0.00000222 0.00121284 - layer.4.k_cache 0.00345388 0.77401512 - layer.4.v_cache 0.00000327 0.00205732 - layer.4.output 0.00018665 0.06640053 - ------------------------------------------------------------------------------------- - TOTAL 0.00264223 0.67383378 - (elements=2,967,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2967552 -Total Bytes 750100 -BPFP 2.0221 bits/point -EBPFP 4.0443 equivalent bits/point -MSE 0.673834 ----------------------- -------------------------------------------------------- -Time: 4.373s Load: 0.011s, Pack+Encode: 2.348s, Decode+Unpack: 2.014s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 207, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 207, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6738 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample16-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample16-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample17-layer4-item1.zst (21/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample17-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 213, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 213, 128) -Output shape: (1, 213, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) -> torch.Size([1, 1, 213, 1024]) - layer.4.output: torch.Size([1, 213, 4096]) -> torch.Size([1, 1, 213, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,784B, BPFP=0.4322 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,592B, BPFP=2.4058 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,992B, BPFP=1.5769 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,096B, BPFP=2.5343 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,376B, BPFP=2.0678 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,424B, BPFP=2.5830 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,864B, BPFP=2.1224 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,564B, BPFP=2.5515 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,876B, BPFP=1.6093 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 70,744B, BPFP=2.5948 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 142,848B, BPFP=1.3099 -⌛️ [2/4] FRONTEND: Frontend time: 2.355s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.output: torch.Size([1, 213, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.905s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 213, 128]) - layer.0.v_cache: torch.Size([1, 8, 213, 128]) - layer.1.k_cache: torch.Size([1, 8, 213, 128]) - layer.1.v_cache: torch.Size([1, 8, 213, 128]) - layer.2.k_cache: torch.Size([1, 8, 213, 128]) - layer.2.v_cache: torch.Size([1, 8, 213, 128]) - layer.3.k_cache: torch.Size([1, 8, 213, 128]) - layer.3.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.k_cache: torch.Size([1, 8, 213, 128]) - layer.4.v_cache: torch.Size([1, 8, 213, 128]) - layer.4.output: torch.Size([1, 213, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02567197 6.75290619 - layer.0.v_cache 0.00000028 0.00015211 - layer.1.k_cache 0.00297252 0.82048601 - layer.1.v_cache 0.00000081 0.00052012 - layer.2.k_cache 0.00119374 0.36412163 - layer.2.v_cache 0.00000110 0.00071385 - layer.3.k_cache 0.00130971 0.38927646 - layer.3.v_cache 0.00000208 0.00114946 - layer.4.k_cache 0.00366947 0.79369866 - layer.4.v_cache 0.00000305 0.00199410 - layer.4.output 0.00013440 0.05499912 - ------------------------------------------------------------------------------------- - TOTAL 0.00252588 0.66750108 - (elements=3,053,568) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3053568 -Total Bytes 701160 -BPFP 1.8370 bits/point -EBPFP 3.6739 equivalent bits/point -MSE 0.667501 ----------------------- -------------------------------------------------------- -Time: 4.272s Load: 0.012s, Pack+Encode: 2.355s, Decode+Unpack: 1.905s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 213, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 213, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6675 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample17-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample17-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample18-layer4-item1.zst (22/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample18-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 191, 128) -Output shape: (1, 191, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.output: torch.Size([1, 191, 4096]) -> torch.Size([1, 1, 191, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,324B, BPFP=0.4223 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,408B, BPFP=2.0618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,848B, BPFP=1.5072 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,308B, BPFP=2.1805 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,368B, BPFP=1.8148 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,808B, BPFP=2.2009 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,260B, BPFP=1.8104 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,468B, BPFP=2.1870 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,256B, BPFP=1.5239 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,044B, BPFP=2.2106 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 137,144B, BPFP=1.4024 -⌛️ [2/4] FRONTEND: Frontend time: 1.964s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.517s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02670301 7.09267693 - layer.0.v_cache 0.00000027 0.00014282 - layer.1.k_cache 0.00305067 0.96510211 - layer.1.v_cache 0.00000084 0.00051435 - layer.2.k_cache 0.00117351 0.37695137 - layer.2.v_cache 0.00000111 0.00069724 - layer.3.k_cache 0.00132215 0.43139513 - layer.3.v_cache 0.00000209 0.00112278 - layer.4.k_cache 0.00346212 0.77431412 - layer.4.v_cache 0.00000303 0.00191375 - layer.4.output 0.00018565 0.05908072 - ------------------------------------------------------------------------------------- - TOTAL 0.00260439 0.70579668 - (elements=2,738,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2738176 -Total Bytes 575236 -BPFP 1.6806 bits/point -EBPFP 3.3613 equivalent bits/point -MSE 0.705797 ----------------------- -------------------------------------------------------- -Time: 3.492s Load: 0.012s, Pack+Encode: 1.964s, Decode+Unpack: 1.517s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7058 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample18-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample18-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample19-layer4-item1.zst (23/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample19-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 194, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 194, 128) -Output shape: (1, 194, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.0.v_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.1.k_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.1.v_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.2.k_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.2.v_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.3.k_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.3.v_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.4.k_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.4.v_cache: torch.Size([1, 8, 194, 128]) -> torch.Size([1, 1, 194, 1024]) - layer.4.output: torch.Size([1, 194, 4096]) -> torch.Size([1, 1, 194, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,168B, BPFP=0.4095 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 61,504B, BPFP=2.4768 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 38,952B, BPFP=1.5686 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 66,300B, BPFP=2.6699 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 47,560B, BPFP=1.9153 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 66,580B, BPFP=2.6812 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 52,060B, BPFP=2.0965 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 65,244B, BPFP=2.6274 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 38,228B, BPFP=1.5395 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 66,856B, BPFP=2.6923 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 135,776B, BPFP=1.3669 -⌛️ [2/4] FRONTEND: Frontend time: 2.397s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 194, 128]) - layer.0.v_cache: torch.Size([1, 8, 194, 128]) - layer.1.k_cache: torch.Size([1, 8, 194, 128]) - layer.1.v_cache: torch.Size([1, 8, 194, 128]) - layer.2.k_cache: torch.Size([1, 8, 194, 128]) - layer.2.v_cache: torch.Size([1, 8, 194, 128]) - layer.3.k_cache: torch.Size([1, 8, 194, 128]) - layer.3.v_cache: torch.Size([1, 8, 194, 128]) - layer.4.k_cache: torch.Size([1, 8, 194, 128]) - layer.4.v_cache: torch.Size([1, 8, 194, 128]) - layer.4.output: torch.Size([1, 194, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.910s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 194, 128]) - layer.0.v_cache: torch.Size([1, 8, 194, 128]) - layer.1.k_cache: torch.Size([1, 8, 194, 128]) - layer.1.v_cache: torch.Size([1, 8, 194, 128]) - layer.2.k_cache: torch.Size([1, 8, 194, 128]) - layer.2.v_cache: torch.Size([1, 8, 194, 128]) - layer.3.k_cache: torch.Size([1, 8, 194, 128]) - layer.3.v_cache: torch.Size([1, 8, 194, 128]) - layer.4.k_cache: torch.Size([1, 8, 194, 128]) - layer.4.v_cache: torch.Size([1, 8, 194, 128]) - layer.4.output: torch.Size([1, 194, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02707747 7.20646746 - layer.0.v_cache 0.00000027 0.00014493 - layer.1.k_cache 0.00295501 0.79032395 - layer.1.v_cache 0.00000076 0.00052274 - layer.2.k_cache 0.00116763 0.37852627 - layer.2.v_cache 0.00000111 0.00073524 - layer.3.k_cache 0.00133119 0.40535248 - layer.3.v_cache 0.00000207 0.00115195 - layer.4.k_cache 0.00346189 0.76500820 - layer.4.v_cache 0.00000304 0.00198189 - layer.4.output 0.00015844 0.06794970 - ------------------------------------------------------------------------------------- - TOTAL 0.00261673 0.70157242 - (elements=2,781,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2781184 -Total Bytes 649228 -BPFP 1.8675 bits/point -EBPFP 3.7350 equivalent bits/point -MSE 0.701572 ----------------------- -------------------------------------------------------- -Time: 4.318s Load: 0.011s, Pack+Encode: 2.397s, Decode+Unpack: 1.910s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 194, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 194, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7016 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample19-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample19-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample2-layer4-item1.zst (24/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample2-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 241, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.013s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 241, 128) -Output shape: (1, 241, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.0.v_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.1.k_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.1.v_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.2.k_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.2.v_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.3.k_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.3.v_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.4.k_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.4.v_cache: torch.Size([1, 8, 241, 128]) -> torch.Size([1, 1, 241, 1024]) - layer.4.output: torch.Size([1, 241, 4096]) -> torch.Size([1, 1, 241, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 13,088B, BPFP=0.4243 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 66,532B, BPFP=2.1568 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,676B, BPFP=1.5779 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,756B, BPFP=2.2937 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 59,600B, BPFP=1.9321 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 72,112B, BPFP=2.3377 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,840B, BPFP=1.9398 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 71,504B, BPFP=2.3179 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 51,272B, BPFP=1.6621 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 72,556B, BPFP=2.3520 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 165,108B, BPFP=1.3381 -⌛️ [2/4] FRONTEND: Frontend time: 2.377s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 241, 128]) - layer.0.v_cache: torch.Size([1, 8, 241, 128]) - layer.1.k_cache: torch.Size([1, 8, 241, 128]) - layer.1.v_cache: torch.Size([1, 8, 241, 128]) - layer.2.k_cache: torch.Size([1, 8, 241, 128]) - layer.2.v_cache: torch.Size([1, 8, 241, 128]) - layer.3.k_cache: torch.Size([1, 8, 241, 128]) - layer.3.v_cache: torch.Size([1, 8, 241, 128]) - layer.4.k_cache: torch.Size([1, 8, 241, 128]) - layer.4.v_cache: torch.Size([1, 8, 241, 128]) - layer.4.output: torch.Size([1, 241, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.960s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 241, 128]) - layer.0.v_cache: torch.Size([1, 8, 241, 128]) - layer.1.k_cache: torch.Size([1, 8, 241, 128]) - layer.1.v_cache: torch.Size([1, 8, 241, 128]) - layer.2.k_cache: torch.Size([1, 8, 241, 128]) - layer.2.v_cache: torch.Size([1, 8, 241, 128]) - layer.3.k_cache: torch.Size([1, 8, 241, 128]) - layer.3.v_cache: torch.Size([1, 8, 241, 128]) - layer.4.k_cache: torch.Size([1, 8, 241, 128]) - layer.4.v_cache: torch.Size([1, 8, 241, 128]) - layer.4.output: torch.Size([1, 241, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02558961 6.99760671 - layer.0.v_cache 0.00000026 0.00014350 - layer.1.k_cache 0.00297512 0.82006849 - layer.1.v_cache 0.00000081 0.00051591 - layer.2.k_cache 0.00122557 0.36811439 - layer.2.v_cache 0.00000115 0.00074310 - layer.3.k_cache 0.00132258 0.41827791 - layer.3.v_cache 0.00000273 0.00121884 - layer.4.k_cache 0.00351858 0.80255323 - layer.4.v_cache 0.00000307 0.00201645 - layer.4.output 0.00016366 0.06319599 - ------------------------------------------------------------------------------------- - TOTAL 0.00252101 0.69028875 - (elements=3,454,976) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3454976 -Total Bytes 751044 -BPFP 1.7390 bits/point -EBPFP 3.4781 equivalent bits/point -MSE 0.690289 ----------------------- -------------------------------------------------------- -Time: 4.351s Load: 0.013s, Pack+Encode: 2.377s, Decode+Unpack: 1.960s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 241, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 241, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6903 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample2-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample20-layer4-item1.zst (25/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample20-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 197, 128) -Output shape: (1, 197, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.output: torch.Size([1, 197, 4096]) -> torch.Size([1, 1, 197, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,700B, BPFP=0.4243 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 62,140B, BPFP=2.4643 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 38,892B, BPFP=1.5424 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 65,980B, BPFP=2.6166 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 51,220B, BPFP=2.0312 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 66,980B, BPFP=2.6562 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 53,404B, BPFP=2.1179 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 66,016B, BPFP=2.6180 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 39,592B, BPFP=1.5701 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 67,496B, BPFP=2.6767 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 125,840B, BPFP=1.2476 -⌛️ [2/4] FRONTEND: Frontend time: 2.375s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.output: torch.Size([1, 197, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.925s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.output: torch.Size([1, 197, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02755299 7.12028697 - layer.0.v_cache 0.00000026 0.00014672 - layer.1.k_cache 0.00304278 0.78777178 - layer.1.v_cache 0.00000080 0.00052380 - layer.2.k_cache 0.00114678 0.38051024 - layer.2.v_cache 0.00000109 0.00070981 - layer.3.k_cache 0.00134691 0.41956329 - layer.3.v_cache 0.00000250 0.00117150 - layer.4.k_cache 0.00338668 0.81881644 - layer.4.v_cache 0.00000296 0.00197807 - layer.4.output 0.00017873 0.07004347 - ------------------------------------------------------------------------------------- - TOTAL 0.00265705 0.70083232 - (elements=2,824,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2824192 -Total Bytes 648260 -BPFP 1.8363 bits/point -EBPFP 3.6726 equivalent bits/point -MSE 0.700832 ----------------------- -------------------------------------------------------- -Time: 4.311s Load: 0.010s, Pack+Encode: 2.375s, Decode+Unpack: 1.925s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7008 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample20-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample20-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample21-layer4-item1.zst (26/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample21-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 182, 128) -Output shape: (1, 182, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.output: torch.Size([1, 182, 4096]) -> torch.Size([1, 1, 182, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,496B, BPFP=0.4505 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,416B, BPFP=2.1641 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,108B, BPFP=1.5500 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,360B, BPFP=2.2905 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,344B, BPFP=1.9464 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,700B, BPFP=2.3051 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,068B, BPFP=1.9346 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,308B, BPFP=2.2883 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,188B, BPFP=1.5963 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,276B, BPFP=2.3298 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 134,980B, BPFP=1.4485 -⌛️ [2/4] FRONTEND: Frontend time: 1.948s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.517s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02714607 6.71380079 - layer.0.v_cache 0.00000026 0.00014780 - layer.1.k_cache 0.00303102 0.82908647 - layer.1.v_cache 0.00000080 0.00053956 - layer.2.k_cache 0.00117008 0.36668488 - layer.2.v_cache 0.00000108 0.00073791 - layer.3.k_cache 0.00136151 0.41992439 - layer.3.v_cache 0.00000204 0.00120845 - layer.4.k_cache 0.00337920 0.87231881 - layer.4.v_cache 0.00000313 0.00208218 - layer.4.output 0.00018729 0.07013293 - ------------------------------------------------------------------------------------- - TOTAL 0.00263174 0.67764735 - (elements=2,609,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2609152 -Total Bytes 574244 -BPFP 1.7607 bits/point -EBPFP 3.5214 equivalent bits/point -MSE 0.677647 ----------------------- -------------------------------------------------------- -Time: 3.475s Load: 0.010s, Pack+Encode: 1.948s, Decode+Unpack: 1.517s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6776 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample21-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample21-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample22-layer4-item1.zst (27/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample22-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 191, 128) -Output shape: (1, 191, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.output: torch.Size([1, 191, 4096]) -> torch.Size([1, 1, 191, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,608B, BPFP=0.4339 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,384B, BPFP=2.0609 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,500B, BPFP=1.4930 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,020B, BPFP=2.1687 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,248B, BPFP=1.8099 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,772B, BPFP=2.1994 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,500B, BPFP=1.8202 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,336B, BPFP=2.1816 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,856B, BPFP=1.5075 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,868B, BPFP=2.2034 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 167,616B, BPFP=1.7140 -⌛️ [2/4] FRONTEND: Frontend time: 1.967s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.534s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02679409 7.00329654 - layer.0.v_cache 0.00000027 0.00014429 - layer.1.k_cache 0.00297761 0.94979195 - layer.1.v_cache 0.00000080 0.00050511 - layer.2.k_cache 0.00115400 0.37446147 - layer.2.v_cache 0.00000116 0.00068603 - layer.3.k_cache 0.00135265 0.45105851 - layer.3.v_cache 0.00000211 0.00110189 - layer.4.k_cache 0.00347104 0.79382508 - layer.4.v_cache 0.00000296 0.00181161 - layer.4.output 0.00020113 0.05534002 - ------------------------------------------------------------------------------------- - TOTAL 0.00261152 0.69986018 - (elements=2,738,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2738176 -Total Bytes 604708 -BPFP 1.7667 bits/point -EBPFP 3.5335 equivalent bits/point -MSE 0.699860 ----------------------- -------------------------------------------------------- -Time: 3.512s Load: 0.011s, Pack+Encode: 1.967s, Decode+Unpack: 1.534s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6999 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample22-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample22-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample23-layer4-item1.zst (28/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample23-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 185, 128) -Output shape: (1, 185, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.0.v_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.1.k_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.1.v_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.2.k_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.2.v_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.3.k_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.3.v_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.4.k_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.4.v_cache: torch.Size([1, 8, 185, 128]) -> torch.Size([1, 1, 185, 1024]) - layer.4.output: torch.Size([1, 185, 4096]) -> torch.Size([1, 1, 185, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,084B, BPFP=0.4258 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,740B, BPFP=2.1005 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,168B, BPFP=1.5274 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,236B, BPFP=2.2481 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,920B, BPFP=1.8970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,752B, BPFP=2.2699 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,788B, BPFP=1.8914 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,192B, BPFP=2.2463 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,776B, BPFP=1.5953 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,944B, BPFP=2.2780 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 132,516B, BPFP=1.3990 -⌛️ [2/4] FRONTEND: Frontend time: 1.937s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 185, 128]) - layer.0.v_cache: torch.Size([1, 8, 185, 128]) - layer.1.k_cache: torch.Size([1, 8, 185, 128]) - layer.1.v_cache: torch.Size([1, 8, 185, 128]) - layer.2.k_cache: torch.Size([1, 8, 185, 128]) - layer.2.v_cache: torch.Size([1, 8, 185, 128]) - layer.3.k_cache: torch.Size([1, 8, 185, 128]) - layer.3.v_cache: torch.Size([1, 8, 185, 128]) - layer.4.k_cache: torch.Size([1, 8, 185, 128]) - layer.4.v_cache: torch.Size([1, 8, 185, 128]) - layer.4.output: torch.Size([1, 185, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.572s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 185, 128]) - layer.0.v_cache: torch.Size([1, 8, 185, 128]) - layer.1.k_cache: torch.Size([1, 8, 185, 128]) - layer.1.v_cache: torch.Size([1, 8, 185, 128]) - layer.2.k_cache: torch.Size([1, 8, 185, 128]) - layer.2.v_cache: torch.Size([1, 8, 185, 128]) - layer.3.k_cache: torch.Size([1, 8, 185, 128]) - layer.3.v_cache: torch.Size([1, 8, 185, 128]) - layer.4.k_cache: torch.Size([1, 8, 185, 128]) - layer.4.v_cache: torch.Size([1, 8, 185, 128]) - layer.4.output: torch.Size([1, 185, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02696336 6.87293470 - layer.0.v_cache 0.00000026 0.00014625 - layer.1.k_cache 0.00303561 0.75725048 - layer.1.v_cache 0.00000078 0.00053526 - layer.2.k_cache 0.00115494 0.36920018 - layer.2.v_cache 0.00000115 0.00075968 - layer.3.k_cache 0.00132996 0.43614090 - layer.3.v_cache 0.00000212 0.00120409 - layer.4.k_cache 0.00355001 0.81589817 - layer.4.v_cache 0.00000299 0.00197865 - layer.4.output 0.00021000 0.06063253 - ------------------------------------------------------------------------------------- - TOTAL 0.00263437 0.67846989 - (elements=2,652,160) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2652160 -Total Bytes 570116 -BPFP 1.7197 bits/point -EBPFP 3.4394 equivalent bits/point -MSE 0.678470 ----------------------- -------------------------------------------------------- -Time: 3.519s Load: 0.010s, Pack+Encode: 1.937s, Decode+Unpack: 1.572s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 185, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6785 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample23-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample23-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample24-layer4-item1.zst (29/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample24-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 184, 128) -Output shape: (1, 184, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.output: torch.Size([1, 184, 4096]) -> torch.Size([1, 1, 184, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,904B, BPFP=0.4205 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,112B, BPFP=2.1277 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,120B, BPFP=1.5336 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,100B, BPFP=2.2546 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,084B, BPFP=1.9142 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,616B, BPFP=2.2765 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,064B, BPFP=1.9134 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,080B, BPFP=2.2537 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,720B, BPFP=1.6016 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,080B, BPFP=2.2962 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 121,276B, BPFP=1.2873 -⌛️ [2/4] FRONTEND: Frontend time: 1.988s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.output: torch.Size([1, 184, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.545s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.output: torch.Size([1, 184, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02665583 7.08816130 - layer.0.v_cache 0.00000026 0.00014679 - layer.1.k_cache 0.00299416 0.80004808 - layer.1.v_cache 0.00000077 0.00051635 - layer.2.k_cache 0.00115256 0.37950798 - layer.2.v_cache 0.00000113 0.00070964 - layer.3.k_cache 0.00134142 0.43164286 - layer.3.v_cache 0.00000202 0.00114780 - layer.4.k_cache 0.00346119 0.79732663 - layer.4.v_cache 0.00000292 0.00195271 - layer.4.output 0.00021714 0.07735402 - ------------------------------------------------------------------------------------- - TOTAL 0.00260577 0.70075545 - (elements=2,637,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2637824 -Total Bytes 559156 -BPFP 1.6958 bits/point -EBPFP 3.3916 equivalent bits/point -MSE 0.700755 ----------------------- -------------------------------------------------------- -Time: 3.543s Load: 0.010s, Pack+Encode: 1.988s, Decode+Unpack: 1.545s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7008 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample24-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample24-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample25-layer4-item1.zst (30/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample25-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 195, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 195, 128) -Output shape: (1, 195, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.0.v_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.1.k_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.1.v_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.2.k_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.2.v_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.3.k_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.3.v_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.4.k_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.4.v_cache: torch.Size([1, 8, 195, 128]) -> torch.Size([1, 1, 195, 1024]) - layer.4.output: torch.Size([1, 195, 4096]) -> torch.Size([1, 1, 195, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,620B, BPFP=0.4255 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 61,432B, BPFP=2.4612 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 38,068B, BPFP=1.5252 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 66,656B, BPFP=2.6705 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 48,488B, BPFP=1.9426 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 67,144B, BPFP=2.6901 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 52,696B, BPFP=2.1112 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 66,092B, BPFP=2.6479 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 40,016B, BPFP=1.6032 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 67,332B, BPFP=2.6976 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 149,712B, BPFP=1.4995 -⌛️ [2/4] FRONTEND: Frontend time: 2.426s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 195, 128]) - layer.0.v_cache: torch.Size([1, 8, 195, 128]) - layer.1.k_cache: torch.Size([1, 8, 195, 128]) - layer.1.v_cache: torch.Size([1, 8, 195, 128]) - layer.2.k_cache: torch.Size([1, 8, 195, 128]) - layer.2.v_cache: torch.Size([1, 8, 195, 128]) - layer.3.k_cache: torch.Size([1, 8, 195, 128]) - layer.3.v_cache: torch.Size([1, 8, 195, 128]) - layer.4.k_cache: torch.Size([1, 8, 195, 128]) - layer.4.v_cache: torch.Size([1, 8, 195, 128]) - layer.4.output: torch.Size([1, 195, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.842s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 195, 128]) - layer.0.v_cache: torch.Size([1, 8, 195, 128]) - layer.1.k_cache: torch.Size([1, 8, 195, 128]) - layer.1.v_cache: torch.Size([1, 8, 195, 128]) - layer.2.k_cache: torch.Size([1, 8, 195, 128]) - layer.2.v_cache: torch.Size([1, 8, 195, 128]) - layer.3.k_cache: torch.Size([1, 8, 195, 128]) - layer.3.v_cache: torch.Size([1, 8, 195, 128]) - layer.4.k_cache: torch.Size([1, 8, 195, 128]) - layer.4.v_cache: torch.Size([1, 8, 195, 128]) - layer.4.output: torch.Size([1, 195, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02622716 6.97000451 - layer.0.v_cache 0.00000027 0.00014457 - layer.1.k_cache 0.00300573 0.85508939 - layer.1.v_cache 0.00000076 0.00051536 - layer.2.k_cache 0.00116358 0.37724066 - layer.2.v_cache 0.00000112 0.00072409 - layer.3.k_cache 0.00131201 0.39292524 - layer.3.v_cache 0.00000217 0.00120891 - layer.4.k_cache 0.00348515 0.74583216 - layer.4.v_cache 0.00000317 0.00203842 - layer.4.output 0.00014777 0.06168414 - ------------------------------------------------------------------------------------- - TOTAL 0.00255658 0.68517571 - (elements=2,795,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2795520 -Total Bytes 668256 -BPFP 1.9124 bits/point -EBPFP 3.8247 equivalent bits/point -MSE 0.685176 ----------------------- -------------------------------------------------------- -Time: 4.278s Load: 0.011s, Pack+Encode: 2.426s, Decode+Unpack: 1.842s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 195, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 195, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6852 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample25-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample25-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample26-layer4-item1.zst (31/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample26-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 190, 128) -Output shape: (1, 190, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,744B, BPFP=0.4418 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,620B, BPFP=2.0814 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 37,420B, BPFP=1.5387 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,388B, BPFP=2.1952 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,372B, BPFP=1.8656 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,256B, BPFP=2.2309 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,908B, BPFP=1.8465 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,928B, BPFP=2.2174 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,824B, BPFP=1.5553 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,304B, BPFP=2.2329 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 167,244B, BPFP=1.7192 -⌛️ [2/4] FRONTEND: Frontend time: 1.981s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.541s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02720581 7.05531905 - layer.0.v_cache 0.00000028 0.00014489 - layer.1.k_cache 0.00301432 0.93940671 - layer.1.v_cache 0.00000084 0.00051559 - layer.2.k_cache 0.00115620 0.37400669 - layer.2.v_cache 0.00000110 0.00070922 - layer.3.k_cache 0.00139960 0.46390742 - layer.3.v_cache 0.00000211 0.00114488 - layer.4.k_cache 0.00346420 0.86764004 - layer.4.v_cache 0.00000301 0.00192146 - layer.4.output 0.00019765 0.07020793 - ------------------------------------------------------------------------------------- - TOTAL 0.00264558 0.71325341 - (elements=2,723,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2723840 -Total Bytes 610008 -BPFP 1.7916 bits/point -EBPFP 3.5832 equivalent bits/point -MSE 0.713253 ----------------------- -------------------------------------------------------- -Time: 3.532s Load: 0.010s, Pack+Encode: 1.981s, Decode+Unpack: 1.541s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7133 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample26-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample26-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample27-layer4-item1.zst (32/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample27-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 177, 128) -Output shape: (1, 177, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.output: torch.Size([1, 177, 4096]) -> torch.Size([1, 1, 177, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,068B, BPFP=0.4444 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,220B, BPFP=2.2166 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,108B, BPFP=1.5496 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,964B, BPFP=2.3377 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,932B, BPFP=1.9832 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,640B, BPFP=2.3676 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,980B, BPFP=1.9853 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,260B, BPFP=2.3508 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,460B, BPFP=1.6534 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,024B, BPFP=2.3845 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 128,836B, BPFP=1.4217 -⌛️ [2/4] FRONTEND: Frontend time: 1.936s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.547s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02683750 6.94390076 - layer.0.v_cache 0.00000026 0.00014841 - layer.1.k_cache 0.00307604 0.91763202 - layer.1.v_cache 0.00000078 0.00054097 - layer.2.k_cache 0.00123050 0.37816292 - layer.2.v_cache 0.00000110 0.00076170 - layer.3.k_cache 0.00133191 0.42124331 - layer.3.v_cache 0.00000206 0.00117825 - layer.4.k_cache 0.00350474 0.83406032 - layer.4.v_cache 0.00000300 0.00200350 - layer.4.output 0.00016181 0.05941516 - ------------------------------------------------------------------------------------- - TOTAL 0.00261679 0.69552091 - (elements=2,537,472) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2537472 -Total Bytes 565492 -BPFP 1.7829 bits/point -EBPFP 3.5657 equivalent bits/point -MSE 0.695521 ----------------------- -------------------------------------------------------- -Time: 3.492s Load: 0.009s, Pack+Encode: 1.936s, Decode+Unpack: 1.547s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6955 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample27-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample27-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample28-layer4-item1.zst (33/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample28-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 182, 128) -Output shape: (1, 182, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) -> torch.Size([1, 1, 182, 1024]) - layer.4.output: torch.Size([1, 182, 4096]) -> torch.Size([1, 1, 182, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,248B, BPFP=0.4399 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,188B, BPFP=2.1544 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,324B, BPFP=1.5592 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,232B, BPFP=2.2850 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,992B, BPFP=1.9313 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,000B, BPFP=2.3180 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,552B, BPFP=1.9124 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,356B, BPFP=2.2904 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 38,040B, BPFP=1.6329 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,396B, BPFP=2.3350 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 151,180B, BPFP=1.6224 -⌛️ [2/4] FRONTEND: Frontend time: 1.968s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.523s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 182, 128]) - layer.0.v_cache: torch.Size([1, 8, 182, 128]) - layer.1.k_cache: torch.Size([1, 8, 182, 128]) - layer.1.v_cache: torch.Size([1, 8, 182, 128]) - layer.2.k_cache: torch.Size([1, 8, 182, 128]) - layer.2.v_cache: torch.Size([1, 8, 182, 128]) - layer.3.k_cache: torch.Size([1, 8, 182, 128]) - layer.3.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.k_cache: torch.Size([1, 8, 182, 128]) - layer.4.v_cache: torch.Size([1, 8, 182, 128]) - layer.4.output: torch.Size([1, 182, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02800116 6.94655198 - layer.0.v_cache 0.00000026 0.00014631 - layer.1.k_cache 0.00305201 0.72828222 - layer.1.v_cache 0.00000078 0.00051872 - layer.2.k_cache 0.00115625 0.36933207 - layer.2.v_cache 0.00000114 0.00074210 - layer.3.k_cache 0.00134464 0.42380016 - layer.3.v_cache 0.00000201 0.00113225 - layer.4.k_cache 0.00345014 0.84468104 - layer.4.v_cache 0.00000301 0.00200279 - layer.4.output 0.00022774 0.07610559 - ------------------------------------------------------------------------------------- - TOTAL 0.00270874 0.68725800 - (elements=2,609,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2609152 -Total Bytes 590508 -BPFP 1.8106 bits/point -EBPFP 3.6211 equivalent bits/point -MSE 0.687258 ----------------------- -------------------------------------------------------- -Time: 3.501s Load: 0.010s, Pack+Encode: 1.968s, Decode+Unpack: 1.523s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 182, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6873 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample28-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample28-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample29-layer4-item1.zst (34/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample29-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 186, 128) -Output shape: (1, 186, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.0.v_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.1.k_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.1.v_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.2.k_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.2.v_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.3.k_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.3.v_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.4.k_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.4.v_cache: torch.Size([1, 8, 186, 128]) -> torch.Size([1, 1, 186, 1024]) - layer.4.output: torch.Size([1, 186, 4096]) -> torch.Size([1, 1, 186, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,124B, BPFP=0.4252 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,656B, BPFP=2.0857 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,500B, BPFP=1.5331 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,148B, BPFP=2.2324 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,308B, BPFP=1.9031 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,812B, BPFP=2.2602 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,628B, BPFP=1.8745 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,108B, BPFP=2.2307 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,264B, BPFP=1.5652 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,452B, BPFP=2.2871 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 141,032B, BPFP=1.4809 -⌛️ [2/4] FRONTEND: Frontend time: 1.979s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 186, 128]) - layer.0.v_cache: torch.Size([1, 8, 186, 128]) - layer.1.k_cache: torch.Size([1, 8, 186, 128]) - layer.1.v_cache: torch.Size([1, 8, 186, 128]) - layer.2.k_cache: torch.Size([1, 8, 186, 128]) - layer.2.v_cache: torch.Size([1, 8, 186, 128]) - layer.3.k_cache: torch.Size([1, 8, 186, 128]) - layer.3.v_cache: torch.Size([1, 8, 186, 128]) - layer.4.k_cache: torch.Size([1, 8, 186, 128]) - layer.4.v_cache: torch.Size([1, 8, 186, 128]) - layer.4.output: torch.Size([1, 186, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.559s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 186, 128]) - layer.0.v_cache: torch.Size([1, 8, 186, 128]) - layer.1.k_cache: torch.Size([1, 8, 186, 128]) - layer.1.v_cache: torch.Size([1, 8, 186, 128]) - layer.2.k_cache: torch.Size([1, 8, 186, 128]) - layer.2.v_cache: torch.Size([1, 8, 186, 128]) - layer.3.k_cache: torch.Size([1, 8, 186, 128]) - layer.3.v_cache: torch.Size([1, 8, 186, 128]) - layer.4.k_cache: torch.Size([1, 8, 186, 128]) - layer.4.v_cache: torch.Size([1, 8, 186, 128]) - layer.4.output: torch.Size([1, 186, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02748687 7.29145206 - layer.0.v_cache 0.00000027 0.00014733 - layer.1.k_cache 0.00295559 0.81049798 - layer.1.v_cache 0.00000084 0.00054103 - layer.2.k_cache 0.00114893 0.37515087 - layer.2.v_cache 0.00000110 0.00073777 - layer.3.k_cache 0.00136328 0.42881135 - layer.3.v_cache 0.00000212 0.00121196 - layer.4.k_cache 0.00348854 0.81258236 - layer.4.v_cache 0.00000314 0.00209157 - layer.4.output 0.00017293 0.06919784 - ------------------------------------------------------------------------------------- - TOTAL 0.00265303 0.71428683 - (elements=2,666,496) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2666496 -Total Bytes 579032 -BPFP 1.7372 bits/point -EBPFP 3.4744 equivalent bits/point -MSE 0.714287 ----------------------- -------------------------------------------------------- -Time: 3.548s Load: 0.010s, Pack+Encode: 1.979s, Decode+Unpack: 1.559s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 186, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7143 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample29-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample29-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample3-layer4-item1.zst (35/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample3-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 225, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 225, 128) -Output shape: (1, 225, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.0.v_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.1.k_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.1.v_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.2.k_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.2.v_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.3.k_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.3.v_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.4.k_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.4.v_cache: torch.Size([1, 8, 225, 128]) -> torch.Size([1, 1, 225, 1024]) - layer.4.output: torch.Size([1, 225, 4096]) -> torch.Size([1, 1, 225, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 12,940B, BPFP=0.4493 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 66,236B, BPFP=2.2999 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,452B, BPFP=1.5782 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,648B, BPFP=2.4183 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,736B, BPFP=2.0394 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,860B, BPFP=2.4604 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,020B, BPFP=2.0493 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,320B, BPFP=2.4417 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,304B, BPFP=1.6772 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,972B, BPFP=2.4990 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 177,880B, BPFP=1.5441 -⌛️ [2/4] FRONTEND: Frontend time: 2.331s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 225, 128]) - layer.0.v_cache: torch.Size([1, 8, 225, 128]) - layer.1.k_cache: torch.Size([1, 8, 225, 128]) - layer.1.v_cache: torch.Size([1, 8, 225, 128]) - layer.2.k_cache: torch.Size([1, 8, 225, 128]) - layer.2.v_cache: torch.Size([1, 8, 225, 128]) - layer.3.k_cache: torch.Size([1, 8, 225, 128]) - layer.3.v_cache: torch.Size([1, 8, 225, 128]) - layer.4.k_cache: torch.Size([1, 8, 225, 128]) - layer.4.v_cache: torch.Size([1, 8, 225, 128]) - layer.4.output: torch.Size([1, 225, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.917s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 225, 128]) - layer.0.v_cache: torch.Size([1, 8, 225, 128]) - layer.1.k_cache: torch.Size([1, 8, 225, 128]) - layer.1.v_cache: torch.Size([1, 8, 225, 128]) - layer.2.k_cache: torch.Size([1, 8, 225, 128]) - layer.2.v_cache: torch.Size([1, 8, 225, 128]) - layer.3.k_cache: torch.Size([1, 8, 225, 128]) - layer.3.v_cache: torch.Size([1, 8, 225, 128]) - layer.4.k_cache: torch.Size([1, 8, 225, 128]) - layer.4.v_cache: torch.Size([1, 8, 225, 128]) - layer.4.output: torch.Size([1, 225, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02545540 7.05843750 - layer.0.v_cache 0.00000025 0.00014141 - layer.1.k_cache 0.00297359 0.80058980 - layer.1.v_cache 0.00000080 0.00051353 - layer.2.k_cache 0.00113723 0.41048730 - layer.2.v_cache 0.00000119 0.00074794 - layer.3.k_cache 0.00129491 0.40990957 - layer.3.v_cache 0.00000221 0.00119488 - layer.4.k_cache 0.00357589 0.82379951 - layer.4.v_cache 0.00000317 0.00208792 - layer.4.output 0.00016372 0.05973825 - ------------------------------------------------------------------------------------- - TOTAL 0.00250711 0.69620445 - (elements=3,225,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3225600 -Total Bytes 751368 -BPFP 1.8635 bits/point -EBPFP 3.7270 equivalent bits/point -MSE 0.696204 ----------------------- -------------------------------------------------------- -Time: 4.260s Load: 0.012s, Pack+Encode: 2.331s, Decode+Unpack: 1.917s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 225, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 225, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6962 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample3-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample30-layer4-item1.zst (36/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample30-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 179, 128) -Output shape: (1, 179, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,884B, BPFP=0.4314 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,436B, BPFP=2.2013 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,440B, BPFP=1.5904 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,672B, BPFP=2.3425 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,812B, BPFP=1.9558 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,128B, BPFP=2.3624 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,996B, BPFP=1.9639 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,468B, BPFP=2.3336 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,748B, BPFP=1.6039 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,564B, BPFP=2.3815 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 130,728B, BPFP=1.4264 -⌛️ [2/4] FRONTEND: Frontend time: 1.974s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.578s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02765450 7.10626118 - layer.0.v_cache 0.00000027 0.00014804 - layer.1.k_cache 0.00304088 0.79356572 - layer.1.v_cache 0.00000084 0.00055213 - layer.2.k_cache 0.00119692 0.36621034 - layer.2.v_cache 0.00000108 0.00073492 - layer.3.k_cache 0.00131095 0.40320195 - layer.3.v_cache 0.00000209 0.00119649 - layer.4.k_cache 0.00344948 0.81230649 - layer.4.v_cache 0.00000315 0.00208966 - layer.4.output 0.00017252 0.06852312 - ------------------------------------------------------------------------------------- - TOTAL 0.00266787 0.69716853 - (elements=2,566,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2566144 -Total Bytes 569876 -BPFP 1.7766 bits/point -EBPFP 3.5532 equivalent bits/point -MSE 0.697169 ----------------------- -------------------------------------------------------- -Time: 3.562s Load: 0.009s, Pack+Encode: 1.974s, Decode+Unpack: 1.578s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6972 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample30-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample30-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample31-layer4-item1.zst (37/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample31-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 187, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 187, 128) -Output shape: (1, 187, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.0.v_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.1.k_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.1.v_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.2.k_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.2.v_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.3.k_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.3.v_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.4.k_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.4.v_cache: torch.Size([1, 8, 187, 128]) -> torch.Size([1, 1, 187, 1024]) - layer.4.output: torch.Size([1, 187, 4096]) -> torch.Size([1, 1, 187, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,080B, BPFP=0.4211 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,472B, BPFP=2.0668 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,776B, BPFP=1.5364 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,124B, BPFP=2.2194 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,660B, BPFP=1.8658 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,784B, BPFP=2.2470 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,172B, BPFP=1.8454 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,924B, BPFP=2.2111 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,496B, BPFP=1.5665 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,244B, BPFP=2.2662 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 122,220B, BPFP=1.2765 -⌛️ [2/4] FRONTEND: Frontend time: 1.965s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 187, 128]) - layer.0.v_cache: torch.Size([1, 8, 187, 128]) - layer.1.k_cache: torch.Size([1, 8, 187, 128]) - layer.1.v_cache: torch.Size([1, 8, 187, 128]) - layer.2.k_cache: torch.Size([1, 8, 187, 128]) - layer.2.v_cache: torch.Size([1, 8, 187, 128]) - layer.3.k_cache: torch.Size([1, 8, 187, 128]) - layer.3.v_cache: torch.Size([1, 8, 187, 128]) - layer.4.k_cache: torch.Size([1, 8, 187, 128]) - layer.4.v_cache: torch.Size([1, 8, 187, 128]) - layer.4.output: torch.Size([1, 187, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.547s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 187, 128]) - layer.0.v_cache: torch.Size([1, 8, 187, 128]) - layer.1.k_cache: torch.Size([1, 8, 187, 128]) - layer.1.v_cache: torch.Size([1, 8, 187, 128]) - layer.2.k_cache: torch.Size([1, 8, 187, 128]) - layer.2.v_cache: torch.Size([1, 8, 187, 128]) - layer.3.k_cache: torch.Size([1, 8, 187, 128]) - layer.3.v_cache: torch.Size([1, 8, 187, 128]) - layer.4.k_cache: torch.Size([1, 8, 187, 128]) - layer.4.v_cache: torch.Size([1, 8, 187, 128]) - layer.4.output: torch.Size([1, 187, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02664549 6.82787760 - layer.0.v_cache 0.00000027 0.00014806 - layer.1.k_cache 0.00298854 0.69479550 - layer.1.v_cache 0.00000078 0.00053719 - layer.2.k_cache 0.00115871 0.37862706 - layer.2.v_cache 0.00000111 0.00073685 - layer.3.k_cache 0.00133556 0.43245224 - layer.3.v_cache 0.00000211 0.00115753 - layer.4.k_cache 0.00339894 0.80598238 - layer.4.v_cache 0.00000310 0.00203132 - layer.4.output 0.00017976 0.06976004 - ------------------------------------------------------------------------------------- - TOTAL 0.00258955 0.67309899 - (elements=2,680,832) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2680832 -Total Bytes 558952 -BPFP 1.6680 bits/point -EBPFP 3.3360 equivalent bits/point -MSE 0.673099 ----------------------- -------------------------------------------------------- -Time: 3.521s Load: 0.010s, Pack+Encode: 1.965s, Decode+Unpack: 1.547s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 187, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 187, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6731 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample31-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample31-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample33-layer4-item1.zst (38/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample33-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 190, 128) -Output shape: (1, 190, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) -> torch.Size([1, 1, 190, 1024]) - layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,960B, BPFP=0.4507 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,948B, BPFP=2.0949 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,832B, BPFP=1.5145 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,472B, BPFP=2.1987 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,188B, BPFP=1.8581 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,296B, BPFP=2.2326 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,236B, BPFP=1.8600 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,824B, BPFP=2.2132 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 38,432B, BPFP=1.5803 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,328B, BPFP=2.2339 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 145,836B, BPFP=1.4991 -⌛️ [2/4] FRONTEND: Frontend time: 1.962s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.535s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 190, 128]) - layer.0.v_cache: torch.Size([1, 8, 190, 128]) - layer.1.k_cache: torch.Size([1, 8, 190, 128]) - layer.1.v_cache: torch.Size([1, 8, 190, 128]) - layer.2.k_cache: torch.Size([1, 8, 190, 128]) - layer.2.v_cache: torch.Size([1, 8, 190, 128]) - layer.3.k_cache: torch.Size([1, 8, 190, 128]) - layer.3.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.k_cache: torch.Size([1, 8, 190, 128]) - layer.4.v_cache: torch.Size([1, 8, 190, 128]) - layer.4.output: torch.Size([1, 190, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02674542 7.19506386 - layer.0.v_cache 0.00000026 0.00014723 - layer.1.k_cache 0.00305433 0.97575057 - layer.1.v_cache 0.00000079 0.00050988 - layer.2.k_cache 0.00120723 0.37746947 - layer.2.v_cache 0.00000116 0.00074711 - layer.3.k_cache 0.00130860 0.46477906 - layer.3.v_cache 0.00000216 0.00117322 - layer.4.k_cache 0.00354577 0.83502157 - layer.4.v_cache 0.00000304 0.00200006 - layer.4.output 0.00015926 0.06424867 - ------------------------------------------------------------------------------------- - TOTAL 0.00260756 0.72211834 - (elements=2,723,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2723840 -Total Bytes 589352 -BPFP 1.7309 bits/point -EBPFP 3.4619 equivalent bits/point -MSE 0.722118 ----------------------- -------------------------------------------------------- -Time: 3.508s Load: 0.012s, Pack+Encode: 1.962s, Decode+Unpack: 1.535s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 190, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7221 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample33-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample33-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample34-layer4-item1.zst (39/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample34-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,804B, BPFP=0.4479 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,208B, BPFP=2.2939 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,536B, BPFP=1.5779 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,296B, BPFP=2.4349 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,340B, BPFP=2.0715 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,872B, BPFP=2.4613 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,328B, BPFP=2.0252 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,304B, BPFP=2.4353 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,912B, BPFP=1.6864 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,416B, BPFP=2.4861 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 150,368B, BPFP=1.7175 -⌛️ [2/4] FRONTEND: Frontend time: 1.984s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.566s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02696512 6.84970860 - layer.0.v_cache 0.00000027 0.00015039 - layer.1.k_cache 0.00309770 0.83057827 - layer.1.v_cache 0.00000080 0.00054827 - layer.2.k_cache 0.00123733 0.36009863 - layer.2.v_cache 0.00000111 0.00075540 - layer.3.k_cache 0.00133744 0.41351189 - layer.3.v_cache 0.00000211 0.00120331 - layer.4.k_cache 0.00342193 0.75920863 - layer.4.v_cache 0.00000311 0.00211414 - layer.4.output 0.00017162 0.06673189 - ------------------------------------------------------------------------------------- - TOTAL 0.00262524 0.67748608 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 586384 -BPFP 1.9136 bits/point -EBPFP 3.8272 equivalent bits/point -MSE 0.677486 ----------------------- -------------------------------------------------------- -Time: 3.559s Load: 0.009s, Pack+Encode: 1.984s, Decode+Unpack: 1.566s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6775 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample34-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample34-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample35-layer4-item1.zst (40/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample35-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 191, 128) -Output shape: (1, 191, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) -> torch.Size([1, 1, 191, 1024]) - layer.4.output: torch.Size([1, 191, 4096]) -> torch.Size([1, 1, 191, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,832B, BPFP=0.4022 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,540B, BPFP=2.0263 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,344B, BPFP=1.4866 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,276B, BPFP=2.1383 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,916B, BPFP=1.7963 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,836B, BPFP=2.1612 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,048B, BPFP=1.8017 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,412B, BPFP=2.1438 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,716B, BPFP=1.5018 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,140B, BPFP=2.1736 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 114,476B, BPFP=1.1706 -⌛️ [2/4] FRONTEND: Frontend time: 1.958s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.561s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 191, 128]) - layer.0.v_cache: torch.Size([1, 8, 191, 128]) - layer.1.k_cache: torch.Size([1, 8, 191, 128]) - layer.1.v_cache: torch.Size([1, 8, 191, 128]) - layer.2.k_cache: torch.Size([1, 8, 191, 128]) - layer.2.v_cache: torch.Size([1, 8, 191, 128]) - layer.3.k_cache: torch.Size([1, 8, 191, 128]) - layer.3.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.k_cache: torch.Size([1, 8, 191, 128]) - layer.4.v_cache: torch.Size([1, 8, 191, 128]) - layer.4.output: torch.Size([1, 191, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02691182 6.39480223 - layer.0.v_cache 0.00000026 0.00013510 - layer.1.k_cache 0.00302612 0.82335875 - layer.1.v_cache 0.00000075 0.00049214 - layer.2.k_cache 0.00114460 0.38600910 - layer.2.v_cache 0.00000105 0.00065747 - layer.3.k_cache 0.00140477 0.44441071 - layer.3.v_cache 0.00000198 0.00107215 - layer.4.k_cache 0.00340838 0.83042868 - layer.4.v_cache 0.00000288 0.00174513 - layer.4.output 0.00019978 0.06314575 - ------------------------------------------------------------------------------------- - TOTAL 0.00262155 0.65254960 - (elements=2,738,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2738176 -Total Bytes 545536 -BPFP 1.5939 bits/point -EBPFP 3.1877 equivalent bits/point -MSE 0.652550 ----------------------- -------------------------------------------------------- -Time: 3.528s Load: 0.009s, Pack+Encode: 1.958s, Decode+Unpack: 1.561s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 191, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6525 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample35-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample35-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample36-layer4-item1.zst (41/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample36-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 178, 128) -Output shape: (1, 178, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,752B, BPFP=0.4280 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,948B, BPFP=2.1922 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,560B, BPFP=1.5607 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,284B, BPFP=2.3387 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,800B, BPFP=1.9663 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,872B, BPFP=2.3645 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,828B, BPFP=1.9675 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,168B, BPFP=2.3336 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,156B, BPFP=1.6308 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,480B, BPFP=2.3912 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 120,148B, BPFP=1.3183 -⌛️ [2/4] FRONTEND: Frontend time: 2.008s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.524s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02810422 6.78813943 - layer.0.v_cache 0.00000028 0.00014750 - layer.1.k_cache 0.00299977 0.82680606 - layer.1.v_cache 0.00000079 0.00053793 - layer.2.k_cache 0.00115350 0.37134758 - layer.2.v_cache 0.00000110 0.00074516 - layer.3.k_cache 0.00134504 0.42526297 - layer.3.v_cache 0.00000210 0.00118874 - layer.4.k_cache 0.00349525 0.82840634 - layer.4.v_cache 0.00000313 0.00208518 - layer.4.output 0.00017533 0.06181130 - ------------------------------------------------------------------------------------- - TOTAL 0.00270046 0.67799372 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 556996 -BPFP 1.7462 bits/point -EBPFP 3.4924 equivalent bits/point -MSE 0.677994 ----------------------- -------------------------------------------------------- -Time: 3.541s Load: 0.009s, Pack+Encode: 2.008s, Decode+Unpack: 1.524s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6780 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample36-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample37-layer4-item1.zst (42/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample37-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 170, 128) -Output shape: (1, 170, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.output: torch.Size([1, 170, 4096]) -> torch.Size([1, 1, 170, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,700B, BPFP=0.4458 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,212B, BPFP=2.3075 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,816B, BPFP=1.6000 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,968B, BPFP=2.4342 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,936B, BPFP=2.0191 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,428B, BPFP=2.4553 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,456B, BPFP=2.0430 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,764B, BPFP=2.4248 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,076B, BPFP=1.7039 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,860B, BPFP=2.4752 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 142,616B, BPFP=1.6385 -⌛️ [2/4] FRONTEND: Frontend time: 1.968s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.output: torch.Size([1, 170, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.574s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.output: torch.Size([1, 170, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02684929 6.70129754 - layer.0.v_cache 0.00000027 0.00014927 - layer.1.k_cache 0.00302298 0.86466082 - layer.1.v_cache 0.00000079 0.00052666 - layer.2.k_cache 0.00119467 0.35959912 - layer.2.v_cache 0.00000116 0.00071880 - layer.3.k_cache 0.00130934 0.40094735 - layer.3.v_cache 0.00000210 0.00115809 - layer.4.k_cache 0.00343955 0.78348425 - layer.4.v_cache 0.00000302 0.00197428 - layer.4.output 0.00019176 0.07711284 - ------------------------------------------------------------------------------------- - TOTAL 0.00261359 0.67306911 - (elements=2,437,120) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2437120 -Total Bytes 575832 -BPFP 1.8902 bits/point -EBPFP 3.7804 equivalent bits/point -MSE 0.673069 ----------------------- -------------------------------------------------------- -Time: 3.552s Load: 0.009s, Pack+Encode: 1.968s, Decode+Unpack: 1.574s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6731 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample37-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample37-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample38-layer4-item1.zst (43/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample38-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 175, 128) -Output shape: (1, 175, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,676B, BPFP=0.4320 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,928B, BPFP=2.2289 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,180B, BPFP=1.6152 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,000B, BPFP=2.3661 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,068B, BPFP=2.0120 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,552B, BPFP=2.3907 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,752B, BPFP=1.9979 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,996B, BPFP=2.3659 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,332B, BPFP=1.6220 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,996B, BPFP=2.4105 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 128,996B, BPFP=1.4397 -⌛️ [2/4] FRONTEND: Frontend time: 1.962s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.556s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02712274 6.94639021 - layer.0.v_cache 0.00000026 0.00014788 - layer.1.k_cache 0.00313634 0.84926985 - layer.1.v_cache 0.00000079 0.00053668 - layer.2.k_cache 0.00115072 0.37888284 - layer.2.v_cache 0.00000109 0.00072281 - layer.3.k_cache 0.00137237 0.42463780 - layer.3.v_cache 0.00000208 0.00119479 - layer.4.k_cache 0.00360495 0.82820391 - layer.4.v_cache 0.00000302 0.00205161 - layer.4.output 0.00022301 0.08328599 - ------------------------------------------------------------------------------------- - TOTAL 0.00266332 0.69751302 - (elements=2,508,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2508800 -Total Bytes 564476 -BPFP 1.8000 bits/point -EBPFP 3.6000 equivalent bits/point -MSE 0.697513 ----------------------- -------------------------------------------------------- -Time: 3.526s Load: 0.009s, Pack+Encode: 1.962s, Decode+Unpack: 1.556s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6975 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample38-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample39-layer4-item1.zst (44/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample39-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 177, 128) -Output shape: (1, 177, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.output: torch.Size([1, 177, 4096]) -> torch.Size([1, 1, 177, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,112B, BPFP=0.4463 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,388B, BPFP=2.2240 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,300B, BPFP=1.5581 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,604B, BPFP=2.3660 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,192B, BPFP=1.9947 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,052B, BPFP=2.3858 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,724B, BPFP=1.9740 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,504B, BPFP=2.3616 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,928B, BPFP=1.6299 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,460B, BPFP=2.4038 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 145,912B, BPFP=1.6101 -⌛️ [2/4] FRONTEND: Frontend time: 1.988s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.532s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02784215 7.07476841 - layer.0.v_cache 0.00000027 0.00015286 - layer.1.k_cache 0.00307979 0.89094293 - layer.1.v_cache 0.00000080 0.00054662 - layer.2.k_cache 0.00119829 0.37417732 - layer.2.v_cache 0.00000109 0.00074273 - layer.3.k_cache 0.00135088 0.40785032 - layer.3.v_cache 0.00000204 0.00117693 - layer.4.k_cache 0.00346965 0.79639353 - layer.4.v_cache 0.00000308 0.00205859 - layer.4.output 0.00016766 0.06593838 - ------------------------------------------------------------------------------------- - TOTAL 0.00268705 0.70089741 - (elements=2,537,472) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2537472 -Total Bytes 584176 -BPFP 1.8418 bits/point -EBPFP 3.6835 equivalent bits/point -MSE 0.700897 ----------------------- -------------------------------------------------------- -Time: 3.530s Load: 0.010s, Pack+Encode: 1.988s, Decode+Unpack: 1.532s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7009 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample39-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample39-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample4-layer4-item1.zst (45/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample4-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 258, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.014s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 258, 128) -Output shape: (1, 258, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.0.v_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.1.k_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.1.v_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.2.k_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.2.v_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.3.k_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.3.v_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.4.k_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.4.v_cache: torch.Size([1, 8, 258, 128]) -> torch.Size([1, 1, 258, 1024]) - layer.4.output: torch.Size([1, 258, 4096]) -> torch.Size([1, 1, 258, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 14,444B, BPFP=0.4374 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 77,616B, BPFP=2.3503 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 51,324B, BPFP=1.5541 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 83,732B, BPFP=2.5355 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 62,568B, BPFP=1.8946 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 84,556B, BPFP=2.5604 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 67,880B, BPFP=2.0555 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 83,544B, BPFP=2.5298 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 51,780B, BPFP=1.5680 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 85,136B, BPFP=2.5780 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 192,212B, BPFP=1.4551 -⌛️ [2/4] FRONTEND: Frontend time: 2.375s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 258, 128]) - layer.0.v_cache: torch.Size([1, 8, 258, 128]) - layer.1.k_cache: torch.Size([1, 8, 258, 128]) - layer.1.v_cache: torch.Size([1, 8, 258, 128]) - layer.2.k_cache: torch.Size([1, 8, 258, 128]) - layer.2.v_cache: torch.Size([1, 8, 258, 128]) - layer.3.k_cache: torch.Size([1, 8, 258, 128]) - layer.3.v_cache: torch.Size([1, 8, 258, 128]) - layer.4.k_cache: torch.Size([1, 8, 258, 128]) - layer.4.v_cache: torch.Size([1, 8, 258, 128]) - layer.4.output: torch.Size([1, 258, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.779s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 258, 128]) - layer.0.v_cache: torch.Size([1, 8, 258, 128]) - layer.1.k_cache: torch.Size([1, 8, 258, 128]) - layer.1.v_cache: torch.Size([1, 8, 258, 128]) - layer.2.k_cache: torch.Size([1, 8, 258, 128]) - layer.2.v_cache: torch.Size([1, 8, 258, 128]) - layer.3.k_cache: torch.Size([1, 8, 258, 128]) - layer.3.v_cache: torch.Size([1, 8, 258, 128]) - layer.4.k_cache: torch.Size([1, 8, 258, 128]) - layer.4.v_cache: torch.Size([1, 8, 258, 128]) - layer.4.output: torch.Size([1, 258, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02582260 7.08598813 - layer.0.v_cache 0.00000026 0.00014304 - layer.1.k_cache 0.00299646 0.77326954 - layer.1.v_cache 0.00000083 0.00052807 - layer.2.k_cache 0.00117653 0.37131483 - layer.2.v_cache 0.00000118 0.00075522 - layer.3.k_cache 0.00132295 0.40819419 - layer.3.v_cache 0.00000232 0.00124671 - layer.4.k_cache 0.00348270 0.77293532 - layer.4.v_cache 0.00000323 0.00212881 - layer.4.output 0.00015595 0.06221960 - ------------------------------------------------------------------------------------- - TOTAL 0.00253092 0.69038445 - (elements=3,698,688) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3698688 -Total Bytes 854792 -BPFP 1.8489 bits/point -EBPFP 3.6977 equivalent bits/point -MSE 0.690384 ----------------------- -------------------------------------------------------- -Time: 4.167s Load: 0.014s, Pack+Encode: 2.375s, Decode+Unpack: 1.779s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 258, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 258, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6904 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample4-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample4-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample40-layer4-item1.zst (46/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample40-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,740B, BPFP=0.4398 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,084B, BPFP=2.2617 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,316B, BPFP=1.5948 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,164B, BPFP=2.4008 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,264B, BPFP=2.0441 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,772B, BPFP=2.4283 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,696B, BPFP=2.0184 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,240B, BPFP=2.4043 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,872B, BPFP=1.6651 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,268B, BPFP=2.4507 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 126,528B, BPFP=1.4285 -⌛️ [2/4] FRONTEND: Frontend time: 1.932s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.543s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02754376 6.93911090 - layer.0.v_cache 0.00000027 0.00015081 - layer.1.k_cache 0.00303993 0.80446594 - layer.1.v_cache 0.00000078 0.00053684 - layer.2.k_cache 0.00116504 0.37207922 - layer.2.v_cache 0.00000115 0.00076311 - layer.3.k_cache 0.00135444 0.43021728 - layer.3.v_cache 0.00000213 0.00121824 - layer.4.k_cache 0.00351023 0.81441978 - layer.4.v_cache 0.00000311 0.00208374 - layer.4.output 0.00018471 0.06969470 - ------------------------------------------------------------------------------------- - TOTAL 0.00266855 0.68884462 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 562944 -BPFP 1.8159 bits/point -EBPFP 3.6317 equivalent bits/point -MSE 0.688845 ----------------------- -------------------------------------------------------- -Time: 3.484s Load: 0.009s, Pack+Encode: 1.932s, Decode+Unpack: 1.543s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6888 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample40-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample41-layer4-item1.zst (47/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample41-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 179, 128) -Output shape: (1, 179, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,100B, BPFP=0.4408 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,296B, BPFP=2.1952 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,796B, BPFP=1.5623 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,284B, BPFP=2.3256 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,700B, BPFP=1.9509 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,956B, BPFP=2.3549 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,660B, BPFP=1.9492 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,196B, BPFP=2.3218 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,616B, BPFP=1.6418 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,372B, BPFP=2.3731 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 131,768B, BPFP=1.4378 -⌛️ [2/4] FRONTEND: Frontend time: 1.986s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.520s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02616998 6.81241885 - layer.0.v_cache 0.00000026 0.00014779 - layer.1.k_cache 0.00300457 0.80249313 - layer.1.v_cache 0.00000081 0.00053185 - layer.2.k_cache 0.00116479 0.37708854 - layer.2.v_cache 0.00000111 0.00073574 - layer.3.k_cache 0.00135393 0.41620338 - layer.3.v_cache 0.00000203 0.00117506 - layer.4.k_cache 0.00348554 0.82574659 - layer.4.v_cache 0.00000307 0.00204255 - layer.4.output 0.00016512 0.07118480 - ------------------------------------------------------------------------------------- - TOTAL 0.00256047 0.68023734 - (elements=2,566,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2566144 -Total Bytes 569744 -BPFP 1.7762 bits/point -EBPFP 3.5524 equivalent bits/point -MSE 0.680237 ----------------------- -------------------------------------------------------- -Time: 3.516s Load: 0.010s, Pack+Encode: 1.986s, Decode+Unpack: 1.520s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6802 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample41-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample41-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample42-layer4-item1.zst (48/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample42-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 157, 128) -Output shape: (1, 157, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,952B, BPFP=0.4455 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,592B, BPFP=2.5175 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,840B, BPFP=1.6839 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,000B, BPFP=2.6373 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,412B, BPFP=2.2100 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,732B, BPFP=2.6738 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,268B, BPFP=2.2028 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,372B, BPFP=2.6559 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,752B, BPFP=1.7791 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,312B, BPFP=2.7026 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 158,784B, BPFP=1.9753 -⌛️ [2/4] FRONTEND: Frontend time: 1.966s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.541s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02756658 7.59350197 - layer.0.v_cache 0.00000027 0.00015214 - layer.1.k_cache 0.00308485 0.82845700 - layer.1.v_cache 0.00000086 0.00054346 - layer.2.k_cache 0.00117752 0.36114679 - layer.2.v_cache 0.00000111 0.00074712 - layer.3.k_cache 0.00136410 0.40840752 - layer.3.v_cache 0.00000207 0.00117299 - layer.4.k_cache 0.00338255 0.86308007 - layer.4.v_cache 0.00000300 0.00209050 - layer.4.output 0.00016164 0.06830261 - ------------------------------------------------------------------------------------- - TOTAL 0.00265925 0.73803643 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 591016 -BPFP 2.1007 bits/point -EBPFP 4.2014 equivalent bits/point -MSE 0.738036 ----------------------- -------------------------------------------------------- -Time: 3.515s Load: 0.008s, Pack+Encode: 1.966s, Decode+Unpack: 1.541s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7380 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample42-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample43-layer4-item1.zst (49/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample43-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 174, 128) -Output shape: (1, 174, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.output: torch.Size([1, 174, 4096]) -> torch.Size([1, 1, 174, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,020B, BPFP=0.4499 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,436B, BPFP=2.2645 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,704B, BPFP=1.6031 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,592B, BPFP=2.4062 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,308B, BPFP=2.0343 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,452B, BPFP=2.4449 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,244B, BPFP=2.0314 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,908B, BPFP=2.4204 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,316B, BPFP=1.6755 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,700B, BPFP=2.4560 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 164,024B, BPFP=1.8411 -⌛️ [2/4] FRONTEND: Frontend time: 1.963s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.output: torch.Size([1, 174, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.563s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.output: torch.Size([1, 174, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02674412 7.12409359 - layer.0.v_cache 0.00000027 0.00015407 - layer.1.k_cache 0.00309791 0.82597491 - layer.1.v_cache 0.00000087 0.00055423 - layer.2.k_cache 0.00117281 0.36865011 - layer.2.v_cache 0.00000114 0.00076804 - layer.3.k_cache 0.00132088 0.41554168 - layer.3.v_cache 0.00000225 0.00125778 - layer.4.k_cache 0.00347387 0.80772444 - layer.4.v_cache 0.00000314 0.00204439 - layer.4.output 0.00018404 0.07367041 - ------------------------------------------------------------------------------------- - TOTAL 0.00261096 0.70296035 - (elements=2,494,464) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2494464 -Total Bytes 604704 -BPFP 1.9393 bits/point -EBPFP 3.8787 equivalent bits/point -MSE 0.702960 ----------------------- -------------------------------------------------------- -Time: 3.535s Load: 0.009s, Pack+Encode: 1.963s, Decode+Unpack: 1.563s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7030 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample43-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample43-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample44-layer4-item1.zst (50/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample44-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 177, 128) -Output shape: (1, 177, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) -> torch.Size([1, 1, 177, 1024]) - layer.4.output: torch.Size([1, 177, 4096]) -> torch.Size([1, 1, 177, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,864B, BPFP=0.4354 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,092B, BPFP=2.2110 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,676B, BPFP=1.5747 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,176B, BPFP=2.3471 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,644B, BPFP=1.9705 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,888B, BPFP=2.3785 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,764B, BPFP=1.9758 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,176B, BPFP=2.3471 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,776B, BPFP=1.6674 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,324B, BPFP=2.3978 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 119,732B, BPFP=1.3212 -⌛️ [2/4] FRONTEND: Frontend time: 1.959s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.571s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 177, 128]) - layer.0.v_cache: torch.Size([1, 8, 177, 128]) - layer.1.k_cache: torch.Size([1, 8, 177, 128]) - layer.1.v_cache: torch.Size([1, 8, 177, 128]) - layer.2.k_cache: torch.Size([1, 8, 177, 128]) - layer.2.v_cache: torch.Size([1, 8, 177, 128]) - layer.3.k_cache: torch.Size([1, 8, 177, 128]) - layer.3.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.k_cache: torch.Size([1, 8, 177, 128]) - layer.4.v_cache: torch.Size([1, 8, 177, 128]) - layer.4.output: torch.Size([1, 177, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02710410 6.95970300 - layer.0.v_cache 0.00000026 0.00014973 - layer.1.k_cache 0.00310284 0.82655671 - layer.1.v_cache 0.00000080 0.00053528 - layer.2.k_cache 0.00117362 0.37507246 - layer.2.v_cache 0.00000106 0.00072847 - layer.3.k_cache 0.00136356 0.42976961 - layer.3.v_cache 0.00000205 0.00115952 - layer.4.k_cache 0.00349505 0.83073279 - layer.4.v_cache 0.00000305 0.00203158 - layer.4.output 0.00021483 0.06981899 - ------------------------------------------------------------------------------------- - TOTAL 0.00265041 0.69326537 - (elements=2,537,472) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2537472 -Total Bytes 557112 -BPFP 1.7564 bits/point -EBPFP 3.5129 equivalent bits/point -MSE 0.693265 ----------------------- -------------------------------------------------------- -Time: 3.540s Load: 0.010s, Pack+Encode: 1.959s, Decode+Unpack: 1.571s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 177, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 177, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6933 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample44-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample45-layer4-item1.zst (51/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample45-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 165, 128) -Output shape: (1, 165, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.output: torch.Size([1, 165, 4096]) -> torch.Size([1, 1, 165, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,864B, BPFP=0.4197 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 48,780B, BPFP=2.3097 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,132B, BPFP=1.5688 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 50,916B, BPFP=2.4108 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,896B, BPFP=2.0784 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,020B, BPFP=2.4631 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,140B, BPFP=2.0900 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,328B, BPFP=2.4303 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,924B, BPFP=1.6536 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 52,480B, BPFP=2.4848 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 98,272B, BPFP=1.1633 -⌛️ [2/4] FRONTEND: Frontend time: 1.954s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.559s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02820997 7.83623861 - layer.0.v_cache 0.00000025 0.00014890 - layer.1.k_cache 0.00309343 0.70965474 - layer.1.v_cache 0.00000073 0.00049634 - layer.2.k_cache 0.00118989 0.40632740 - layer.2.v_cache 0.00000098 0.00066577 - layer.3.k_cache 0.00136500 0.43193785 - layer.3.v_cache 0.00000190 0.00109513 - layer.4.k_cache 0.00352692 0.91430248 - layer.4.v_cache 0.00000287 0.00192995 - layer.4.output 0.00017443 0.06569042 - ------------------------------------------------------------------------------------- - TOTAL 0.00272069 0.75468277 - (elements=2,365,440) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2365440 -Total Bytes 518752 -BPFP 1.7544 bits/point -EBPFP 3.5089 equivalent bits/point -MSE 0.754683 ----------------------- -------------------------------------------------------- -Time: 3.522s Load: 0.009s, Pack+Encode: 1.954s, Decode+Unpack: 1.559s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7547 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample45-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample45-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample47-layer4-item1.zst (52/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample47-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 175, 128) -Output shape: (1, 175, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,484B, BPFP=0.4234 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,424B, BPFP=2.2064 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,388B, BPFP=1.6245 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,216B, BPFP=2.3311 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,044B, BPFP=2.0109 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,136B, BPFP=2.3721 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,740B, BPFP=1.9973 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,476B, BPFP=2.3427 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,448B, BPFP=1.6271 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,516B, BPFP=2.3891 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 108,484B, BPFP=1.2108 -⌛️ [2/4] FRONTEND: Frontend time: 2.004s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.573s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02684375 6.95224470 - layer.0.v_cache 0.00000026 0.00014111 - layer.1.k_cache 0.00307220 0.77180926 - layer.1.v_cache 0.00000076 0.00051746 - layer.2.k_cache 0.00114531 0.36294146 - layer.2.v_cache 0.00000107 0.00071607 - layer.3.k_cache 0.00135081 0.43734645 - layer.3.v_cache 0.00000217 0.00119235 - layer.4.k_cache 0.00336849 0.85833278 - layer.4.v_cache 0.00000291 0.00199472 - layer.4.output 0.00018329 0.07231701 - ------------------------------------------------------------------------------------- - TOTAL 0.00260863 0.69117889 - (elements=2,508,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2508800 -Total Bytes 541356 -BPFP 1.7263 bits/point -EBPFP 3.4525 equivalent bits/point -MSE 0.691179 ----------------------- -------------------------------------------------------- -Time: 3.586s Load: 0.009s, Pack+Encode: 2.004s, Decode+Unpack: 1.573s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6912 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample47-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample47-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample48-layer4-item1.zst (53/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample48-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 179, 128) -Output shape: (1, 179, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,752B, BPFP=0.4256 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,024B, BPFP=2.1833 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,268B, BPFP=1.5829 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,264B, BPFP=2.3247 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,764B, BPFP=1.9537 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,924B, BPFP=2.3535 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,636B, BPFP=1.9481 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,380B, BPFP=2.3298 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,336B, BPFP=1.6295 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,424B, BPFP=2.3753 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 139,128B, BPFP=1.5181 -⌛️ [2/4] FRONTEND: Frontend time: 1.974s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02697467 6.99457912 - layer.0.v_cache 0.00000027 0.00014666 - layer.1.k_cache 0.00293179 0.78247190 - layer.1.v_cache 0.00000082 0.00052788 - layer.2.k_cache 0.00115626 0.37566111 - layer.2.v_cache 0.00000116 0.00074601 - layer.3.k_cache 0.00134185 0.41918353 - layer.3.v_cache 0.00000205 0.00117454 - layer.4.k_cache 0.00342791 0.81917866 - layer.4.v_cache 0.00000307 0.00201922 - layer.4.output 0.00018990 0.07272565 - ------------------------------------------------------------------------------------- - TOTAL 0.00261425 0.69189937 - (elements=2,566,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2566144 -Total Bytes 576900 -BPFP 1.7985 bits/point -EBPFP 3.5970 equivalent bits/point -MSE 0.691899 ----------------------- -------------------------------------------------------- -Time: 3.536s Load: 0.010s, Pack+Encode: 1.974s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6919 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample48-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample49-layer4-item1.zst (54/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample49-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 169, 128) -Output shape: (1, 169, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.output: torch.Size([1, 169, 4096]) -> torch.Size([1, 1, 169, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,456B, BPFP=0.4371 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,008B, BPFP=2.3118 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,740B, BPFP=1.6060 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,452B, BPFP=2.4710 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,212B, BPFP=2.0901 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,980B, BPFP=2.4954 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,652B, BPFP=2.0642 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,368B, BPFP=2.4671 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,372B, BPFP=1.7276 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,356B, BPFP=2.5128 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 132,184B, BPFP=1.5276 -⌛️ [2/4] FRONTEND: Frontend time: 1.975s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.581s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02698109 7.42211264 - layer.0.v_cache 0.00000027 0.00015589 - layer.1.k_cache 0.00303223 0.78575911 - layer.1.v_cache 0.00000079 0.00055553 - layer.2.k_cache 0.00117179 0.36626042 - layer.2.v_cache 0.00000111 0.00076851 - layer.3.k_cache 0.00133305 0.41274568 - layer.3.v_cache 0.00000207 0.00119066 - layer.4.k_cache 0.00348428 0.82823416 - layer.4.v_cache 0.00000320 0.00217070 - layer.4.output 0.00019031 0.08021524 - ------------------------------------------------------------------------------------- - TOTAL 0.00262651 0.72434388 - (elements=2,422,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2422784 -Total Bytes 568780 -BPFP 1.8781 bits/point -EBPFP 3.7562 equivalent bits/point -MSE 0.724344 ----------------------- -------------------------------------------------------- -Time: 3.565s Load: 0.009s, Pack+Encode: 1.975s, Decode+Unpack: 1.581s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7243 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample49-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample49-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample5-layer4-item1.zst (55/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample5-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 221, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 221, 128) -Output shape: (1, 221, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.0.v_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.1.k_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.1.v_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.2.k_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.2.v_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.3.k_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.3.v_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.4.k_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.4.v_cache: torch.Size([1, 8, 221, 128]) -> torch.Size([1, 1, 221, 1024]) - layer.4.output: torch.Size([1, 221, 4096]) -> torch.Size([1, 1, 221, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,996B, BPFP=0.4241 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 66,360B, BPFP=2.3459 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,888B, BPFP=1.6222 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,684B, BPFP=2.4634 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,960B, BPFP=2.0843 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,624B, BPFP=2.4966 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,552B, BPFP=2.0699 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,100B, BPFP=2.4781 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,752B, BPFP=1.6881 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,708B, BPFP=2.5349 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 153,840B, BPFP=1.3596 -⌛️ [2/4] FRONTEND: Frontend time: 2.377s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 221, 128]) - layer.0.v_cache: torch.Size([1, 8, 221, 128]) - layer.1.k_cache: torch.Size([1, 8, 221, 128]) - layer.1.v_cache: torch.Size([1, 8, 221, 128]) - layer.2.k_cache: torch.Size([1, 8, 221, 128]) - layer.2.v_cache: torch.Size([1, 8, 221, 128]) - layer.3.k_cache: torch.Size([1, 8, 221, 128]) - layer.3.v_cache: torch.Size([1, 8, 221, 128]) - layer.4.k_cache: torch.Size([1, 8, 221, 128]) - layer.4.v_cache: torch.Size([1, 8, 221, 128]) - layer.4.output: torch.Size([1, 221, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.026s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 221, 128]) - layer.0.v_cache: torch.Size([1, 8, 221, 128]) - layer.1.k_cache: torch.Size([1, 8, 221, 128]) - layer.1.v_cache: torch.Size([1, 8, 221, 128]) - layer.2.k_cache: torch.Size([1, 8, 221, 128]) - layer.2.v_cache: torch.Size([1, 8, 221, 128]) - layer.3.k_cache: torch.Size([1, 8, 221, 128]) - layer.3.v_cache: torch.Size([1, 8, 221, 128]) - layer.4.k_cache: torch.Size([1, 8, 221, 128]) - layer.4.v_cache: torch.Size([1, 8, 221, 128]) - layer.4.output: torch.Size([1, 221, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02713613 6.96858982 - layer.0.v_cache 0.00000026 0.00014350 - layer.1.k_cache 0.00294223 0.82950202 - layer.1.v_cache 0.00000078 0.00050263 - layer.2.k_cache 0.00116104 0.36249629 - layer.2.v_cache 0.00000109 0.00070701 - layer.3.k_cache 0.00132333 0.41321036 - layer.3.v_cache 0.00000207 0.00115414 - layer.4.k_cache 0.00348143 0.84339125 - layer.4.v_cache 0.00000308 0.00201947 - layer.4.output 0.00015143 0.06395576 - ------------------------------------------------------------------------------------- - TOTAL 0.00261837 0.69125282 - (elements=3,168,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3168256 -Total Bytes 725464 -BPFP 1.8318 bits/point -EBPFP 3.6637 equivalent bits/point -MSE 0.691253 ----------------------- -------------------------------------------------------- -Time: 4.416s Load: 0.012s, Pack+Encode: 2.377s, Decode+Unpack: 2.026s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 221, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 221, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6913 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample5-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample5-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample50-layer4-item1.zst (56/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample50-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 175, 128) -Output shape: (1, 175, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,368B, BPFP=0.4182 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,416B, BPFP=2.2061 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,400B, BPFP=1.6250 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,712B, BPFP=2.3532 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,548B, BPFP=1.9888 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,544B, BPFP=2.3904 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,580B, BPFP=1.9902 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,900B, BPFP=2.3616 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,996B, BPFP=1.6070 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,976B, BPFP=2.4096 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 123,792B, BPFP=1.3816 -⌛️ [2/4] FRONTEND: Frontend time: 1.965s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.591s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02716402 7.01696847 - layer.0.v_cache 0.00000026 0.00014296 - layer.1.k_cache 0.00305459 0.81231951 - layer.1.v_cache 0.00000078 0.00052808 - layer.2.k_cache 0.00115271 0.36812866 - layer.2.v_cache 0.00000108 0.00072736 - layer.3.k_cache 0.00136835 0.41021855 - layer.3.v_cache 0.00000209 0.00117023 - layer.4.k_cache 0.00338224 0.88125575 - layer.4.v_cache 0.00000309 0.00200458 - layer.4.output 0.00017352 0.06579922 - ------------------------------------------------------------------------------------- - TOTAL 0.00263024 0.69690436 - (elements=2,508,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2508800 -Total Bytes 557232 -BPFP 1.7769 bits/point -EBPFP 3.5538 equivalent bits/point -MSE 0.696904 ----------------------- -------------------------------------------------------- -Time: 3.566s Load: 0.009s, Pack+Encode: 1.965s, Decode+Unpack: 1.591s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6969 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample50-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample51-layer4-item1.zst (57/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample51-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 184, 128) -Output shape: (1, 184, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) -> torch.Size([1, 1, 184, 1024]) - layer.4.output: torch.Size([1, 184, 4096]) -> torch.Size([1, 1, 184, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,700B, BPFP=0.4543 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,488B, BPFP=2.1437 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,840B, BPFP=1.5642 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,648B, BPFP=2.2779 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,036B, BPFP=1.9122 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,176B, BPFP=2.3003 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,180B, BPFP=1.9183 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,732B, BPFP=2.2814 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,916B, BPFP=1.6099 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,920B, BPFP=2.3319 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 162,744B, BPFP=1.7275 -⌛️ [2/4] FRONTEND: Frontend time: 1.945s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.output: torch.Size([1, 184, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 184, 128]) - layer.0.v_cache: torch.Size([1, 8, 184, 128]) - layer.1.k_cache: torch.Size([1, 8, 184, 128]) - layer.1.v_cache: torch.Size([1, 8, 184, 128]) - layer.2.k_cache: torch.Size([1, 8, 184, 128]) - layer.2.v_cache: torch.Size([1, 8, 184, 128]) - layer.3.k_cache: torch.Size([1, 8, 184, 128]) - layer.3.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.k_cache: torch.Size([1, 8, 184, 128]) - layer.4.v_cache: torch.Size([1, 8, 184, 128]) - layer.4.output: torch.Size([1, 184, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02750326 6.85000942 - layer.0.v_cache 0.00000026 0.00014850 - layer.1.k_cache 0.00297607 0.84744379 - layer.1.v_cache 0.00000082 0.00055790 - layer.2.k_cache 0.00115877 0.36631356 - layer.2.v_cache 0.00000113 0.00074102 - layer.3.k_cache 0.00134157 0.43307549 - layer.3.v_cache 0.00000216 0.00124221 - layer.4.k_cache 0.00343623 0.81398972 - layer.4.v_cache 0.00000338 0.00212486 - layer.4.output 0.00018269 0.06829404 - ------------------------------------------------------------------------------------- - TOTAL 0.00265389 0.68491590 - (elements=2,637,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2637824 -Total Bytes 605380 -BPFP 1.8360 bits/point -EBPFP 3.6720 equivalent bits/point -MSE 0.684916 ----------------------- -------------------------------------------------------- -Time: 3.506s Load: 0.010s, Pack+Encode: 1.945s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 184, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6849 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample51-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample51-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample52-layer4-item1.zst (58/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample52-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 169, 128) -Output shape: (1, 169, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.output: torch.Size([1, 169, 4096]) -> torch.Size([1, 1, 169, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,256B, BPFP=0.4279 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,564B, BPFP=2.2912 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,336B, BPFP=1.6335 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,008B, BPFP=2.4504 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,536B, BPFP=2.0588 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,464B, BPFP=2.4715 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,532B, BPFP=2.0586 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,028B, BPFP=2.4514 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,016B, BPFP=1.7112 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,036B, BPFP=2.4980 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 136,212B, BPFP=1.5742 -⌛️ [2/4] FRONTEND: Frontend time: 1.955s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.557s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02710367 7.12649085 - layer.0.v_cache 0.00000027 0.00015095 - layer.1.k_cache 0.00315269 0.78198206 - layer.1.v_cache 0.00000078 0.00054430 - layer.2.k_cache 0.00113874 0.36837782 - layer.2.v_cache 0.00000109 0.00075308 - layer.3.k_cache 0.00134854 0.41005761 - layer.3.v_cache 0.00000205 0.00119404 - layer.4.k_cache 0.00348762 0.85928868 - layer.4.v_cache 0.00000318 0.00215337 - layer.4.output 0.00019737 0.07637938 - ------------------------------------------------------------------------------------- - TOTAL 0.00264486 0.70403645 - (elements=2,422,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2422784 -Total Bytes 569988 -BPFP 1.8821 bits/point -EBPFP 3.7642 equivalent bits/point -MSE 0.704036 ----------------------- -------------------------------------------------------- -Time: 3.522s Load: 0.009s, Pack+Encode: 1.955s, Decode+Unpack: 1.557s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7040 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample52-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample53-layer4-item1.zst (59/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample53-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,244B, BPFP=0.4626 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,736B, BPFP=2.2912 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,960B, BPFP=1.6239 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,288B, BPFP=2.4064 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,400B, BPFP=2.0502 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,128B, BPFP=2.4444 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,328B, BPFP=2.0470 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,700B, BPFP=2.4250 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,328B, BPFP=1.6857 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,496B, BPFP=2.4610 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 165,576B, BPFP=1.8693 -⌛️ [2/4] FRONTEND: Frontend time: 1.983s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.545s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02649080 7.04780005 - layer.0.v_cache 0.00000026 0.00015276 - layer.1.k_cache 0.00311801 0.85256005 - layer.1.v_cache 0.00000082 0.00054326 - layer.2.k_cache 0.00118182 0.37799222 - layer.2.v_cache 0.00000117 0.00075723 - layer.3.k_cache 0.00133769 0.42004002 - layer.3.v_cache 0.00000224 0.00123343 - layer.4.k_cache 0.00352722 0.86869583 - layer.4.v_cache 0.00000310 0.00210552 - layer.4.output 0.00018175 0.06061548 - ------------------------------------------------------------------------------------- - TOTAL 0.00259930 0.70102445 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 606184 -BPFP 1.9553 bits/point -EBPFP 3.9107 equivalent bits/point -MSE 0.701024 ----------------------- -------------------------------------------------------- -Time: 3.537s Load: 0.009s, Pack+Encode: 1.983s, Decode+Unpack: 1.545s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7010 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample53-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample54-layer4-item1.zst (60/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample54-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 176, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 176, 128) -Output shape: (1, 176, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.0.v_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.1.k_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.1.v_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.2.k_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.2.v_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.3.k_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.3.v_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.4.k_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.4.v_cache: torch.Size([1, 8, 176, 128]) -> torch.Size([1, 1, 176, 1024]) - layer.4.output: torch.Size([1, 176, 4096]) -> torch.Size([1, 1, 176, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,772B, BPFP=0.4338 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,108B, BPFP=2.2243 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,636B, BPFP=1.5819 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,300B, BPFP=2.3659 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,380B, BPFP=1.9700 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,060B, BPFP=2.3997 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,340B, BPFP=1.9682 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,472B, BPFP=2.3736 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,248B, BPFP=1.6090 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,236B, BPFP=2.4075 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 148,200B, BPFP=1.6446 -⌛️ [2/4] FRONTEND: Frontend time: 1.983s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 176, 128]) - layer.0.v_cache: torch.Size([1, 8, 176, 128]) - layer.1.k_cache: torch.Size([1, 8, 176, 128]) - layer.1.v_cache: torch.Size([1, 8, 176, 128]) - layer.2.k_cache: torch.Size([1, 8, 176, 128]) - layer.2.v_cache: torch.Size([1, 8, 176, 128]) - layer.3.k_cache: torch.Size([1, 8, 176, 128]) - layer.3.v_cache: torch.Size([1, 8, 176, 128]) - layer.4.k_cache: torch.Size([1, 8, 176, 128]) - layer.4.v_cache: torch.Size([1, 8, 176, 128]) - layer.4.output: torch.Size([1, 176, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.531s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 176, 128]) - layer.0.v_cache: torch.Size([1, 8, 176, 128]) - layer.1.k_cache: torch.Size([1, 8, 176, 128]) - layer.1.v_cache: torch.Size([1, 8, 176, 128]) - layer.2.k_cache: torch.Size([1, 8, 176, 128]) - layer.2.v_cache: torch.Size([1, 8, 176, 128]) - layer.3.k_cache: torch.Size([1, 8, 176, 128]) - layer.3.v_cache: torch.Size([1, 8, 176, 128]) - layer.4.k_cache: torch.Size([1, 8, 176, 128]) - layer.4.v_cache: torch.Size([1, 8, 176, 128]) - layer.4.output: torch.Size([1, 176, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02681748 6.89934193 - layer.0.v_cache 0.00000027 0.00015329 - layer.1.k_cache 0.00299919 0.85369544 - layer.1.v_cache 0.00000086 0.00054598 - layer.2.k_cache 0.00118180 0.37352579 - layer.2.v_cache 0.00000111 0.00074274 - layer.3.k_cache 0.00131753 0.39999641 - layer.3.v_cache 0.00000211 0.00119988 - layer.4.k_cache 0.00351615 0.81056855 - layer.4.v_cache 0.00000302 0.00202573 - layer.4.output 0.00017220 0.06752997 - ------------------------------------------------------------------------------------- - TOTAL 0.00260917 0.68656540 - (elements=2,523,136) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2523136 -Total Bytes 583752 -BPFP 1.8509 bits/point -EBPFP 3.7018 equivalent bits/point -MSE 0.686565 ----------------------- -------------------------------------------------------- -Time: 3.524s Load: 0.010s, Pack+Encode: 1.983s, Decode+Unpack: 1.531s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 176, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 176, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6866 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample54-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample54-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample55-layer4-item1.zst (61/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample55-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 164, 128) -Output shape: (1, 164, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,856B, BPFP=0.4219 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,680B, BPFP=2.3666 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,752B, BPFP=1.6079 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,396B, BPFP=2.4960 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,724B, BPFP=2.0829 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,036B, BPFP=2.5265 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,104B, BPFP=2.1010 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,352B, BPFP=2.4939 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,816B, BPFP=1.7062 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,448B, BPFP=2.5461 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 107,328B, BPFP=1.2782 -⌛️ [2/4] FRONTEND: Frontend time: 1.962s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.549s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02653997 7.26916504 - layer.0.v_cache 0.00000026 0.00014999 - layer.1.k_cache 0.00301226 0.82512655 - layer.1.v_cache 0.00000076 0.00053549 - layer.2.k_cache 0.00116255 0.37175411 - layer.2.v_cache 0.00000105 0.00071301 - layer.3.k_cache 0.00132644 0.40805328 - layer.3.v_cache 0.00000197 0.00113637 - layer.4.k_cache 0.00346445 0.81612266 - layer.4.v_cache 0.00000298 0.00201343 - layer.4.output 0.00016888 0.06807416 - ------------------------------------------------------------------------------------- - TOTAL 0.00258487 0.71193333 - (elements=2,351,104) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2351104 -Total Bytes 534492 -BPFP 1.8187 bits/point -EBPFP 3.6374 equivalent bits/point -MSE 0.711933 ----------------------- -------------------------------------------------------- -Time: 3.521s Load: 0.009s, Pack+Encode: 1.962s, Decode+Unpack: 1.549s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7119 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample55-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample55-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample56-layer4-item1.zst (62/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample56-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 179, 128) -Output shape: (1, 179, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) -> torch.Size([1, 1, 179, 1024]) - layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,540B, BPFP=0.4164 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,304B, BPFP=2.1519 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,748B, BPFP=1.6039 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,840B, BPFP=2.3062 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,868B, BPFP=1.9583 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,500B, BPFP=2.3350 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,724B, BPFP=1.9520 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,984B, BPFP=2.3125 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,704B, BPFP=1.6456 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,132B, BPFP=2.3626 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 117,008B, BPFP=1.2767 -⌛️ [2/4] FRONTEND: Frontend time: 1.977s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.547s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 179, 128]) - layer.0.v_cache: torch.Size([1, 8, 179, 128]) - layer.1.k_cache: torch.Size([1, 8, 179, 128]) - layer.1.v_cache: torch.Size([1, 8, 179, 128]) - layer.2.k_cache: torch.Size([1, 8, 179, 128]) - layer.2.v_cache: torch.Size([1, 8, 179, 128]) - layer.3.k_cache: torch.Size([1, 8, 179, 128]) - layer.3.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.k_cache: torch.Size([1, 8, 179, 128]) - layer.4.v_cache: torch.Size([1, 8, 179, 128]) - layer.4.output: torch.Size([1, 179, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02669685 7.02285374 - layer.0.v_cache 0.00000027 0.00014790 - layer.1.k_cache 0.00301279 0.70270683 - layer.1.v_cache 0.00000079 0.00052611 - layer.2.k_cache 0.00116913 0.36546931 - layer.2.v_cache 0.00000107 0.00071719 - layer.3.k_cache 0.00133970 0.40989502 - layer.3.v_cache 0.00000203 0.00114571 - layer.4.k_cache 0.00352847 0.82914546 - layer.4.v_cache 0.00000295 0.00197948 - layer.4.output 0.00017485 0.06291551 - ------------------------------------------------------------------------------------- - TOTAL 0.00260382 0.68473206 - (elements=2,566,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2566144 -Total Bytes 553352 -BPFP 1.7251 bits/point -EBPFP 3.4502 equivalent bits/point -MSE 0.684732 ----------------------- -------------------------------------------------------- -Time: 3.534s Load: 0.010s, Pack+Encode: 1.977s, Decode+Unpack: 1.547s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 179, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6847 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample56-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample56-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample57-layer4-item1.zst (63/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample57-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 188, 128) -Output shape: (1, 188, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.0.v_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.1.k_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.1.v_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.2.k_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.2.v_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.3.k_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.3.v_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.4.k_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.4.v_cache: torch.Size([1, 8, 188, 128]) -> torch.Size([1, 1, 188, 1024]) - layer.4.output: torch.Size([1, 188, 4096]) -> torch.Size([1, 1, 188, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,936B, BPFP=0.4129 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,676B, BPFP=2.0643 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,488B, BPFP=1.5163 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,848B, BPFP=2.1961 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,364B, BPFP=1.8436 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,304B, BPFP=2.2151 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,332B, BPFP=1.8423 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,896B, BPFP=2.1981 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,296B, BPFP=1.5499 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,760B, BPFP=2.2340 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 112,028B, BPFP=1.1639 -⌛️ [2/4] FRONTEND: Frontend time: 1.956s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 188, 128]) - layer.0.v_cache: torch.Size([1, 8, 188, 128]) - layer.1.k_cache: torch.Size([1, 8, 188, 128]) - layer.1.v_cache: torch.Size([1, 8, 188, 128]) - layer.2.k_cache: torch.Size([1, 8, 188, 128]) - layer.2.v_cache: torch.Size([1, 8, 188, 128]) - layer.3.k_cache: torch.Size([1, 8, 188, 128]) - layer.3.v_cache: torch.Size([1, 8, 188, 128]) - layer.4.k_cache: torch.Size([1, 8, 188, 128]) - layer.4.v_cache: torch.Size([1, 8, 188, 128]) - layer.4.output: torch.Size([1, 188, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.540s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 188, 128]) - layer.0.v_cache: torch.Size([1, 8, 188, 128]) - layer.1.k_cache: torch.Size([1, 8, 188, 128]) - layer.1.v_cache: torch.Size([1, 8, 188, 128]) - layer.2.k_cache: torch.Size([1, 8, 188, 128]) - layer.2.v_cache: torch.Size([1, 8, 188, 128]) - layer.3.k_cache: torch.Size([1, 8, 188, 128]) - layer.3.v_cache: torch.Size([1, 8, 188, 128]) - layer.4.k_cache: torch.Size([1, 8, 188, 128]) - layer.4.v_cache: torch.Size([1, 8, 188, 128]) - layer.4.output: torch.Size([1, 188, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02614099 6.95994925 - layer.0.v_cache 0.00000027 0.00014409 - layer.1.k_cache 0.00304419 0.79245920 - layer.1.v_cache 0.00000081 0.00051292 - layer.2.k_cache 0.00119225 0.40837779 - layer.2.v_cache 0.00000108 0.00068425 - layer.3.k_cache 0.00133471 0.44160819 - layer.3.v_cache 0.00000207 0.00108800 - layer.4.k_cache 0.00351311 0.81193737 - layer.4.v_cache 0.00000296 0.00185509 - layer.4.output 0.00017320 0.06829873 - ------------------------------------------------------------------------------------- - TOTAL 0.00256609 0.69227222 - (elements=2,695,168) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2695168 -Total Bytes 546928 -BPFP 1.6234 bits/point -EBPFP 3.2469 equivalent bits/point -MSE 0.692272 ----------------------- -------------------------------------------------------- -Time: 3.506s Load: 0.010s, Pack+Encode: 1.956s, Decode+Unpack: 1.540s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 188, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6923 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample57-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample57-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample59-layer4-item1.zst (64/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample59-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 197, 128) -Output shape: (1, 197, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) -> torch.Size([1, 1, 197, 1024]) - layer.4.output: torch.Size([1, 197, 4096]) -> torch.Size([1, 1, 197, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,192B, BPFP=0.4042 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 61,584B, BPFP=2.4423 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 39,472B, BPFP=1.5654 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 65,812B, BPFP=2.6099 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,736B, BPFP=1.9724 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 67,528B, BPFP=2.6780 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 54,116B, BPFP=2.1461 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 66,384B, BPFP=2.6326 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 40,192B, BPFP=1.5939 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 67,964B, BPFP=2.6953 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 127,384B, BPFP=1.2629 -⌛️ [2/4] FRONTEND: Frontend time: 2.361s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.output: torch.Size([1, 197, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.979s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 197, 128]) - layer.0.v_cache: torch.Size([1, 8, 197, 128]) - layer.1.k_cache: torch.Size([1, 8, 197, 128]) - layer.1.v_cache: torch.Size([1, 8, 197, 128]) - layer.2.k_cache: torch.Size([1, 8, 197, 128]) - layer.2.v_cache: torch.Size([1, 8, 197, 128]) - layer.3.k_cache: torch.Size([1, 8, 197, 128]) - layer.3.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.k_cache: torch.Size([1, 8, 197, 128]) - layer.4.v_cache: torch.Size([1, 8, 197, 128]) - layer.4.output: torch.Size([1, 197, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02717010 6.98478211 - layer.0.v_cache 0.00000028 0.00015001 - layer.1.k_cache 0.00303619 0.79245382 - layer.1.v_cache 0.00000078 0.00049113 - layer.2.k_cache 0.00117018 0.37868566 - layer.2.v_cache 0.00000106 0.00067899 - layer.3.k_cache 0.00130579 0.39586109 - layer.3.v_cache 0.00000202 0.00107565 - layer.4.k_cache 0.00343555 0.78019095 - layer.4.v_cache 0.00000289 0.00184117 - layer.4.output 0.00018088 0.06304178 - ------------------------------------------------------------------------------------- - TOTAL 0.00263203 0.68488412 - (elements=2,824,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2824192 -Total Bytes 650364 -BPFP 1.8423 bits/point -EBPFP 3.6845 equivalent bits/point -MSE 0.684884 ----------------------- -------------------------------------------------------- -Time: 4.350s Load: 0.011s, Pack+Encode: 2.361s, Decode+Unpack: 1.979s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 197, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6849 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample59-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample59-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample6-layer4-item1.zst (65/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample6-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 204, 128) -Output shape: (1, 204, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) -> torch.Size([1, 1, 204, 1024]) - layer.4.output: torch.Size([1, 204, 4096]) -> torch.Size([1, 1, 204, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,624B, BPFP=0.4452 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,488B, BPFP=2.4697 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,224B, BPFP=1.6170 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 66,856B, BPFP=2.5604 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,088B, BPFP=2.0714 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 67,600B, BPFP=2.5888 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,316B, BPFP=2.1567 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 68,560B, BPFP=2.6256 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 41,304B, BPFP=1.5818 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 68,836B, BPFP=2.6362 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 152,716B, BPFP=1.4621 -⌛️ [2/4] FRONTEND: Frontend time: 2.377s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.output: torch.Size([1, 204, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.036s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 204, 128]) - layer.0.v_cache: torch.Size([1, 8, 204, 128]) - layer.1.k_cache: torch.Size([1, 8, 204, 128]) - layer.1.v_cache: torch.Size([1, 8, 204, 128]) - layer.2.k_cache: torch.Size([1, 8, 204, 128]) - layer.2.v_cache: torch.Size([1, 8, 204, 128]) - layer.3.k_cache: torch.Size([1, 8, 204, 128]) - layer.3.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.k_cache: torch.Size([1, 8, 204, 128]) - layer.4.v_cache: torch.Size([1, 8, 204, 128]) - layer.4.output: torch.Size([1, 204, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02651415 7.10240084 - layer.0.v_cache 0.00000026 0.00014908 - layer.1.k_cache 0.00301626 0.85225685 - layer.1.v_cache 0.00000082 0.00053788 - layer.2.k_cache 0.00116162 0.36934924 - layer.2.v_cache 0.00000113 0.00073417 - layer.3.k_cache 0.00136563 0.42152438 - layer.3.v_cache 0.00000207 0.00117696 - layer.4.k_cache 0.00348750 0.75656786 - layer.4.v_cache 0.00000309 0.00205639 - layer.4.output 0.00017661 0.06636844 - ------------------------------------------------------------------------------------- - TOTAL 0.00258993 0.69801624 - (elements=2,924,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2924544 -Total Bytes 694612 -BPFP 1.9001 bits/point -EBPFP 3.8002 equivalent bits/point -MSE 0.698016 ----------------------- -------------------------------------------------------- -Time: 4.425s Load: 0.012s, Pack+Encode: 2.377s, Decode+Unpack: 2.036s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 204, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6980 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample6-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample6-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample60-layer4-item1.zst (66/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample60-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,592B, BPFP=0.4382 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,776B, BPFP=2.2741 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,752B, BPFP=1.5877 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,648B, BPFP=2.4053 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,724B, BPFP=2.0433 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,620B, BPFP=2.4497 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,500B, BPFP=2.0331 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,076B, BPFP=2.4249 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,804B, BPFP=1.6815 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,124B, BPFP=2.4728 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 142,128B, BPFP=1.6234 -⌛️ [2/4] FRONTEND: Frontend time: 1.978s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.550s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02690465 6.98513883 - layer.0.v_cache 0.00000028 0.00015347 - layer.1.k_cache 0.00309750 0.87196662 - layer.1.v_cache 0.00000081 0.00052174 - layer.2.k_cache 0.00115835 0.37504658 - layer.2.v_cache 0.00000109 0.00072133 - layer.3.k_cache 0.00135207 0.41298261 - layer.3.v_cache 0.00000214 0.00117584 - layer.4.k_cache 0.00342229 0.84486693 - layer.4.v_cache 0.00000302 0.00197627 - layer.4.output 0.00017203 0.07220119 - ------------------------------------------------------------------------------------- - TOTAL 0.00261645 0.69881107 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 575744 -BPFP 1.8789 bits/point -EBPFP 3.7577 equivalent bits/point -MSE 0.698811 ----------------------- -------------------------------------------------------- -Time: 3.539s Load: 0.011s, Pack+Encode: 1.978s, Decode+Unpack: 1.550s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6988 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample60-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample62-layer4-item1.zst (67/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample62-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 180, 128) -Output shape: (1, 180, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.0.v_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.1.k_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.1.v_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.2.k_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.2.v_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.3.k_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.3.v_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.4.k_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.4.v_cache: torch.Size([1, 8, 180, 128]) -> torch.Size([1, 1, 180, 1024]) - layer.4.output: torch.Size([1, 180, 4096]) -> torch.Size([1, 1, 180, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 10,148B, BPFP=0.4405 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,228B, BPFP=2.1800 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,392B, BPFP=1.5795 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,204B, BPFP=2.3092 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,708B, BPFP=1.9405 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,908B, BPFP=2.3398 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,640B, BPFP=1.9375 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,300B, BPFP=2.3134 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,980B, BPFP=1.6484 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,400B, BPFP=2.3611 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 139,176B, BPFP=1.5102 -⌛️ [2/4] FRONTEND: Frontend time: 1.956s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 180, 128]) - layer.0.v_cache: torch.Size([1, 8, 180, 128]) - layer.1.k_cache: torch.Size([1, 8, 180, 128]) - layer.1.v_cache: torch.Size([1, 8, 180, 128]) - layer.2.k_cache: torch.Size([1, 8, 180, 128]) - layer.2.v_cache: torch.Size([1, 8, 180, 128]) - layer.3.k_cache: torch.Size([1, 8, 180, 128]) - layer.3.v_cache: torch.Size([1, 8, 180, 128]) - layer.4.k_cache: torch.Size([1, 8, 180, 128]) - layer.4.v_cache: torch.Size([1, 8, 180, 128]) - layer.4.output: torch.Size([1, 180, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.542s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 180, 128]) - layer.0.v_cache: torch.Size([1, 8, 180, 128]) - layer.1.k_cache: torch.Size([1, 8, 180, 128]) - layer.1.v_cache: torch.Size([1, 8, 180, 128]) - layer.2.k_cache: torch.Size([1, 8, 180, 128]) - layer.2.v_cache: torch.Size([1, 8, 180, 128]) - layer.3.k_cache: torch.Size([1, 8, 180, 128]) - layer.3.v_cache: torch.Size([1, 8, 180, 128]) - layer.4.k_cache: torch.Size([1, 8, 180, 128]) - layer.4.v_cache: torch.Size([1, 8, 180, 128]) - layer.4.output: torch.Size([1, 180, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02696468 6.80405138 - layer.0.v_cache 0.00000027 0.00014995 - layer.1.k_cache 0.00304294 0.72250697 - layer.1.v_cache 0.00000080 0.00053930 - layer.2.k_cache 0.00116858 0.36253984 - layer.2.v_cache 0.00000111 0.00072505 - layer.3.k_cache 0.00132793 0.40464630 - layer.3.v_cache 0.00000207 0.00118581 - layer.4.k_cache 0.00340096 0.78755442 - layer.4.v_cache 0.00000322 0.00211979 - layer.4.output 0.00018870 0.07144206 - ------------------------------------------------------------------------------------- - TOTAL 0.00261910 0.66941336 - (elements=2,580,480) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2580480 -Total Bytes 578084 -BPFP 1.7922 bits/point -EBPFP 3.5844 equivalent bits/point -MSE 0.669413 ----------------------- -------------------------------------------------------- -Time: 3.507s Load: 0.009s, Pack+Encode: 1.956s, Decode+Unpack: 1.542s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 180, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6694 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample62-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample62-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample63-layer4-item1.zst (68/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample63-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 165, 128) -Output shape: (1, 165, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.output: torch.Size([1, 165, 4096]) -> torch.Size([1, 1, 165, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,132B, BPFP=0.4324 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,700B, BPFP=2.3532 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,940B, BPFP=1.6544 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,636B, BPFP=2.4922 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,812B, BPFP=2.0744 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,384B, BPFP=2.5277 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,896B, BPFP=2.0784 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,664B, BPFP=2.4936 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,992B, BPFP=1.7042 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,704B, BPFP=2.5428 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 156,068B, BPFP=1.8474 -⌛️ [2/4] FRONTEND: Frontend time: 1.997s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.579s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02707518 7.30971976 - layer.0.v_cache 0.00000026 0.00015564 - layer.1.k_cache 0.00315113 0.89499364 - layer.1.v_cache 0.00000089 0.00057313 - layer.2.k_cache 0.00116140 0.37975075 - layer.2.v_cache 0.00000111 0.00077306 - layer.3.k_cache 0.00133385 0.41484819 - layer.3.v_cache 0.00000208 0.00118696 - layer.4.k_cache 0.00353406 0.86559753 - layer.4.v_cache 0.00000307 0.00207489 - layer.4.output 0.00018338 0.06485363 - ------------------------------------------------------------------------------------- - TOTAL 0.00264261 0.72350629 - (elements=2,365,440) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2365440 -Total Bytes 585928 -BPFP 1.9816 bits/point -EBPFP 3.9633 equivalent bits/point -MSE 0.723506 ----------------------- -------------------------------------------------------- -Time: 3.586s Load: 0.010s, Pack+Encode: 1.997s, Decode+Unpack: 1.579s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7235 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample63-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample63-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample64-layer4-item1.zst (69/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample64-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,408B, BPFP=0.4298 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,744B, BPFP=2.2727 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,568B, BPFP=1.6250 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,824B, BPFP=2.4134 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,548B, BPFP=2.0353 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,556B, BPFP=2.4468 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,640B, BPFP=2.0395 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,980B, BPFP=2.4205 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,108B, BPFP=1.6497 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,112B, BPFP=2.4722 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 117,440B, BPFP=1.3414 -⌛️ [2/4] FRONTEND: Frontend time: 1.967s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.565s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02638843 7.02424488 - layer.0.v_cache 0.00000026 0.00014857 - layer.1.k_cache 0.00303726 0.83582364 - layer.1.v_cache 0.00000078 0.00052428 - layer.2.k_cache 0.00118268 0.37786526 - layer.2.v_cache 0.00000107 0.00072410 - layer.3.k_cache 0.00135949 0.42257806 - layer.3.v_cache 0.00000203 0.00116322 - layer.4.k_cache 0.00348397 0.82810242 - layer.4.v_cache 0.00000306 0.00197317 - layer.4.output 0.00019589 0.06987609 - ------------------------------------------------------------------------------------- - TOTAL 0.00258876 0.69804657 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 550928 -BPFP 1.7979 bits/point -EBPFP 3.5958 equivalent bits/point -MSE 0.698047 ----------------------- -------------------------------------------------------- -Time: 3.542s Load: 0.009s, Pack+Encode: 1.967s, Decode+Unpack: 1.565s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6980 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample64-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample64-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample65-layer4-item1.zst (70/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample65-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 166, 128) -Output shape: (1, 166, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,228B, BPFP=0.4343 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,512B, BPFP=2.3302 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,896B, BPFP=1.6423 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,544B, BPFP=2.4729 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,744B, BPFP=2.1058 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,384B, BPFP=2.5124 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,560B, BPFP=2.0971 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,532B, BPFP=2.4723 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,192B, BPFP=1.6562 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,616B, BPFP=2.5233 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 120,948B, BPFP=1.4231 -⌛️ [2/4] FRONTEND: Frontend time: 1.952s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.535s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02749621 7.00128468 - layer.0.v_cache 0.00000027 0.00014880 - layer.1.k_cache 0.00299443 0.75554404 - layer.1.v_cache 0.00000078 0.00053695 - layer.2.k_cache 0.00116568 0.37257348 - layer.2.v_cache 0.00000116 0.00075704 - layer.3.k_cache 0.00133954 0.42306027 - layer.3.v_cache 0.00000205 0.00118136 - layer.4.k_cache 0.00345084 0.81596981 - layer.4.v_cache 0.00000307 0.00206510 - layer.4.output 0.00017739 0.06361336 - ------------------------------------------------------------------------------------- - TOTAL 0.00265454 0.68768393 - (elements=2,379,776) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2379776 -Total Bytes 551156 -BPFP 1.8528 bits/point -EBPFP 3.7056 equivalent bits/point -MSE 0.687684 ----------------------- -------------------------------------------------------- -Time: 3.496s Load: 0.010s, Pack+Encode: 1.952s, Decode+Unpack: 1.535s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6877 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample65-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample66-layer4-item1.zst (71/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample66-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 164, 128) -Output shape: (1, 164, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,188B, BPFP=0.4377 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,604B, BPFP=2.3630 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,820B, BPFP=1.6587 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,352B, BPFP=2.4939 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,244B, BPFP=2.1077 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,048B, BPFP=2.5271 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,180B, BPFP=2.1046 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,628B, BPFP=2.5071 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,404B, BPFP=1.6865 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,556B, BPFP=2.5513 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 133,792B, BPFP=1.5934 -⌛️ [2/4] FRONTEND: Frontend time: 1.957s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.557s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02831957 7.14473520 - layer.0.v_cache 0.00000026 0.00015046 - layer.1.k_cache 0.00319086 0.81871786 - layer.1.v_cache 0.00000080 0.00054371 - layer.2.k_cache 0.00118664 0.35721653 - layer.2.v_cache 0.00000113 0.00073492 - layer.3.k_cache 0.00134609 0.41231942 - layer.3.v_cache 0.00000208 0.00121378 - layer.4.k_cache 0.00339094 0.79505399 - layer.4.v_cache 0.00000313 0.00213182 - layer.4.output 0.00016991 0.08128498 - ------------------------------------------------------------------------------------- - TOTAL 0.00272294 0.70413983 - (elements=2,351,104) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2351104 -Total Bytes 562816 -BPFP 1.9151 bits/point -EBPFP 3.8301 equivalent bits/point -MSE 0.704140 ----------------------- -------------------------------------------------------- -Time: 3.523s Load: 0.009s, Pack+Encode: 1.957s, Decode+Unpack: 1.557s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7041 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample66-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample67-layer4-item1.zst (72/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample67-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 169, 128) -Output shape: (1, 169, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.output: torch.Size([1, 169, 4096]) -> torch.Size([1, 1, 169, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,104B, BPFP=0.4209 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,788B, BPFP=2.3016 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,608B, BPFP=1.5999 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,972B, BPFP=2.4488 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,388B, BPFP=2.0520 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,448B, BPFP=2.4708 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,340B, BPFP=2.0497 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,928B, BPFP=2.4467 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,460B, BPFP=1.6392 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,988B, BPFP=2.4957 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 104,972B, BPFP=1.2132 -⌛️ [2/4] FRONTEND: Frontend time: 1.963s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.551s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02632406 7.20944160 - layer.0.v_cache 0.00000026 0.00015134 - layer.1.k_cache 0.00304039 0.86565440 - layer.1.v_cache 0.00000079 0.00054076 - layer.2.k_cache 0.00117699 0.37155788 - layer.2.v_cache 0.00000106 0.00071517 - layer.3.k_cache 0.00132669 0.41178425 - layer.3.v_cache 0.00000197 0.00110720 - layer.4.k_cache 0.00349736 0.81237667 - layer.4.v_cache 0.00000306 0.00201537 - layer.4.output 0.00016247 0.07145453 - ------------------------------------------------------------------------------------- - TOTAL 0.00257304 0.71151162 - (elements=2,422,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2422784 -Total Bytes 535996 -BPFP 1.7699 bits/point -EBPFP 3.5397 equivalent bits/point -MSE 0.711512 ----------------------- -------------------------------------------------------- -Time: 3.525s Load: 0.011s, Pack+Encode: 1.963s, Decode+Unpack: 1.551s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7115 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample67-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample67-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample68-layer4-item1.zst (73/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample68-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 167, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 167, 128) -Output shape: (1, 167, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) -> torch.Size([1, 1, 167, 1024]) - layer.4.output: torch.Size([1, 167, 4096]) -> torch.Size([1, 1, 167, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,372B, BPFP=0.4384 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,020B, BPFP=2.3400 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,976B, BPFP=1.6362 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,828B, BPFP=2.4714 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,580B, BPFP=2.0387 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,640B, BPFP=2.5094 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,424B, BPFP=2.0782 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,944B, BPFP=2.4768 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,228B, BPFP=1.6948 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,188B, BPFP=2.5350 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 143,280B, BPFP=1.6757 -⌛️ [2/4] FRONTEND: Frontend time: 1.966s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.output: torch.Size([1, 167, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.517s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 167, 128]) - layer.0.v_cache: torch.Size([1, 8, 167, 128]) - layer.1.k_cache: torch.Size([1, 8, 167, 128]) - layer.1.v_cache: torch.Size([1, 8, 167, 128]) - layer.2.k_cache: torch.Size([1, 8, 167, 128]) - layer.2.v_cache: torch.Size([1, 8, 167, 128]) - layer.3.k_cache: torch.Size([1, 8, 167, 128]) - layer.3.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.k_cache: torch.Size([1, 8, 167, 128]) - layer.4.v_cache: torch.Size([1, 8, 167, 128]) - layer.4.output: torch.Size([1, 167, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02744324 7.31091089 - layer.0.v_cache 0.00000027 0.00014978 - layer.1.k_cache 0.00306544 0.77072820 - layer.1.v_cache 0.00000083 0.00055220 - layer.2.k_cache 0.00116567 0.37429120 - layer.2.v_cache 0.00000114 0.00075160 - layer.3.k_cache 0.00131614 0.39562070 - layer.3.v_cache 0.00000212 0.00120496 - layer.4.k_cache 0.00345897 0.81689608 - layer.4.v_cache 0.00000319 0.00206425 - layer.4.output 0.00018850 0.07082037 - ------------------------------------------------------------------------------------- - TOTAL 0.00265793 0.71117510 - (elements=2,394,112) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2394112 -Total Bytes 575480 -BPFP 1.9230 bits/point -EBPFP 3.8460 equivalent bits/point -MSE 0.711175 ----------------------- -------------------------------------------------------- -Time: 3.493s Load: 0.010s, Pack+Encode: 1.966s, Decode+Unpack: 1.517s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 167, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 167, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7112 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample68-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample69-layer4-item1.zst (74/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample69-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 189, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 189, 128) -Output shape: (1, 189, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.0.v_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.1.k_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.1.v_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.2.k_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.2.v_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.3.k_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.3.v_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.4.k_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.4.v_cache: torch.Size([1, 8, 189, 128]) -> torch.Size([1, 1, 189, 1024]) - layer.4.output: torch.Size([1, 189, 4096]) -> torch.Size([1, 1, 189, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,120B, BPFP=0.4597 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,540B, BPFP=2.0891 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 37,380B, BPFP=1.5451 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,688B, BPFP=2.2192 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,024B, BPFP=1.8611 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,208B, BPFP=2.2407 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,344B, BPFP=1.8330 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,864B, BPFP=2.2265 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,880B, BPFP=1.5658 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,700B, BPFP=2.2611 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 173,360B, BPFP=1.7915 -⌛️ [2/4] FRONTEND: Frontend time: 1.981s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 189, 128]) - layer.0.v_cache: torch.Size([1, 8, 189, 128]) - layer.1.k_cache: torch.Size([1, 8, 189, 128]) - layer.1.v_cache: torch.Size([1, 8, 189, 128]) - layer.2.k_cache: torch.Size([1, 8, 189, 128]) - layer.2.v_cache: torch.Size([1, 8, 189, 128]) - layer.3.k_cache: torch.Size([1, 8, 189, 128]) - layer.3.v_cache: torch.Size([1, 8, 189, 128]) - layer.4.k_cache: torch.Size([1, 8, 189, 128]) - layer.4.v_cache: torch.Size([1, 8, 189, 128]) - layer.4.output: torch.Size([1, 189, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.527s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 189, 128]) - layer.0.v_cache: torch.Size([1, 8, 189, 128]) - layer.1.k_cache: torch.Size([1, 8, 189, 128]) - layer.1.v_cache: torch.Size([1, 8, 189, 128]) - layer.2.k_cache: torch.Size([1, 8, 189, 128]) - layer.2.v_cache: torch.Size([1, 8, 189, 128]) - layer.3.k_cache: torch.Size([1, 8, 189, 128]) - layer.3.v_cache: torch.Size([1, 8, 189, 128]) - layer.4.k_cache: torch.Size([1, 8, 189, 128]) - layer.4.v_cache: torch.Size([1, 8, 189, 128]) - layer.4.output: torch.Size([1, 189, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02668087 7.30649014 - layer.0.v_cache 0.00000027 0.00014718 - layer.1.k_cache 0.00297688 0.78254627 - layer.1.v_cache 0.00000086 0.00052334 - layer.2.k_cache 0.00116960 0.38397592 - layer.2.v_cache 0.00000119 0.00070275 - layer.3.k_cache 0.00130480 0.43303567 - layer.3.v_cache 0.00000220 0.00117045 - layer.4.k_cache 0.00357267 0.81947585 - layer.4.v_cache 0.00000319 0.00198820 - layer.4.output 0.00018018 0.06183701 - ------------------------------------------------------------------------------------- - TOTAL 0.00260238 0.71267170 - (elements=2,709,504) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2709504 -Total Bytes 616108 -BPFP 1.8191 bits/point -EBPFP 3.6382 equivalent bits/point -MSE 0.712672 ----------------------- -------------------------------------------------------- -Time: 3.518s Load: 0.010s, Pack+Encode: 1.981s, Decode+Unpack: 1.527s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 189, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 189, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7127 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample69-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample69-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample70-layer4-item1.zst (75/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample70-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,968B, BPFP=0.4501 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,380B, BPFP=2.2751 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,604B, BPFP=1.6078 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,140B, BPFP=2.3997 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,648B, BPFP=2.0163 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,796B, BPFP=2.4294 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,856B, BPFP=2.0257 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,272B, BPFP=2.4057 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,600B, BPFP=1.6980 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,096B, BPFP=2.4429 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 153,728B, BPFP=1.7355 -⌛️ [2/4] FRONTEND: Frontend time: 1.992s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02663673 6.96230991 - layer.0.v_cache 0.00000027 0.00015379 - layer.1.k_cache 0.00311122 0.89092286 - layer.1.v_cache 0.00000080 0.00053056 - layer.2.k_cache 0.00117338 0.35899289 - layer.2.v_cache 0.00000114 0.00072823 - layer.3.k_cache 0.00131734 0.41349687 - layer.3.v_cache 0.00000215 0.00118671 - layer.4.k_cache 0.00350522 0.79897736 - layer.4.v_cache 0.00000298 0.00198526 - layer.4.output 0.00018231 0.06214713 - ------------------------------------------------------------------------------------- - TOTAL 0.00260575 0.69127664 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 591088 -BPFP 1.9066 bits/point -EBPFP 3.8133 equivalent bits/point -MSE 0.691277 ----------------------- -------------------------------------------------------- -Time: 3.554s Load: 0.010s, Pack+Encode: 1.992s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6913 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample70-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample70-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample71-layer4-item1.zst (76/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample71-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 166, 128) -Output shape: (1, 166, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,340B, BPFP=0.4396 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,644B, BPFP=2.3364 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,044B, BPFP=1.6022 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,436B, BPFP=2.4678 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,644B, BPFP=2.1011 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,348B, BPFP=2.5107 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,532B, BPFP=2.0958 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,744B, BPFP=2.4823 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,056B, BPFP=1.6498 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,700B, BPFP=2.5273 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 132,040B, BPFP=1.5536 -⌛️ [2/4] FRONTEND: Frontend time: 1.981s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.562s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02747041 7.11156712 - layer.0.v_cache 0.00000026 0.00015030 - layer.1.k_cache 0.00305587 0.79364492 - layer.1.v_cache 0.00000080 0.00053809 - layer.2.k_cache 0.00117233 0.35628473 - layer.2.v_cache 0.00000108 0.00073901 - layer.3.k_cache 0.00134723 0.41663613 - layer.3.v_cache 0.00000213 0.00121535 - layer.4.k_cache 0.00344432 0.84906171 - layer.4.v_cache 0.00000302 0.00204304 - layer.4.output 0.00018389 0.06893542 - ------------------------------------------------------------------------------------- - TOTAL 0.00265950 0.70054443 - (elements=2,379,776) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2379776 -Total Bytes 561528 -BPFP 1.8877 bits/point -EBPFP 3.7753 equivalent bits/point -MSE 0.700544 ----------------------- -------------------------------------------------------- -Time: 3.554s Load: 0.010s, Pack+Encode: 1.981s, Decode+Unpack: 1.562s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7005 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample71-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample71-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample72-layer4-item1.zst (77/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample72-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,908B, BPFP=0.4527 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,444B, BPFP=2.3046 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,304B, BPFP=1.6129 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,300B, BPFP=2.4351 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,964B, BPFP=2.0543 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,988B, BPFP=2.4666 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,652B, BPFP=2.0400 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,036B, BPFP=2.4231 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,812B, BPFP=1.6818 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,176B, BPFP=2.4751 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 137,960B, BPFP=1.5757 -⌛️ [2/4] FRONTEND: Frontend time: 1.972s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.574s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02713734 6.87004009 - layer.0.v_cache 0.00000027 0.00015159 - layer.1.k_cache 0.00309222 0.85276036 - layer.1.v_cache 0.00000080 0.00053611 - layer.2.k_cache 0.00115323 0.37979220 - layer.2.v_cache 0.00000110 0.00074210 - layer.3.k_cache 0.00132483 0.41008437 - layer.3.v_cache 0.00000201 0.00114352 - layer.4.k_cache 0.00341669 0.82881343 - layer.4.v_cache 0.00000301 0.00201468 - layer.4.output 0.00019064 0.07176932 - ------------------------------------------------------------------------------------- - TOTAL 0.00263529 0.68808255 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 574544 -BPFP 1.8749 bits/point -EBPFP 3.7499 equivalent bits/point -MSE 0.688083 ----------------------- -------------------------------------------------------- -Time: 3.556s Load: 0.010s, Pack+Encode: 1.972s, Decode+Unpack: 1.574s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6881 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample72-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample72-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample73-layer4-item1.zst (78/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample73-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 168, 128) -Output shape: (1, 168, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.0.v_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.1.k_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.1.v_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.2.k_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.2.v_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.3.k_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.3.v_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.4.k_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.4.v_cache: torch.Size([1, 8, 168, 128]) -> torch.Size([1, 1, 168, 1024]) - layer.4.output: torch.Size([1, 168, 4096]) -> torch.Size([1, 1, 168, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,492B, BPFP=0.4414 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,852B, BPFP=2.3183 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,692B, BPFP=1.6133 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,912B, BPFP=2.4606 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,048B, BPFP=2.0484 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,528B, BPFP=2.4892 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,892B, BPFP=2.0876 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,132B, BPFP=2.4708 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,376B, BPFP=1.6916 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,200B, BPFP=2.5205 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 120,892B, BPFP=1.4055 -⌛️ [2/4] FRONTEND: Frontend time: 1.962s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 168, 128]) - layer.0.v_cache: torch.Size([1, 8, 168, 128]) - layer.1.k_cache: torch.Size([1, 8, 168, 128]) - layer.1.v_cache: torch.Size([1, 8, 168, 128]) - layer.2.k_cache: torch.Size([1, 8, 168, 128]) - layer.2.v_cache: torch.Size([1, 8, 168, 128]) - layer.3.k_cache: torch.Size([1, 8, 168, 128]) - layer.3.v_cache: torch.Size([1, 8, 168, 128]) - layer.4.k_cache: torch.Size([1, 8, 168, 128]) - layer.4.v_cache: torch.Size([1, 8, 168, 128]) - layer.4.output: torch.Size([1, 168, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.553s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 168, 128]) - layer.0.v_cache: torch.Size([1, 8, 168, 128]) - layer.1.k_cache: torch.Size([1, 8, 168, 128]) - layer.1.v_cache: torch.Size([1, 8, 168, 128]) - layer.2.k_cache: torch.Size([1, 8, 168, 128]) - layer.2.v_cache: torch.Size([1, 8, 168, 128]) - layer.3.k_cache: torch.Size([1, 8, 168, 128]) - layer.3.v_cache: torch.Size([1, 8, 168, 128]) - layer.4.k_cache: torch.Size([1, 8, 168, 128]) - layer.4.v_cache: torch.Size([1, 8, 168, 128]) - layer.4.output: torch.Size([1, 168, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02628393 7.23645020 - layer.0.v_cache 0.00000027 0.00015151 - layer.1.k_cache 0.00308696 0.86528079 - layer.1.v_cache 0.00000088 0.00054158 - layer.2.k_cache 0.00116165 0.37781843 - layer.2.v_cache 0.00000115 0.00075175 - layer.3.k_cache 0.00130884 0.42618352 - layer.3.v_cache 0.00000264 0.00124486 - layer.4.k_cache 0.00339222 0.79676583 - layer.4.v_cache 0.00000308 0.00211192 - layer.4.output 0.00017467 0.07725616 - ------------------------------------------------------------------------------------- - TOTAL 0.00256716 0.71545179 - (elements=2,408,448) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2408448 -Total Bytes 554016 -BPFP 1.8402 bits/point -EBPFP 3.6805 equivalent bits/point -MSE 0.715452 ----------------------- -------------------------------------------------------- -Time: 3.524s Load: 0.009s, Pack+Encode: 1.962s, Decode+Unpack: 1.553s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 168, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7155 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample73-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample73-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample76-layer4-item1.zst (79/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample76-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 164, 128) -Output shape: (1, 164, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) -> torch.Size([1, 1, 164, 1024]) - layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,124B, BPFP=0.4346 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,416B, BPFP=2.3540 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,736B, BPFP=1.6547 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,416B, BPFP=2.4970 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,996B, BPFP=2.0958 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,296B, BPFP=2.5389 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,472B, BPFP=2.1185 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,536B, BPFP=2.5027 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,296B, BPFP=1.6814 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,720B, BPFP=2.5591 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 124,944B, BPFP=1.4880 -⌛️ [2/4] FRONTEND: Frontend time: 1.960s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.542s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 164, 128]) - layer.0.v_cache: torch.Size([1, 8, 164, 128]) - layer.1.k_cache: torch.Size([1, 8, 164, 128]) - layer.1.v_cache: torch.Size([1, 8, 164, 128]) - layer.2.k_cache: torch.Size([1, 8, 164, 128]) - layer.2.v_cache: torch.Size([1, 8, 164, 128]) - layer.3.k_cache: torch.Size([1, 8, 164, 128]) - layer.3.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.k_cache: torch.Size([1, 8, 164, 128]) - layer.4.v_cache: torch.Size([1, 8, 164, 128]) - layer.4.output: torch.Size([1, 164, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02737338 7.22773557 - layer.0.v_cache 0.00000027 0.00015106 - layer.1.k_cache 0.00299196 0.85481104 - layer.1.v_cache 0.00000082 0.00053753 - layer.2.k_cache 0.00116550 0.37077778 - layer.2.v_cache 0.00000112 0.00073937 - layer.3.k_cache 0.00132297 0.41746488 - layer.3.v_cache 0.00000204 0.00116086 - layer.4.k_cache 0.00347967 0.80705847 - layer.4.v_cache 0.00000307 0.00206173 - layer.4.output 0.00016775 0.07589911 - ------------------------------------------------------------------------------------- - TOTAL 0.00264370 0.71329248 - (elements=2,351,104) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2351104 -Total Bytes 553952 -BPFP 1.8849 bits/point -EBPFP 3.7698 equivalent bits/point -MSE 0.713292 ----------------------- -------------------------------------------------------- -Time: 3.512s Load: 0.010s, Pack+Encode: 1.960s, Decode+Unpack: 1.542s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 164, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7133 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample76-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample76-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample77-layer4-item1.zst (80/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample77-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 175, 128) -Output shape: (1, 175, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,768B, BPFP=0.4361 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,956B, BPFP=2.2302 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,768B, BPFP=1.5968 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,900B, BPFP=2.3616 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,108B, BPFP=2.0137 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,660B, BPFP=2.3955 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,460B, BPFP=1.9848 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,188B, BPFP=2.3745 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,128B, BPFP=1.6575 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,996B, BPFP=2.4105 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 163,052B, BPFP=1.8198 -⌛️ [2/4] FRONTEND: Frontend time: 1.999s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.572s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02704878 7.11632045 - layer.0.v_cache 0.00000028 0.00014893 - layer.1.k_cache 0.00295012 0.81285967 - layer.1.v_cache 0.00000089 0.00054692 - layer.2.k_cache 0.00115680 0.36732827 - layer.2.v_cache 0.00000112 0.00075157 - layer.3.k_cache 0.00133986 0.42535719 - layer.3.v_cache 0.00000218 0.00121467 - layer.4.k_cache 0.00335044 0.88170253 - layer.4.v_cache 0.00000310 0.00204632 - layer.4.output 0.00017962 0.06738668 - ------------------------------------------------------------------------------------- - TOTAL 0.00261229 0.70555880 - (elements=2,508,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2508800 -Total Bytes 598984 -BPFP 1.9100 bits/point -EBPFP 3.8201 equivalent bits/point -MSE 0.705559 ----------------------- -------------------------------------------------------- -Time: 3.580s Load: 0.010s, Pack+Encode: 1.999s, Decode+Unpack: 1.572s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7056 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample77-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample77-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample78-layer4-item1.zst (81/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample78-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 166, 128) -Output shape: (1, 166, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) -> torch.Size([1, 1, 166, 1024]) - layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,212B, BPFP=0.4335 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,608B, BPFP=2.3347 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,448B, BPFP=1.6212 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,404B, BPFP=2.4663 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,292B, BPFP=2.0845 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,964B, BPFP=2.4927 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,636B, BPFP=2.1007 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,376B, BPFP=2.4650 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,356B, BPFP=1.6640 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,512B, BPFP=2.5184 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 125,768B, BPFP=1.4798 -⌛️ [2/4] FRONTEND: Frontend time: 1.969s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.547s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 166, 128]) - layer.0.v_cache: torch.Size([1, 8, 166, 128]) - layer.1.k_cache: torch.Size([1, 8, 166, 128]) - layer.1.v_cache: torch.Size([1, 8, 166, 128]) - layer.2.k_cache: torch.Size([1, 8, 166, 128]) - layer.2.v_cache: torch.Size([1, 8, 166, 128]) - layer.3.k_cache: torch.Size([1, 8, 166, 128]) - layer.3.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.k_cache: torch.Size([1, 8, 166, 128]) - layer.4.v_cache: torch.Size([1, 8, 166, 128]) - layer.4.output: torch.Size([1, 166, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02721562 6.98317342 - layer.0.v_cache 0.00000027 0.00015048 - layer.1.k_cache 0.00308722 0.83615351 - layer.1.v_cache 0.00000079 0.00053689 - layer.2.k_cache 0.00116669 0.36977828 - layer.2.v_cache 0.00000107 0.00072318 - layer.3.k_cache 0.00132320 0.41484460 - layer.3.v_cache 0.00000211 0.00119257 - layer.4.k_cache 0.00344917 0.82114787 - layer.4.v_cache 0.00000308 0.00208274 - layer.4.output 0.00019248 0.06790527 - ------------------------------------------------------------------------------------- - TOTAL 0.00264423 0.69295747 - (elements=2,379,776) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2379776 -Total Bytes 554576 -BPFP 1.8643 bits/point -EBPFP 3.7286 equivalent bits/point -MSE 0.692957 ----------------------- -------------------------------------------------------- -Time: 3.525s Load: 0.009s, Pack+Encode: 1.969s, Decode+Unpack: 1.547s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 166, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6930 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample78-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample78-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample79-layer4-item1.zst (82/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample79-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 157, 128) -Output shape: (1, 157, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,860B, BPFP=0.4409 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,476B, BPFP=2.5117 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,284B, BPFP=1.6562 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,712B, BPFP=2.6230 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,628B, BPFP=2.1710 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,480B, BPFP=2.6612 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,268B, BPFP=2.2028 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,088B, BPFP=2.6417 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,540B, BPFP=1.7188 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,228B, BPFP=2.6984 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 114,464B, BPFP=1.4240 -⌛️ [2/4] FRONTEND: Frontend time: 1.970s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02715074 7.55324676 - layer.0.v_cache 0.00000026 0.00015441 - layer.1.k_cache 0.00316500 0.72423102 - layer.1.v_cache 0.00000094 0.00053459 - layer.2.k_cache 0.00119041 0.39728094 - layer.2.v_cache 0.00000113 0.00073788 - layer.3.k_cache 0.00134161 0.41684553 - layer.3.v_cache 0.00000209 0.00117461 - layer.4.k_cache 0.00347424 0.84862659 - layer.4.v_cache 0.00000301 0.00200901 - layer.4.output 0.00014407 0.07604947 - ------------------------------------------------------------------------------------- - TOTAL 0.00263612 0.73207423 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 543028 -BPFP 1.9301 bits/point -EBPFP 3.8602 equivalent bits/point -MSE 0.732074 ----------------------- -------------------------------------------------------- -Time: 3.531s Load: 0.009s, Pack+Encode: 1.970s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7321 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample79-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample79-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample8-layer4-item1.zst (83/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample8-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 235, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.013s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 235, 128) -Output shape: (1, 235, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.0.v_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.1.k_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.1.v_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.2.k_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.2.v_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.3.k_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.3.v_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.4.k_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.4.v_cache: torch.Size([1, 8, 235, 128]) -> torch.Size([1, 1, 235, 1024]) - layer.4.output: torch.Size([1, 235, 4096]) -> torch.Size([1, 1, 235, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 12,604B, BPFP=0.4190 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,588B, BPFP=2.1805 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,760B, BPFP=1.5545 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,096B, BPFP=2.2971 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,340B, BPFP=1.9395 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,236B, BPFP=2.3350 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,112B, BPFP=1.9652 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,796B, BPFP=2.3203 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 49,796B, BPFP=1.6555 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,140B, BPFP=2.3650 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 156,324B, BPFP=1.2992 -⌛️ [2/4] FRONTEND: Frontend time: 2.385s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 235, 128]) - layer.0.v_cache: torch.Size([1, 8, 235, 128]) - layer.1.k_cache: torch.Size([1, 8, 235, 128]) - layer.1.v_cache: torch.Size([1, 8, 235, 128]) - layer.2.k_cache: torch.Size([1, 8, 235, 128]) - layer.2.v_cache: torch.Size([1, 8, 235, 128]) - layer.3.k_cache: torch.Size([1, 8, 235, 128]) - layer.3.v_cache: torch.Size([1, 8, 235, 128]) - layer.4.k_cache: torch.Size([1, 8, 235, 128]) - layer.4.v_cache: torch.Size([1, 8, 235, 128]) - layer.4.output: torch.Size([1, 235, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.935s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 235, 128]) - layer.0.v_cache: torch.Size([1, 8, 235, 128]) - layer.1.k_cache: torch.Size([1, 8, 235, 128]) - layer.1.v_cache: torch.Size([1, 8, 235, 128]) - layer.2.k_cache: torch.Size([1, 8, 235, 128]) - layer.2.v_cache: torch.Size([1, 8, 235, 128]) - layer.3.k_cache: torch.Size([1, 8, 235, 128]) - layer.3.v_cache: torch.Size([1, 8, 235, 128]) - layer.4.k_cache: torch.Size([1, 8, 235, 128]) - layer.4.v_cache: torch.Size([1, 8, 235, 128]) - layer.4.output: torch.Size([1, 235, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02682624 6.74952315 - layer.0.v_cache 0.00000026 0.00014332 - layer.1.k_cache 0.00299644 0.82773619 - layer.1.v_cache 0.00000075 0.00048542 - layer.2.k_cache 0.00117112 0.36677964 - layer.2.v_cache 0.00000116 0.00066158 - layer.3.k_cache 0.00131946 0.40539720 - layer.3.v_cache 0.00000202 0.00106729 - layer.4.k_cache 0.00363359 0.83362401 - layer.4.v_cache 0.00000293 0.00182224 - layer.4.output 0.00013720 0.05790972 - ------------------------------------------------------------------------------------- - TOTAL 0.00260734 0.67277707 - (elements=3,368,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3368960 -Total Bytes 728792 -BPFP 1.7306 bits/point -EBPFP 3.4612 equivalent bits/point -MSE 0.672777 ----------------------- -------------------------------------------------------- -Time: 4.333s Load: 0.013s, Pack+Encode: 2.385s, Decode+Unpack: 1.935s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 235, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 235, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6728 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample8-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample8-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample81-layer4-item1.zst (84/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample81-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 160, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 160, 128) -Output shape: (1, 160, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.0.v_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.1.k_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.1.v_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.2.k_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.2.v_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.3.k_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.3.v_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.4.k_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.4.v_cache: torch.Size([1, 8, 160, 128]) -> torch.Size([1, 1, 160, 1024]) - layer.4.output: torch.Size([1, 160, 4096]) -> torch.Size([1, 1, 160, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,028B, BPFP=0.4408 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,780B, BPFP=2.4307 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,772B, BPFP=1.6490 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,320B, BPFP=2.5547 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,524B, BPFP=2.1252 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,352B, BPFP=2.6051 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,392B, BPFP=2.1676 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,668B, BPFP=2.5717 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,952B, BPFP=1.7066 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,008B, BPFP=2.6371 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 126,212B, BPFP=1.5407 -⌛️ [2/4] FRONTEND: Frontend time: 1.947s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 160, 128]) - layer.0.v_cache: torch.Size([1, 8, 160, 128]) - layer.1.k_cache: torch.Size([1, 8, 160, 128]) - layer.1.v_cache: torch.Size([1, 8, 160, 128]) - layer.2.k_cache: torch.Size([1, 8, 160, 128]) - layer.2.v_cache: torch.Size([1, 8, 160, 128]) - layer.3.k_cache: torch.Size([1, 8, 160, 128]) - layer.3.v_cache: torch.Size([1, 8, 160, 128]) - layer.4.k_cache: torch.Size([1, 8, 160, 128]) - layer.4.v_cache: torch.Size([1, 8, 160, 128]) - layer.4.output: torch.Size([1, 160, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 160, 128]) - layer.0.v_cache: torch.Size([1, 8, 160, 128]) - layer.1.k_cache: torch.Size([1, 8, 160, 128]) - layer.1.v_cache: torch.Size([1, 8, 160, 128]) - layer.2.k_cache: torch.Size([1, 8, 160, 128]) - layer.2.v_cache: torch.Size([1, 8, 160, 128]) - layer.3.k_cache: torch.Size([1, 8, 160, 128]) - layer.3.v_cache: torch.Size([1, 8, 160, 128]) - layer.4.k_cache: torch.Size([1, 8, 160, 128]) - layer.4.v_cache: torch.Size([1, 8, 160, 128]) - layer.4.output: torch.Size([1, 160, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02677556 7.30297623 - layer.0.v_cache 0.00000027 0.00015257 - layer.1.k_cache 0.00313261 0.84546795 - layer.1.v_cache 0.00000088 0.00056550 - layer.2.k_cache 0.00115055 0.36760426 - layer.2.v_cache 0.00000113 0.00078710 - layer.3.k_cache 0.00131762 0.41041093 - layer.3.v_cache 0.00000244 0.00123959 - layer.4.k_cache 0.00341674 0.77880564 - layer.4.v_cache 0.00000319 0.00219891 - layer.4.output 0.00016336 0.06807387 - ------------------------------------------------------------------------------------- - TOTAL 0.00260389 0.71303601 - (elements=2,293,760) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2293760 -Total Bytes 554008 -BPFP 1.9322 bits/point -EBPFP 3.8645 equivalent bits/point -MSE 0.713036 ----------------------- -------------------------------------------------------- -Time: 3.507s Load: 0.008s, Pack+Encode: 1.947s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 160, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 160, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7130 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample81-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample81-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample82-layer4-item1.zst (85/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample82-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 163, 128) -Output shape: (1, 163, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.output: torch.Size([1, 163, 4096]) -> torch.Size([1, 1, 163, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,048B, BPFP=0.4337 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,508B, BPFP=2.3729 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,928B, BPFP=1.6262 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,232B, BPFP=2.5035 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,284B, BPFP=2.1225 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,928B, BPFP=2.5368 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,024B, BPFP=2.1100 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,408B, BPFP=2.5119 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,400B, BPFP=1.7446 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,464B, BPFP=2.5625 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 143,408B, BPFP=1.7184 -⌛️ [2/4] FRONTEND: Frontend time: 2.007s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.585s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02751794 7.27919203 - layer.0.v_cache 0.00000027 0.00014958 - layer.1.k_cache 0.00304081 0.90879532 - layer.1.v_cache 0.00000086 0.00055023 - layer.2.k_cache 0.00115337 0.36238395 - layer.2.v_cache 0.00000112 0.00074420 - layer.3.k_cache 0.00134529 0.40930279 - layer.3.v_cache 0.00000213 0.00121254 - layer.4.k_cache 0.00339313 0.79944709 - layer.4.v_cache 0.00000308 0.00211503 - layer.4.output 0.00017108 0.07738645 - ------------------------------------------------------------------------------------- - TOTAL 0.00265302 0.71953133 - (elements=2,336,768) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2336768 -Total Bytes 571632 -BPFP 1.9570 bits/point -EBPFP 3.9140 equivalent bits/point -MSE 0.719531 ----------------------- -------------------------------------------------------- -Time: 3.601s Load: 0.010s, Pack+Encode: 2.007s, Decode+Unpack: 1.585s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7195 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample82-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample82-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample84-layer4-item1.zst (86/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample84-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,340B, BPFP=0.4218 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,476B, BPFP=2.2343 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,628B, BPFP=1.5638 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,072B, BPFP=2.3967 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,652B, BPFP=2.0164 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,704B, BPFP=2.4252 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,880B, BPFP=2.0267 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,264B, BPFP=2.4053 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,044B, BPFP=1.6729 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,144B, BPFP=2.4451 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 115,048B, BPFP=1.2989 -⌛️ [2/4] FRONTEND: Frontend time: 1.987s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.561s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02651478 7.01413052 - layer.0.v_cache 0.00000027 0.00014778 - layer.1.k_cache 0.00306594 0.84840420 - layer.1.v_cache 0.00000080 0.00052363 - layer.2.k_cache 0.00117582 0.37522015 - layer.2.v_cache 0.00000111 0.00075626 - layer.3.k_cache 0.00132699 0.41998895 - layer.3.v_cache 0.00000224 0.00118786 - layer.4.k_cache 0.00342058 0.80689438 - layer.4.v_cache 0.00000347 0.00202141 - layer.4.output 0.00021934 0.06898694 - ------------------------------------------------------------------------------------- - TOTAL 0.00259924 0.69608735 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 549252 -BPFP 1.7717 bits/point -EBPFP 3.5434 equivalent bits/point -MSE 0.696087 ----------------------- -------------------------------------------------------- -Time: 3.558s Load: 0.009s, Pack+Encode: 1.987s, Decode+Unpack: 1.561s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6961 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample84-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample84-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample85-layer4-item1.zst (87/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample85-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 170, 128) -Output shape: (1, 170, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) -> torch.Size([1, 1, 170, 1024]) - layer.4.output: torch.Size([1, 170, 4096]) -> torch.Size([1, 1, 170, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,640B, BPFP=0.4430 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,000B, BPFP=2.2978 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,160B, BPFP=1.6158 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,140B, BPFP=2.4421 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,512B, BPFP=2.0456 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,564B, BPFP=2.4616 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,364B, BPFP=2.0388 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,908B, BPFP=2.4314 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,360B, BPFP=1.6710 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,864B, BPFP=2.4754 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 142,436B, BPFP=1.6364 -⌛️ [2/4] FRONTEND: Frontend time: 1.952s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.output: torch.Size([1, 170, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.542s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 170, 128]) - layer.0.v_cache: torch.Size([1, 8, 170, 128]) - layer.1.k_cache: torch.Size([1, 8, 170, 128]) - layer.1.v_cache: torch.Size([1, 8, 170, 128]) - layer.2.k_cache: torch.Size([1, 8, 170, 128]) - layer.2.v_cache: torch.Size([1, 8, 170, 128]) - layer.3.k_cache: torch.Size([1, 8, 170, 128]) - layer.3.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.k_cache: torch.Size([1, 8, 170, 128]) - layer.4.v_cache: torch.Size([1, 8, 170, 128]) - layer.4.output: torch.Size([1, 170, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02725268 6.91794434 - layer.0.v_cache 0.00000026 0.00015181 - layer.1.k_cache 0.00300898 0.78198440 - layer.1.v_cache 0.00000090 0.00057098 - layer.2.k_cache 0.00114627 0.37067683 - layer.2.v_cache 0.00000112 0.00075471 - layer.3.k_cache 0.00135579 0.41289224 - layer.3.v_cache 0.00000210 0.00118795 - layer.4.k_cache 0.00341688 0.82261326 - layer.4.v_cache 0.00000298 0.00201549 - layer.4.output 0.00021049 0.07687726 - ------------------------------------------------------------------------------------- - TOTAL 0.00264500 0.68702150 - (elements=2,437,120) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2437120 -Total Bytes 575948 -BPFP 1.8906 bits/point -EBPFP 3.7812 equivalent bits/point -MSE 0.687022 ----------------------- -------------------------------------------------------- -Time: 3.503s Load: 0.009s, Pack+Encode: 1.952s, Decode+Unpack: 1.542s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 170, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6870 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample85-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample85-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample86-layer4-item1.zst (88/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample86-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 163, 128) -Output shape: (1, 163, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) -> torch.Size([1, 1, 163, 1024]) - layer.4.output: torch.Size([1, 163, 4096]) -> torch.Size([1, 1, 163, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,984B, BPFP=0.4306 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,564B, BPFP=2.3756 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,496B, BPFP=1.6534 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,436B, BPFP=2.5132 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,488B, BPFP=2.1323 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,044B, BPFP=2.5424 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,172B, BPFP=2.1171 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,488B, BPFP=2.5157 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,028B, BPFP=1.7747 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,600B, BPFP=2.5690 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 128,128B, BPFP=1.5353 -⌛️ [2/4] FRONTEND: Frontend time: 1.969s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.562s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 163, 128]) - layer.0.v_cache: torch.Size([1, 8, 163, 128]) - layer.1.k_cache: torch.Size([1, 8, 163, 128]) - layer.1.v_cache: torch.Size([1, 8, 163, 128]) - layer.2.k_cache: torch.Size([1, 8, 163, 128]) - layer.2.v_cache: torch.Size([1, 8, 163, 128]) - layer.3.k_cache: torch.Size([1, 8, 163, 128]) - layer.3.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.k_cache: torch.Size([1, 8, 163, 128]) - layer.4.v_cache: torch.Size([1, 8, 163, 128]) - layer.4.output: torch.Size([1, 163, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02663420 7.25025313 - layer.0.v_cache 0.00000027 0.00015591 - layer.1.k_cache 0.00304468 0.86801044 - layer.1.v_cache 0.00000080 0.00054513 - layer.2.k_cache 0.00117203 0.38188696 - layer.2.v_cache 0.00000109 0.00074692 - layer.3.k_cache 0.00131422 0.41309535 - layer.3.v_cache 0.00000209 0.00122250 - layer.4.k_cache 0.00347827 0.79777124 - layer.4.v_cache 0.00000310 0.00209392 - layer.4.output 0.00017721 0.07194390 - ------------------------------------------------------------------------------------- - TOTAL 0.00259711 0.71453979 - (elements=2,336,768) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2336768 -Total Bytes 558428 -BPFP 1.9118 bits/point -EBPFP 3.8236 equivalent bits/point -MSE 0.714540 ----------------------- -------------------------------------------------------- -Time: 3.540s Load: 0.009s, Pack+Encode: 1.969s, Decode+Unpack: 1.562s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 163, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7145 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample86-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample87-layer4-item1.zst (89/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample87-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 175, 128) -Output shape: (1, 175, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) -> torch.Size([1, 1, 175, 1024]) - layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,984B, BPFP=0.4457 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,296B, BPFP=2.2454 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,852B, BPFP=1.6005 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,108B, BPFP=2.3709 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,224B, BPFP=2.0189 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,924B, BPFP=2.4073 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,988B, BPFP=2.0084 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,204B, BPFP=2.3752 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,056B, BPFP=1.6543 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,256B, BPFP=2.4221 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 147,792B, BPFP=1.6495 -⌛️ [2/4] FRONTEND: Frontend time: 1.977s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.515s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 175, 128]) - layer.0.v_cache: torch.Size([1, 8, 175, 128]) - layer.1.k_cache: torch.Size([1, 8, 175, 128]) - layer.1.v_cache: torch.Size([1, 8, 175, 128]) - layer.2.k_cache: torch.Size([1, 8, 175, 128]) - layer.2.v_cache: torch.Size([1, 8, 175, 128]) - layer.3.k_cache: torch.Size([1, 8, 175, 128]) - layer.3.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.k_cache: torch.Size([1, 8, 175, 128]) - layer.4.v_cache: torch.Size([1, 8, 175, 128]) - layer.4.output: torch.Size([1, 175, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02610176 7.05002790 - layer.0.v_cache 0.00000027 0.00015134 - layer.1.k_cache 0.00311052 0.93071455 - layer.1.v_cache 0.00000081 0.00054658 - layer.2.k_cache 0.00119380 0.37455693 - layer.2.v_cache 0.00000114 0.00074748 - layer.3.k_cache 0.00132266 0.40711121 - layer.3.v_cache 0.00000212 0.00118440 - layer.4.k_cache 0.00346741 0.77740766 - layer.4.v_cache 0.00000313 0.00207388 - layer.4.output 0.00016538 0.06734088 - ------------------------------------------------------------------------------------- - TOTAL 0.00256179 0.70099182 - (elements=2,508,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2508800 -Total Bytes 585684 -BPFP 1.8676 bits/point -EBPFP 3.7352 equivalent bits/point -MSE 0.700992 ----------------------- -------------------------------------------------------- -Time: 3.501s Load: 0.009s, Pack+Encode: 1.977s, Decode+Unpack: 1.515s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 175, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7010 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample87-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample87-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample88-layer4-item1.zst (90/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample88-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.011s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 172, 128) -Output shape: (1, 172, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.0.v_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.1.k_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.1.v_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.2.k_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.2.v_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.3.k_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.3.v_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.4.k_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.4.v_cache: torch.Size([1, 8, 172, 128]) -> torch.Size([1, 1, 172, 1024]) - layer.4.output: torch.Size([1, 172, 4096]) -> torch.Size([1, 1, 172, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,860B, BPFP=0.4479 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,264B, BPFP=2.2831 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,944B, BPFP=1.6326 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,928B, BPFP=2.4041 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,236B, BPFP=2.0093 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,464B, BPFP=2.4284 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,740B, BPFP=2.0322 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,744B, BPFP=2.3957 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,304B, BPFP=1.6490 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,012B, BPFP=2.4533 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 136,580B, BPFP=1.5509 -⌛️ [2/4] FRONTEND: Frontend time: 1.925s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 172, 128]) - layer.0.v_cache: torch.Size([1, 8, 172, 128]) - layer.1.k_cache: torch.Size([1, 8, 172, 128]) - layer.1.v_cache: torch.Size([1, 8, 172, 128]) - layer.2.k_cache: torch.Size([1, 8, 172, 128]) - layer.2.v_cache: torch.Size([1, 8, 172, 128]) - layer.3.k_cache: torch.Size([1, 8, 172, 128]) - layer.3.v_cache: torch.Size([1, 8, 172, 128]) - layer.4.k_cache: torch.Size([1, 8, 172, 128]) - layer.4.v_cache: torch.Size([1, 8, 172, 128]) - layer.4.output: torch.Size([1, 172, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.411s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 172, 128]) - layer.0.v_cache: torch.Size([1, 8, 172, 128]) - layer.1.k_cache: torch.Size([1, 8, 172, 128]) - layer.1.v_cache: torch.Size([1, 8, 172, 128]) - layer.2.k_cache: torch.Size([1, 8, 172, 128]) - layer.2.v_cache: torch.Size([1, 8, 172, 128]) - layer.3.k_cache: torch.Size([1, 8, 172, 128]) - layer.3.v_cache: torch.Size([1, 8, 172, 128]) - layer.4.k_cache: torch.Size([1, 8, 172, 128]) - layer.4.v_cache: torch.Size([1, 8, 172, 128]) - layer.4.output: torch.Size([1, 172, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02694283 6.93292094 - layer.0.v_cache 0.00000026 0.00015237 - layer.1.k_cache 0.00313726 0.78183658 - layer.1.v_cache 0.00000082 0.00053654 - layer.2.k_cache 0.00118743 0.37503282 - layer.2.v_cache 0.00000107 0.00072734 - layer.3.k_cache 0.00136375 0.42292852 - layer.3.v_cache 0.00000202 0.00114980 - layer.4.k_cache 0.00345272 0.82930081 - layer.4.v_cache 0.00000296 0.00201868 - layer.4.output 0.00020827 0.07066828 - ------------------------------------------------------------------------------------- - TOTAL 0.00263744 0.68780554 - (elements=2,465,792) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2465792 -Total Bytes 571076 -BPFP 1.8528 bits/point -EBPFP 3.7056 equivalent bits/point -MSE 0.687806 ----------------------- -------------------------------------------------------- -Time: 3.347s Load: 0.011s, Pack+Encode: 1.925s, Decode+Unpack: 1.411s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 172, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6878 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample88-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample88-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample89-layer4-item1.zst (91/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample89-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 169, 128) -Output shape: (1, 169, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.output: torch.Size([1, 169, 4096]) -> torch.Size([1, 1, 169, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,484B, BPFP=0.4384 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,768B, BPFP=2.3007 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,060B, BPFP=1.6207 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,688B, BPFP=2.4357 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,532B, BPFP=2.0586 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,184B, BPFP=2.4586 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,628B, BPFP=2.0631 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,872B, BPFP=2.4442 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,052B, BPFP=1.6666 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,660B, BPFP=2.4806 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 108,584B, BPFP=1.2549 -⌛️ [2/4] FRONTEND: Frontend time: 1.975s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.603s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02715943 6.74454006 - layer.0.v_cache 0.00000026 0.00015115 - layer.1.k_cache 0.00301211 0.77122976 - layer.1.v_cache 0.00000078 0.00052087 - layer.2.k_cache 0.00115479 0.37354671 - layer.2.v_cache 0.00000109 0.00073046 - layer.3.k_cache 0.00130335 0.42372605 - layer.3.v_cache 0.00000203 0.00115023 - layer.4.k_cache 0.00348027 0.81796915 - layer.4.v_cache 0.00000300 0.00195635 - layer.4.output 0.00019495 0.07125500 - ------------------------------------------------------------------------------------- - TOTAL 0.00263549 0.67289577 - (elements=2,422,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2422784 -Total Bytes 540512 -BPFP 1.7848 bits/point -EBPFP 3.5695 equivalent bits/point -MSE 0.672896 ----------------------- -------------------------------------------------------- -Time: 3.587s Load: 0.008s, Pack+Encode: 1.975s, Decode+Unpack: 1.603s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6729 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample89-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample89-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample9-layer4-item1.zst (92/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample9-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 216, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.012s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 216, 128) -Output shape: (1, 216, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.0.v_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.1.k_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.1.v_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.2.k_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.2.v_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.3.k_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.3.v_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.4.k_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.4.v_cache: torch.Size([1, 8, 216, 128]) -> torch.Size([1, 1, 216, 1024]) - layer.4.output: torch.Size([1, 216, 4096]) -> torch.Size([1, 1, 216, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 11,668B, BPFP=0.4220 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,876B, BPFP=2.3827 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,396B, BPFP=1.6058 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,092B, BPFP=2.5352 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,100B, BPFP=2.0652 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,200B, BPFP=2.5752 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,984B, BPFP=2.0972 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,620B, BPFP=2.5543 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,148B, BPFP=1.5968 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 70,524B, BPFP=2.5508 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 188,624B, BPFP=1.7056 -⌛️ [2/4] FRONTEND: Frontend time: 2.367s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 216, 128]) - layer.0.v_cache: torch.Size([1, 8, 216, 128]) - layer.1.k_cache: torch.Size([1, 8, 216, 128]) - layer.1.v_cache: torch.Size([1, 8, 216, 128]) - layer.2.k_cache: torch.Size([1, 8, 216, 128]) - layer.2.v_cache: torch.Size([1, 8, 216, 128]) - layer.3.k_cache: torch.Size([1, 8, 216, 128]) - layer.3.v_cache: torch.Size([1, 8, 216, 128]) - layer.4.k_cache: torch.Size([1, 8, 216, 128]) - layer.4.v_cache: torch.Size([1, 8, 216, 128]) - layer.4.output: torch.Size([1, 216, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.147s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 216, 128]) - layer.0.v_cache: torch.Size([1, 8, 216, 128]) - layer.1.k_cache: torch.Size([1, 8, 216, 128]) - layer.1.v_cache: torch.Size([1, 8, 216, 128]) - layer.2.k_cache: torch.Size([1, 8, 216, 128]) - layer.2.v_cache: torch.Size([1, 8, 216, 128]) - layer.3.k_cache: torch.Size([1, 8, 216, 128]) - layer.3.v_cache: torch.Size([1, 8, 216, 128]) - layer.4.k_cache: torch.Size([1, 8, 216, 128]) - layer.4.v_cache: torch.Size([1, 8, 216, 128]) - layer.4.output: torch.Size([1, 216, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02680264 6.90425392 - layer.0.v_cache 0.00000027 0.00014652 - layer.1.k_cache 0.00301343 0.78831786 - layer.1.v_cache 0.00000084 0.00051800 - layer.2.k_cache 0.00116753 0.36710951 - layer.2.v_cache 0.00000113 0.00071320 - layer.3.k_cache 0.00132686 0.39961024 - layer.3.v_cache 0.00000211 0.00117070 - layer.4.k_cache 0.00344202 0.78943620 - layer.4.v_cache 0.00000310 0.00200918 - layer.4.output 0.00021250 0.06805767 - ------------------------------------------------------------------------------------- - TOTAL 0.00261500 0.68039400 - (elements=3,096,576) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 3096576 -Total Bytes 752232 -BPFP 1.9434 bits/point -EBPFP 3.8868 equivalent bits/point -MSE 0.680394 ----------------------- -------------------------------------------------------- -Time: 4.526s Load: 0.012s, Pack+Encode: 2.367s, Decode+Unpack: 2.147s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 216, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 216, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6804 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample9-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample9-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample90-layer4-item1.zst (93/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample90-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 174, 128) -Output shape: (1, 174, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) -> torch.Size([1, 1, 174, 1024]) - layer.4.output: torch.Size([1, 174, 4096]) -> torch.Size([1, 1, 174, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,628B, BPFP=0.4323 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,860B, BPFP=2.2387 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,588B, BPFP=1.5979 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,668B, BPFP=2.3648 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,120B, BPFP=2.0259 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,300B, BPFP=2.3931 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,816B, BPFP=2.0122 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,820B, BPFP=2.3716 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,364B, BPFP=1.6327 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,892B, BPFP=2.4197 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 120,312B, BPFP=1.3505 -⌛️ [2/4] FRONTEND: Frontend time: 1.973s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.output: torch.Size([1, 174, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.577s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 174, 128]) - layer.0.v_cache: torch.Size([1, 8, 174, 128]) - layer.1.k_cache: torch.Size([1, 8, 174, 128]) - layer.1.v_cache: torch.Size([1, 8, 174, 128]) - layer.2.k_cache: torch.Size([1, 8, 174, 128]) - layer.2.v_cache: torch.Size([1, 8, 174, 128]) - layer.3.k_cache: torch.Size([1, 8, 174, 128]) - layer.3.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.k_cache: torch.Size([1, 8, 174, 128]) - layer.4.v_cache: torch.Size([1, 8, 174, 128]) - layer.4.output: torch.Size([1, 174, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02737866 7.06144346 - layer.0.v_cache 0.00000027 0.00014870 - layer.1.k_cache 0.00309889 0.88963178 - layer.1.v_cache 0.00000078 0.00051697 - layer.2.k_cache 0.00114661 0.38492663 - layer.2.v_cache 0.00000107 0.00072654 - layer.3.k_cache 0.00139116 0.43404467 - layer.3.v_cache 0.00000205 0.00118490 - layer.4.k_cache 0.00338123 0.88490541 - layer.4.v_cache 0.00000306 0.00206072 - layer.4.output 0.00019207 0.08517026 - ------------------------------------------------------------------------------------- - TOTAL 0.00265515 0.71430506 - (elements=2,494,464) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2494464 -Total Bytes 554368 -BPFP 1.7779 bits/point -EBPFP 3.5558 equivalent bits/point -MSE 0.714305 ----------------------- -------------------------------------------------------- -Time: 3.559s Load: 0.009s, Pack+Encode: 1.973s, Decode+Unpack: 1.577s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 174, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7143 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample90-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample90-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample91-layer4-item1.zst (94/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample91-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 169, 128) -Output shape: (1, 169, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) -> torch.Size([1, 1, 169, 1024]) - layer.4.output: torch.Size([1, 169, 4096]) -> torch.Size([1, 1, 169, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,160B, BPFP=0.4234 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,360B, BPFP=2.2818 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 34,808B, BPFP=1.6091 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,728B, BPFP=2.4375 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,928B, BPFP=2.0307 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,452B, BPFP=2.4710 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,356B, BPFP=2.0505 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,024B, BPFP=2.4512 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,112B, BPFP=1.6694 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,008B, BPFP=2.4967 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 105,928B, BPFP=1.2242 -⌛️ [2/4] FRONTEND: Frontend time: 1.978s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.504s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 169, 128]) - layer.0.v_cache: torch.Size([1, 8, 169, 128]) - layer.1.k_cache: torch.Size([1, 8, 169, 128]) - layer.1.v_cache: torch.Size([1, 8, 169, 128]) - layer.2.k_cache: torch.Size([1, 8, 169, 128]) - layer.2.v_cache: torch.Size([1, 8, 169, 128]) - layer.3.k_cache: torch.Size([1, 8, 169, 128]) - layer.3.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.k_cache: torch.Size([1, 8, 169, 128]) - layer.4.v_cache: torch.Size([1, 8, 169, 128]) - layer.4.output: torch.Size([1, 169, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02703965 7.20271545 - layer.0.v_cache 0.00000028 0.00015216 - layer.1.k_cache 0.00299766 0.89461955 - layer.1.v_cache 0.00000081 0.00051861 - layer.2.k_cache 0.00115006 0.37908561 - layer.2.v_cache 0.00000108 0.00071930 - layer.3.k_cache 0.00135242 0.41730061 - layer.3.v_cache 0.00000202 0.00111803 - layer.4.k_cache 0.00346273 0.81904774 - layer.4.v_cache 0.00000299 0.00198877 - layer.4.output 0.00018472 0.07102545 - ------------------------------------------------------------------------------------- - TOTAL 0.00262490 0.71438340 - (elements=2,422,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2422784 -Total Bytes 536864 -BPFP 1.7727 bits/point -EBPFP 3.5454 equivalent bits/point -MSE 0.714383 ----------------------- -------------------------------------------------------- -Time: 3.490s Load: 0.009s, Pack+Encode: 1.978s, Decode+Unpack: 1.504s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 169, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 169, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7144 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample91-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample91-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample92-layer4-item1.zst (95/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample92-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,808B, BPFP=0.4481 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,152B, BPFP=2.2913 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,176B, BPFP=1.6071 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,720B, BPFP=2.4086 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 45,136B, BPFP=2.0621 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,472B, BPFP=2.4430 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,532B, BPFP=2.0345 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,976B, BPFP=2.4203 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 37,040B, BPFP=1.6923 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,244B, BPFP=2.4783 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 128,032B, BPFP=1.4624 -⌛️ [2/4] FRONTEND: Frontend time: 1.974s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.577s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02767900 6.86201058 - layer.0.v_cache 0.00000027 0.00015207 - layer.1.k_cache 0.00300022 0.84340806 - layer.1.v_cache 0.00000079 0.00052304 - layer.2.k_cache 0.00116824 0.37298789 - layer.2.v_cache 0.00000115 0.00074081 - layer.3.k_cache 0.00134906 0.42220686 - layer.3.v_cache 0.00000206 0.00116872 - layer.4.k_cache 0.00347571 0.82313752 - layer.4.v_cache 0.00000302 0.00207136 - layer.4.output 0.00024629 0.07687821 - ------------------------------------------------------------------------------------- - TOTAL 0.00269033 0.68827998 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 563288 -BPFP 1.8382 bits/point -EBPFP 3.6764 equivalent bits/point -MSE 0.688280 ----------------------- -------------------------------------------------------- -Time: 3.561s Load: 0.010s, Pack+Encode: 1.974s, Decode+Unpack: 1.577s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6883 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample92-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample92-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample94-layer4-item1.zst (96/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample94-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 171, 128) -Output shape: (1, 171, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) -> torch.Size([1, 1, 171, 1024]) - layer.4.output: torch.Size([1, 171, 4096]) -> torch.Size([1, 1, 171, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,520B, BPFP=0.4349 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,780B, BPFP=2.2743 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,984B, BPFP=1.6440 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,380B, BPFP=2.3931 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,828B, BPFP=2.0481 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,088B, BPFP=2.4254 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,608B, BPFP=2.0380 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,508B, BPFP=2.3989 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,248B, BPFP=1.6561 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,688B, BPFP=2.4529 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 113,188B, BPFP=1.2928 -⌛️ [2/4] FRONTEND: Frontend time: 1.992s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.544s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 171, 128]) - layer.0.v_cache: torch.Size([1, 8, 171, 128]) - layer.1.k_cache: torch.Size([1, 8, 171, 128]) - layer.1.v_cache: torch.Size([1, 8, 171, 128]) - layer.2.k_cache: torch.Size([1, 8, 171, 128]) - layer.2.v_cache: torch.Size([1, 8, 171, 128]) - layer.3.k_cache: torch.Size([1, 8, 171, 128]) - layer.3.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.k_cache: torch.Size([1, 8, 171, 128]) - layer.4.v_cache: torch.Size([1, 8, 171, 128]) - layer.4.output: torch.Size([1, 171, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02680470 6.72643972 - layer.0.v_cache 0.00000027 0.00014852 - layer.1.k_cache 0.00310069 0.78939141 - layer.1.v_cache 0.00000077 0.00052098 - layer.2.k_cache 0.00116404 0.36717898 - layer.2.v_cache 0.00000106 0.00070061 - layer.3.k_cache 0.00136682 0.42177787 - layer.3.v_cache 0.00000197 0.00112721 - layer.4.k_cache 0.00339793 0.80313494 - layer.4.v_cache 0.00000297 0.00195990 - layer.4.output 0.00018087 0.06691946 - ------------------------------------------------------------------------------------- - TOTAL 0.00261177 0.67000414 - (elements=2,451,456) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2451456 -Total Bytes 545820 -BPFP 1.7812 bits/point -EBPFP 3.5624 equivalent bits/point -MSE 0.670004 ----------------------- -------------------------------------------------------- -Time: 3.545s Load: 0.009s, Pack+Encode: 1.992s, Decode+Unpack: 1.544s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 171, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 171, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6700 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample94-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample94-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample96-layer4-item1.zst (97/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample96-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 156, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 156, 128) -Output shape: (1, 156, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) -> torch.Size([1, 1, 156, 1024]) - layer.4.output: torch.Size([1, 156, 4096]) -> torch.Size([1, 1, 156, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,992B, BPFP=0.4503 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,176B, BPFP=2.5128 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,232B, BPFP=1.6643 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,540B, BPFP=2.6312 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,340B, BPFP=2.1705 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,004B, BPFP=2.6544 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,976B, BPFP=2.2524 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,564B, BPFP=2.6324 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,464B, BPFP=1.7260 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,436B, BPFP=2.6761 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 113,172B, BPFP=1.4169 -⌛️ [2/4] FRONTEND: Frontend time: 1.968s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.output: torch.Size([1, 156, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.508s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 156, 128]) - layer.0.v_cache: torch.Size([1, 8, 156, 128]) - layer.1.k_cache: torch.Size([1, 8, 156, 128]) - layer.1.v_cache: torch.Size([1, 8, 156, 128]) - layer.2.k_cache: torch.Size([1, 8, 156, 128]) - layer.2.v_cache: torch.Size([1, 8, 156, 128]) - layer.3.k_cache: torch.Size([1, 8, 156, 128]) - layer.3.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.k_cache: torch.Size([1, 8, 156, 128]) - layer.4.v_cache: torch.Size([1, 8, 156, 128]) - layer.4.output: torch.Size([1, 156, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02775914 7.51449272 - layer.0.v_cache 0.00000026 0.00015381 - layer.1.k_cache 0.00308099 0.82186890 - layer.1.v_cache 0.00000085 0.00055385 - layer.2.k_cache 0.00113398 0.37323174 - layer.2.v_cache 0.00000135 0.00076161 - layer.3.k_cache 0.00135748 0.42978111 - layer.3.v_cache 0.00000209 0.00121225 - layer.4.k_cache 0.00343364 0.81183747 - layer.4.v_cache 0.00000289 0.00202869 - layer.4.output 0.00020185 0.08400131 - ------------------------------------------------------------------------------------- - TOTAL 0.00268429 0.73513767 - (elements=2,236,416) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2236416 -Total Bytes 539896 -BPFP 1.9313 bits/point -EBPFP 3.8626 equivalent bits/point -MSE 0.735138 ----------------------- -------------------------------------------------------- -Time: 3.485s Load: 0.008s, Pack+Encode: 1.968s, Decode+Unpack: 1.508s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 156, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 156, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7351 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample96-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample96-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample97-layer4-item1.zst (98/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample97-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 162, 128) -Output shape: (1, 162, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) -> torch.Size([1, 1, 162, 1024]) - layer.4.output: torch.Size([1, 162, 4096]) -> torch.Size([1, 1, 162, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,028B, BPFP=0.4354 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,608B, BPFP=2.3924 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,892B, BPFP=1.6345 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,436B, BPFP=2.5287 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,396B, BPFP=2.0928 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,220B, BPFP=2.5666 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,244B, BPFP=2.1337 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,640B, BPFP=2.5386 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,060B, BPFP=1.6908 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,892B, BPFP=2.5990 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 134,468B, BPFP=1.6212 -⌛️ [2/4] FRONTEND: Frontend time: 1.954s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.output: torch.Size([1, 162, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.552s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 162, 128]) - layer.0.v_cache: torch.Size([1, 8, 162, 128]) - layer.1.k_cache: torch.Size([1, 8, 162, 128]) - layer.1.v_cache: torch.Size([1, 8, 162, 128]) - layer.2.k_cache: torch.Size([1, 8, 162, 128]) - layer.2.v_cache: torch.Size([1, 8, 162, 128]) - layer.3.k_cache: torch.Size([1, 8, 162, 128]) - layer.3.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.k_cache: torch.Size([1, 8, 162, 128]) - layer.4.v_cache: torch.Size([1, 8, 162, 128]) - layer.4.output: torch.Size([1, 162, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02742675 7.37397069 - layer.0.v_cache 0.00000027 0.00015379 - layer.1.k_cache 0.00301362 0.80997599 - layer.1.v_cache 0.00000083 0.00056568 - layer.2.k_cache 0.00118087 0.36390672 - layer.2.v_cache 0.00000111 0.00077373 - layer.3.k_cache 0.00131722 0.40297930 - layer.3.v_cache 0.00000209 0.00121321 - layer.4.k_cache 0.00343243 0.81176098 - layer.4.v_cache 0.00000314 0.00213717 - layer.4.output 0.00017655 0.06853638 - ------------------------------------------------------------------------------------- - TOTAL 0.00264889 0.71725591 - (elements=2,322,432) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2322432 -Total Bytes 561884 -BPFP 1.9355 bits/point -EBPFP 3.8710 equivalent bits/point -MSE 0.717256 ----------------------- -------------------------------------------------------- -Time: 3.515s Load: 0.008s, Pack+Encode: 1.954s, Decode+Unpack: 1.552s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 162, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7173 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample97-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample98-layer4-item1.zst (99/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample98-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 157, 128) -Output shape: (1, 157, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,020B, BPFP=0.4488 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,264B, BPFP=2.5012 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,172B, BPFP=1.6507 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,188B, BPFP=2.6467 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,792B, BPFP=2.1791 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,852B, BPFP=2.6797 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,488B, BPFP=2.2138 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,288B, BPFP=2.6517 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,304B, BPFP=1.7070 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,188B, BPFP=2.6965 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 129,352B, BPFP=1.6092 -⌛️ [2/4] FRONTEND: Frontend time: 1.960s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.515s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02749603 7.21471313 - layer.0.v_cache 0.00000027 0.00015264 - layer.1.k_cache 0.00316339 0.86155827 - layer.1.v_cache 0.00000087 0.00055526 - layer.2.k_cache 0.00115124 0.37742486 - layer.2.v_cache 0.00000113 0.00078490 - layer.3.k_cache 0.00133258 0.40626215 - layer.3.v_cache 0.00000205 0.00120228 - layer.4.k_cache 0.00350615 0.80100430 - layer.4.v_cache 0.00000300 0.00207509 - layer.4.output 0.00014089 0.06776572 - ------------------------------------------------------------------------------------- - TOTAL 0.00265859 0.70977112 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 558908 -BPFP 1.9866 bits/point -EBPFP 3.9731 equivalent bits/point -MSE 0.709771 ----------------------- -------------------------------------------------------- -Time: 3.483s Load: 0.008s, Pack+Encode: 1.960s, Decode+Unpack: 1.515s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7098 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample98-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample98-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample99-layer4-item1.zst (100/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample99-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 165, 128) -Output shape: (1, 165, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) -> torch.Size([1, 1, 165, 1024]) - layer.4.output: torch.Size([1, 165, 4096]) -> torch.Size([1, 1, 165, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,008B, BPFP=0.4265 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,644B, BPFP=2.3506 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,920B, BPFP=1.6061 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,012B, BPFP=2.4627 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,056B, BPFP=2.0860 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 52,928B, BPFP=2.5061 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,864B, BPFP=2.0769 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,112B, BPFP=2.4674 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,596B, BPFP=1.6854 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,228B, BPFP=2.5203 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 118,460B, BPFP=1.4022 -⌛️ [2/4] FRONTEND: Frontend time: 2.048s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.582s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 165, 128]) - layer.0.v_cache: torch.Size([1, 8, 165, 128]) - layer.1.k_cache: torch.Size([1, 8, 165, 128]) - layer.1.v_cache: torch.Size([1, 8, 165, 128]) - layer.2.k_cache: torch.Size([1, 8, 165, 128]) - layer.2.v_cache: torch.Size([1, 8, 165, 128]) - layer.3.k_cache: torch.Size([1, 8, 165, 128]) - layer.3.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.k_cache: torch.Size([1, 8, 165, 128]) - layer.4.v_cache: torch.Size([1, 8, 165, 128]) - layer.4.output: torch.Size([1, 165, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02734190 7.05345496 - layer.0.v_cache 0.00000026 0.00015236 - layer.1.k_cache 0.00308234 0.74842580 - layer.1.v_cache 0.00000078 0.00053980 - layer.2.k_cache 0.00115028 0.37040257 - layer.2.v_cache 0.00000111 0.00077242 - layer.3.k_cache 0.00133313 0.41164329 - layer.3.v_cache 0.00000202 0.00118408 - layer.4.k_cache 0.00348877 0.82233249 - layer.4.v_cache 0.00000299 0.00204564 - layer.4.output 0.00017595 0.06987631 - ------------------------------------------------------------------------------------- - TOTAL 0.00265053 0.69217562 - (elements=2,365,440) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2365440 -Total Bytes 544828 -BPFP 1.8426 bits/point -EBPFP 3.6853 equivalent bits/point -MSE 0.692176 ----------------------- -------------------------------------------------------- -Time: 3.638s Load: 0.009s, Pack+Encode: 2.048s, Decode+Unpack: 1.582s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 165, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6922 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_arc_challenge/sample99-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_arc_challenge/sample99-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 1.8352 bits/point -Avg EBPFP 3.6705 equivalent bits/point -Avg MSE 0.695902 -Avg Time 3.692s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:803b478b2189fa9ff3d18986ecfe55b6854fd1ea7ca22a5f2f27f0cc91bebc72 +size 1125213 diff --git a/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/dtufc_elic-featurecoding_qwen_individual.log b/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/dtufc_elic-featurecoding_qwen_individual.log index b76663f7ac7e4e7a77af66b2924d4c11d9f713a9..790f28247bfa37486b74afe964a0ca11f7035141 100644 --- a/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/dtufc_elic-featurecoding_qwen_individual.log +++ b/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/dtufc_elic-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_elic-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/dtufc_elic-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: elic-featurecoding - handler: qwen - checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar - transform_type: kmeans - transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json - bit_depth: 8 - device: cuda:0 -Loading checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 506 -Loaded elic-featurecoding (1-channel) on cuda:0 -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json -Loaded per-key mappings: model=qwen - Keys: ['layer.0.k_cache', 'layer.0.v_cache', 'layer.1.k_cache', 'layer.1.v_cache', 'layer.2.k_cache', 'layer.2.v_cache', 'layer.3.k_cache', 'layer.3.v_cache', 'layer.4.k_cache', 'layer.4.v_cache', 'layer.4.output'] ----------------- ------------------------------------------------------------------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding -Checkpoint codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Transform type kmeans -Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json -Input ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k -Output output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k ----------------- ------------------------------------------------------------------------------------------------------------------- -Files found: 100 ----------------------------------------------------------------------- - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst (1/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,732B, BPFP=0.4395 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,736B, BPFP=2.2460 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 35,192B, BPFP=1.5892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 51,988B, BPFP=2.3477 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 43,804B, BPFP=1.9781 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,244B, BPFP=2.4044 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 45,060B, BPFP=2.0349 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 52,904B, BPFP=2.3891 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,692B, BPFP=1.6570 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,880B, BPFP=2.4332 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 117,156B, BPFP=1.3227 -⌛️ [2/4] FRONTEND: Frontend time: 2.578s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.693s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02440321 7.59576310 - layer.0.v_cache 0.00000027 0.00014893 - layer.1.k_cache 0.00314874 0.84839255 - layer.1.v_cache 0.00000072 0.00049280 - layer.2.k_cache 0.00114712 0.38544610 - layer.2.v_cache 0.00000110 0.00071971 - layer.3.k_cache 0.00139354 0.44243287 - layer.3.v_cache 0.00000204 0.00120573 - layer.4.k_cache 0.00353492 0.89226616 - layer.4.v_cache 0.00000301 0.00204372 - layer.4.output 0.00018661 0.07144650 - ------------------------------------------------------------------------------------- - TOTAL 0.00245579 0.74676412 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 549388 -BPFP 1.7721 bits/point -EBPFP 3.5443 equivalent bits/point -MSE 0.746764 ----------------------- -------------------------------------------------------- -Time: 4.280s Load: 0.009s, Pack+Encode: 2.578s, Decode+Unpack: 1.693s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7468 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample1-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst (2/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 109, 128) -Output shape: (1, 109, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,248B, BPFP=0.4478 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,548B, BPFP=2.4045 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,152B, BPFP=1.6594 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,736B, BPFP=2.5614 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,564B, BPFP=1.9756 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,912B, BPFP=2.5740 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,396B, BPFP=2.1786 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,496B, BPFP=2.5442 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,908B, BPFP=1.7136 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,260B, BPFP=2.5989 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,940B, BPFP=1.2891 -⌛️ [2/4] FRONTEND: Frontend time: 2.061s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.341s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02465495 7.22924133 - layer.0.v_cache 0.00000027 0.00015594 - layer.1.k_cache 0.00330346 0.81017730 - layer.1.v_cache 0.00000080 0.00053040 - layer.2.k_cache 0.00115767 0.37879289 - layer.2.v_cache 0.00000114 0.00076955 - layer.3.k_cache 0.00132842 0.43200754 - layer.3.v_cache 0.00000211 0.00123851 - layer.4.k_cache 0.00335301 0.85036979 - layer.4.v_cache 0.00000290 0.00202210 - layer.4.output 0.00024947 0.08087296 - ------------------------------------------------------------------------------------- - TOTAL 0.00248590 0.71634266 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 360160 -BPFP 1.8439 bits/point -EBPFP 3.6877 equivalent bits/point -MSE 0.716343 ----------------------- -------------------------------------------------------- -Time: 3.407s Load: 0.006s, Pack+Encode: 2.061s, Decode+Unpack: 1.341s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7163 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample100-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst (3/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,896B, BPFP=0.4749 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,720B, BPFP=2.6353 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,568B, BPFP=1.4955 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,648B, BPFP=2.7906 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,824B, BPFP=2.1604 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,576B, BPFP=2.7848 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,932B, BPFP=2.2497 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,488B, BPFP=2.7777 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,212B, BPFP=1.6279 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,780B, BPFP=2.8012 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 74,524B, BPFP=1.5006 -⌛️ [2/4] FRONTEND: Frontend time: 1.772s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.371s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02579871 8.12591867 - layer.0.v_cache 0.00000028 0.00016056 - layer.1.k_cache 0.00331373 0.75017721 - layer.1.v_cache 0.00000086 0.00057120 - layer.2.k_cache 0.00113316 0.36803240 - layer.2.v_cache 0.00000115 0.00081128 - layer.3.k_cache 0.00137718 0.43829916 - layer.3.v_cache 0.00000224 0.00132451 - layer.4.k_cache 0.00334605 0.84982048 - layer.4.v_cache 0.00000324 0.00233221 - layer.4.output 0.00031880 0.07949906 - ------------------------------------------------------------------------------------- - TOTAL 0.00258942 0.77538885 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 345168 -BPFP 1.9857 bits/point -EBPFP 3.9715 equivalent bits/point -MSE 0.775389 ----------------------- -------------------------------------------------------- -Time: 3.149s Load: 0.006s, Pack+Encode: 1.772s, Decode+Unpack: 1.371s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7754 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample108-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst (4/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 50, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 50, 128) -Output shape: (1, 50, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.output: torch.Size([1, 50, 4096]) -> torch.Size([1, 1, 50, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 3,568B, BPFP=0.5575 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,588B, BPFP=2.5919 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 11,392B, BPFP=1.7800 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 17,612B, BPFP=2.7519 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 12,940B, BPFP=2.0219 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 17,592B, BPFP=2.7487 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,072B, BPFP=2.1987 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 17,328B, BPFP=2.7075 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 10,980B, BPFP=1.7156 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 17,568B, BPFP=2.7450 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 41,668B, BPFP=1.6277 -⌛️ [2/4] FRONTEND: Frontend time: 1.842s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.output: torch.Size([1, 50, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.089s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.output: torch.Size([1, 50, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02911560 9.37940552 - layer.0.v_cache 0.00000029 0.00018268 - layer.1.k_cache 0.00403332 0.85373657 - layer.1.v_cache 0.00000087 0.00065648 - layer.2.k_cache 0.00117108 0.45243065 - layer.2.v_cache 0.00000108 0.00088134 - layer.3.k_cache 0.00141424 0.53205544 - layer.3.v_cache 0.00000205 0.00142990 - layer.4.k_cache 0.00319313 1.11699661 - layer.4.v_cache 0.00000283 0.00234177 - layer.4.output 0.00027592 0.11202163 - ------------------------------------------------------------------------------------- - TOTAL 0.00285987 0.91344310 - (elements=716,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 716800 -Total Bytes 181308 -BPFP 2.0235 bits/point -EBPFP 4.0471 equivalent bits/point -MSE 0.913443 ----------------------- -------------------------------------------------------- -Time: 2.935s Load: 0.004s, Pack+Encode: 1.842s, Decode+Unpack: 1.089s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 50, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.9134 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample1087-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst (5/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 57, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 57, 128) -Output shape: (1, 57, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.output: torch.Size([1, 57, 4096]) -> torch.Size([1, 1, 57, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 3,980B, BPFP=0.5455 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,748B, BPFP=2.2955 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,764B, BPFP=1.8865 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 17,784B, BPFP=2.4375 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,536B, BPFP=1.9923 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 18,060B, BPFP=2.4753 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,096B, BPFP=2.0691 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 17,816B, BPFP=2.4419 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,500B, BPFP=1.8503 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 18,108B, BPFP=2.4819 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 49,000B, BPFP=1.6790 -⌛️ [2/4] FRONTEND: Frontend time: 1.588s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.output: torch.Size([1, 57, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.098s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.output: torch.Size([1, 57, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02901734 8.51534981 - layer.0.v_cache 0.00000028 0.00018043 - layer.1.k_cache 0.00371865 0.84039347 - layer.1.v_cache 0.00000081 0.00060686 - layer.2.k_cache 0.00114644 0.41538761 - layer.2.v_cache 0.00000109 0.00083754 - layer.3.k_cache 0.00141512 0.48444828 - layer.3.v_cache 0.00000208 0.00141497 - layer.4.k_cache 0.00325023 1.01342707 - layer.4.v_cache 0.00000285 0.00227173 - layer.4.output 0.00023785 0.09055538 - ------------------------------------------------------------------------------------- - TOTAL 0.00282188 0.83118138 - (elements=817,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 817152 -Total Bytes 198392 -BPFP 1.9423 bits/point -EBPFP 3.8846 equivalent bits/point -MSE 0.831181 ----------------------- -------------------------------------------------------- -Time: 2.691s Load: 0.004s, Pack+Encode: 1.588s, Decode+Unpack: 1.098s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 57, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8312 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample1128-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst (6/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 113, 128) -Output shape: (1, 113, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.output: torch.Size([1, 113, 4096]) -> torch.Size([1, 1, 113, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,636B, BPFP=0.4588 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,636B, BPFP=2.3255 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,548B, BPFP=1.5589 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,772B, BPFP=2.4732 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,540B, BPFP=2.0423 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,152B, BPFP=2.4994 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,392B, BPFP=2.1012 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,860B, BPFP=2.4793 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,860B, BPFP=1.6496 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,428B, BPFP=2.5185 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 83,176B, BPFP=1.4376 -⌛️ [2/4] FRONTEND: Frontend time: 1.834s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.337s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03016301 7.96720954 - layer.0.v_cache 0.00000028 0.00015601 - layer.1.k_cache 0.00339028 0.75456872 - layer.1.v_cache 0.00000081 0.00053772 - layer.2.k_cache 0.00113693 0.36561203 - layer.2.v_cache 0.00000114 0.00076866 - layer.3.k_cache 0.00135151 0.42398024 - layer.3.v_cache 0.00000219 0.00129721 - layer.4.k_cache 0.00328814 0.82821878 - layer.4.v_cache 0.00000314 0.00220788 - layer.4.output 0.00017637 0.07812702 - ------------------------------------------------------------------------------------- - TOTAL 0.00286021 0.76121892 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 374000 -BPFP 1.8470 bits/point -EBPFP 3.6939 equivalent bits/point -MSE 0.761219 ----------------------- -------------------------------------------------------- -Time: 3.178s Load: 0.007s, Pack+Encode: 1.834s, Decode+Unpack: 1.337s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7612 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample117-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst (7/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 108, 128) -Output shape: (1, 108, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,072B, BPFP=0.4392 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,048B, BPFP=2.3906 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,184B, BPFP=1.6047 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,248B, BPFP=2.5498 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,972B, BPFP=2.0234 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,604B, BPFP=2.5755 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,152B, BPFP=2.1811 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,396B, BPFP=2.5605 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,612B, BPFP=1.7080 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,104B, BPFP=2.6117 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,372B, BPFP=1.2365 -⌛️ [2/4] FRONTEND: Frontend time: 1.754s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.output: torch.Size([1, 108, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.339s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.output: torch.Size([1, 108, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02569981 7.57576497 - layer.0.v_cache 0.00000028 0.00015763 - layer.1.k_cache 0.00351470 0.79466728 - layer.1.v_cache 0.00000073 0.00053187 - layer.2.k_cache 0.00116483 0.38501418 - layer.2.v_cache 0.00000104 0.00075257 - layer.3.k_cache 0.00137767 0.43339118 - layer.3.v_cache 0.00000202 0.00125571 - layer.4.k_cache 0.00329212 0.88171606 - layer.4.v_cache 0.00000308 0.00212455 - layer.4.output 0.00021651 0.08316651 - ------------------------------------------------------------------------------------- - TOTAL 0.00256588 0.74343158 - (elements=1,548,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1548288 -Total Bytes 353764 -BPFP 1.8279 bits/point -EBPFP 3.6558 equivalent bits/point -MSE 0.743432 ----------------------- -------------------------------------------------------- -Time: 3.099s Load: 0.007s, Pack+Encode: 1.754s, Decode+Unpack: 1.339s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7434 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample120-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst (8/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 47, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 47, 128) -Output shape: (1, 47, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.output: torch.Size([1, 47, 4096]) -> torch.Size([1, 1, 47, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 3,436B, BPFP=0.5711 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,256B, BPFP=2.7021 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 10,644B, BPFP=1.7693 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 17,104B, BPFP=2.8431 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 12,772B, BPFP=2.1230 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 17,220B, BPFP=2.8624 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,276B, BPFP=2.3730 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 16,880B, BPFP=2.8059 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 10,472B, BPFP=1.7407 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 17,200B, BPFP=2.8590 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 36,976B, BPFP=1.5366 -⌛️ [2/4] FRONTEND: Frontend time: 1.567s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.output: torch.Size([1, 47, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.149s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.output: torch.Size([1, 47, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02987218 9.52902676 - layer.0.v_cache 0.00000031 0.00019733 - layer.1.k_cache 0.00397950 0.84132385 - layer.1.v_cache 0.00000076 0.00061007 - layer.2.k_cache 0.00125400 0.46210415 - layer.2.v_cache 0.00000109 0.00086893 - layer.3.k_cache 0.00146534 0.51007859 - layer.3.v_cache 0.00000206 0.00143520 - layer.4.k_cache 0.00325167 1.11009557 - layer.4.v_cache 0.00000289 0.00231440 - layer.4.output 0.00022000 0.11636817 - ------------------------------------------------------------------------------------- - TOTAL 0.00290784 0.92310911 - (elements=673,792) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 673792 -Total Bytes 173236 -BPFP 2.0568 bits/point -EBPFP 4.1137 equivalent bits/point -MSE 0.923109 ----------------------- -------------------------------------------------------- -Time: 2.720s Load: 0.004s, Pack+Encode: 1.567s, Decode+Unpack: 1.149s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 47, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.9231 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample1295-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst (9/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 154, 128) -Output shape: (1, 154, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.output: torch.Size([1, 154, 4096]) -> torch.Size([1, 1, 154, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,920B, BPFP=0.4525 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,692B, BPFP=2.5209 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 32,492B, BPFP=1.6483 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,780B, BPFP=2.6776 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 41,740B, BPFP=2.1175 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 54,024B, BPFP=2.7407 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,868B, BPFP=2.2254 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,792B, BPFP=2.7289 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 34,036B, BPFP=1.7267 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,580B, BPFP=2.7689 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 140,196B, BPFP=1.7781 -⌛️ [2/4] FRONTEND: Frontend time: 1.943s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.output: torch.Size([1, 154, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.553s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.output: torch.Size([1, 154, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02515895 7.43158672 - layer.0.v_cache 0.00000027 0.00014648 - layer.1.k_cache 0.00328501 0.85498423 - layer.1.v_cache 0.00000091 0.00051230 - layer.2.k_cache 0.00115449 0.37288418 - layer.2.v_cache 0.00000127 0.00076355 - layer.3.k_cache 0.00140159 0.43394188 - layer.3.v_cache 0.00000237 0.00130911 - layer.4.k_cache 0.00336527 0.80543736 - layer.4.v_cache 0.00000311 0.00207144 - layer.4.output 0.00016660 0.07309741 - ------------------------------------------------------------------------------------- - TOTAL 0.00250283 0.72828764 - (elements=2,207,744) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2207744 -Total Bytes 566120 -BPFP 2.0514 bits/point -EBPFP 4.1028 equivalent bits/point -MSE 0.728288 ----------------------- -------------------------------------------------------- -Time: 3.503s Load: 0.008s, Pack+Encode: 1.943s, Decode+Unpack: 1.553s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7283 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample13-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst (10/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 100, 128) -Output shape: (1, 100, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,032B, BPFP=0.4713 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,912B, BPFP=2.5713 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,320B, BPFP=1.6656 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,652B, BPFP=2.7072 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,004B, BPFP=2.1097 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,160B, BPFP=2.7469 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,500B, BPFP=2.3047 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,308B, BPFP=2.7584 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,128B, BPFP=1.7288 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,792B, BPFP=2.7963 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 85,156B, BPFP=1.6632 -⌛️ [2/4] FRONTEND: Frontend time: 1.804s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.333s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02583573 7.68077637 - layer.0.v_cache 0.00000027 0.00015463 - layer.1.k_cache 0.00330333 0.84144608 - layer.1.v_cache 0.00000088 0.00055278 - layer.2.k_cache 0.00114776 0.38825916 - layer.2.v_cache 0.00000114 0.00075239 - layer.3.k_cache 0.00139474 0.46346317 - layer.3.v_cache 0.00000229 0.00133057 - layer.4.k_cache 0.00328125 0.87706924 - layer.4.v_cache 0.00000316 0.00218956 - layer.4.output 0.00020624 0.08438413 - ------------------------------------------------------------------------------------- - TOTAL 0.00255682 0.75668075 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 364964 -BPFP 2.0366 bits/point -EBPFP 4.0733 equivalent bits/point -MSE 0.756681 ----------------------- -------------------------------------------------------- -Time: 3.144s Load: 0.007s, Pack+Encode: 1.804s, Decode+Unpack: 1.333s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7567 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample130-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst (11/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,956B, BPFP=0.4898 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,628B, BPFP=2.6832 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,220B, BPFP=1.5806 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,608B, BPFP=2.7638 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 25,140B, BPFP=2.0674 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,004B, BPFP=2.8786 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,436B, BPFP=2.3385 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,524B, BPFP=2.8391 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,412B, BPFP=1.5964 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,552B, BPFP=2.9237 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,676B, BPFP=1.3502 -⌛️ [2/4] FRONTEND: Frontend time: 1.815s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.370s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02533972 7.60480957 - layer.0.v_cache 0.00000026 0.00015380 - layer.1.k_cache 0.00345234 0.80187667 - layer.1.v_cache 0.00000076 0.00053938 - layer.2.k_cache 0.00126989 0.40421384 - layer.2.v_cache 0.00000106 0.00074186 - layer.3.k_cache 0.00140666 0.44070129 - layer.3.v_cache 0.00000216 0.00130816 - layer.4.k_cache 0.00351815 0.89199195 - layer.4.v_cache 0.00000309 0.00223898 - layer.4.output 0.00017514 0.08431551 - ------------------------------------------------------------------------------------- - TOTAL 0.00254962 0.74898840 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 335156 -BPFP 1.9687 bits/point -EBPFP 3.9375 equivalent bits/point -MSE 0.748988 ----------------------- -------------------------------------------------------- -Time: 3.192s Load: 0.007s, Pack+Encode: 1.815s, Decode+Unpack: 1.370s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7490 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample144-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst (12/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,628B, BPFP=0.4533 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,320B, BPFP=2.6031 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,368B, BPFP=1.5599 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,168B, BPFP=2.7519 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,860B, BPFP=2.0023 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,348B, BPFP=2.7664 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,848B, BPFP=2.2429 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,212B, BPFP=2.7555 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,400B, BPFP=1.5625 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,764B, BPFP=2.7999 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,608B, BPFP=1.2606 -⌛️ [2/4] FRONTEND: Frontend time: 1.790s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.342s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02570412 7.66097802 - layer.0.v_cache 0.00000027 0.00015693 - layer.1.k_cache 0.00344234 0.81102926 - layer.1.v_cache 0.00000085 0.00054536 - layer.2.k_cache 0.00116154 0.39217683 - layer.2.v_cache 0.00000108 0.00077632 - layer.3.k_cache 0.00150460 0.44211972 - layer.3.v_cache 0.00000231 0.00130018 - layer.4.k_cache 0.00334272 0.85720188 - layer.4.v_cache 0.00000310 0.00224942 - layer.4.output 0.00020164 0.08372456 - ------------------------------------------------------------------------------------- - TOTAL 0.00256925 0.75024516 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 329524 -BPFP 1.8957 bits/point -EBPFP 3.7915 equivalent bits/point -MSE 0.750245 ----------------------- -------------------------------------------------------- -Time: 3.138s Load: 0.007s, Pack+Encode: 1.790s, Decode+Unpack: 1.342s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7502 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample145-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst (13/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 98, 128) -Output shape: (1, 98, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,932B, BPFP=0.4729 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,356B, BPFP=2.5794 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,272B, BPFP=1.6161 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,896B, BPFP=2.7022 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,092B, BPFP=2.1598 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,840B, BPFP=2.7774 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,952B, BPFP=2.3080 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,864B, BPFP=2.7793 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,696B, BPFP=1.6499 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,348B, BPFP=2.8179 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 92,984B, BPFP=1.8532 -⌛️ [2/4] FRONTEND: Frontend time: 1.770s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.398s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02660144 7.87051703 - layer.0.v_cache 0.00000028 0.00015731 - layer.1.k_cache 0.00330389 0.76700249 - layer.1.v_cache 0.00000100 0.00056108 - layer.2.k_cache 0.00113731 0.36073731 - layer.2.v_cache 0.00000122 0.00081935 - layer.3.k_cache 0.00133429 0.42839525 - layer.3.v_cache 0.00000236 0.00136948 - layer.4.k_cache 0.00329518 0.86068290 - layer.4.v_cache 0.00000318 0.00223419 - layer.4.output 0.00020743 0.07111234 - ------------------------------------------------------------------------------------- - TOTAL 0.00260785 0.75549470 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 367232 -BPFP 2.0911 bits/point -EBPFP 4.1822 equivalent bits/point -MSE 0.755495 ----------------------- -------------------------------------------------------- -Time: 3.174s Load: 0.006s, Pack+Encode: 1.770s, Decode+Unpack: 1.398s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7555 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample146-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst (14/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 98, 128) -Output shape: (1, 98, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,936B, BPFP=0.4732 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,336B, BPFP=2.5778 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,012B, BPFP=1.5953 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,052B, BPFP=2.7146 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,584B, BPFP=2.1193 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,676B, BPFP=2.7643 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,012B, BPFP=2.3128 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,504B, BPFP=2.7506 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,888B, BPFP=1.6652 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,720B, BPFP=2.7679 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,888B, BPFP=1.4327 -⌛️ [2/4] FRONTEND: Frontend time: 1.819s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.343s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02655566 7.87375065 - layer.0.v_cache 0.00000028 0.00016388 - layer.1.k_cache 0.00343622 0.85355377 - layer.1.v_cache 0.00000088 0.00057309 - layer.2.k_cache 0.00122497 0.37710198 - layer.2.v_cache 0.00000114 0.00081781 - layer.3.k_cache 0.00135168 0.44371990 - layer.3.v_cache 0.00000224 0.00135496 - layer.4.k_cache 0.00342672 0.85515524 - layer.4.v_cache 0.00000300 0.00215763 - layer.4.output 0.00022158 0.07531862 - ------------------------------------------------------------------------------------- - TOTAL 0.00263494 0.76497310 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 344608 -BPFP 1.9623 bits/point -EBPFP 3.9246 equivalent bits/point -MSE 0.764973 ----------------------- -------------------------------------------------------- -Time: 3.169s Load: 0.007s, Pack+Encode: 1.819s, Decode+Unpack: 1.343s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7650 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample147-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst (15/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 100, 128) -Output shape: (1, 100, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,868B, BPFP=0.4584 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,760B, BPFP=2.5594 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,552B, BPFP=1.7619 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,740B, BPFP=2.7141 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,204B, BPFP=2.0472 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,356B, BPFP=2.7622 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,032B, BPFP=2.2681 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,880B, BPFP=2.7250 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,208B, BPFP=1.6569 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,316B, BPFP=2.7591 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,504B, BPFP=1.4161 -⌛️ [2/4] FRONTEND: Frontend time: 1.830s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.324s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02664840 7.42320068 - layer.0.v_cache 0.00000027 0.00015987 - layer.1.k_cache 0.00335783 0.82467392 - layer.1.v_cache 0.00000086 0.00054303 - layer.2.k_cache 0.00109031 0.38004215 - layer.2.v_cache 0.00000115 0.00078552 - layer.3.k_cache 0.00136642 0.44598465 - layer.3.v_cache 0.00000222 0.00131422 - layer.4.k_cache 0.00341075 0.85030212 - layer.4.v_cache 0.00000317 0.00218094 - layer.4.output 0.00031254 0.07828354 - ------------------------------------------------------------------------------------- - TOTAL 0.00265226 0.73159438 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 350420 -BPFP 1.9555 bits/point -EBPFP 3.9109 equivalent bits/point -MSE 0.731594 ----------------------- -------------------------------------------------------- -Time: 3.162s Load: 0.007s, Pack+Encode: 1.830s, Decode+Unpack: 1.324s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7316 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample150-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst (16/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,848B, BPFP=0.4710 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,516B, BPFP=2.6189 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,524B, BPFP=1.5725 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,336B, BPFP=2.7655 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,464B, BPFP=2.1314 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,552B, BPFP=2.7829 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,664B, BPFP=2.2281 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,340B, BPFP=2.7658 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,964B, BPFP=1.5274 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,128B, BPFP=2.8293 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,524B, BPFP=1.3798 -⌛️ [2/4] FRONTEND: Frontend time: 1.819s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02654818 8.08212312 - layer.0.v_cache 0.00000027 0.00016059 - layer.1.k_cache 0.00347710 0.79020832 - layer.1.v_cache 0.00000089 0.00060108 - layer.2.k_cache 0.00114081 0.39204175 - layer.2.v_cache 0.00000111 0.00080780 - layer.3.k_cache 0.00140715 0.45740336 - layer.3.v_cache 0.00000211 0.00133041 - layer.4.k_cache 0.00324570 0.84010771 - layer.4.v_cache 0.00000329 0.00231595 - layer.4.output 0.00023085 0.09059450 - ------------------------------------------------------------------------------------- - TOTAL 0.00262500 0.78067701 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 337860 -BPFP 1.9437 bits/point -EBPFP 3.8874 equivalent bits/point -MSE 0.780677 ----------------------- -------------------------------------------------------- -Time: 3.171s Load: 0.007s, Pack+Encode: 1.819s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7807 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample153-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst (17/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,412B, BPFP=0.4751 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,784B, BPFP=2.7900 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,072B, BPFP=1.4986 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,908B, BPFP=2.9765 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,028B, BPFP=1.8459 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,452B, BPFP=3.0242 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 24,668B, BPFP=2.1654 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 33,900B, BPFP=2.9758 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,608B, BPFP=1.5456 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,616B, BPFP=2.9508 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,420B, BPFP=1.3698 -⌛️ [2/4] FRONTEND: Frontend time: 1.794s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.340s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02697376 8.95066508 - layer.0.v_cache 0.00000027 0.00016344 - layer.1.k_cache 0.00348241 0.78175200 - layer.1.v_cache 0.00000086 0.00055386 - layer.2.k_cache 0.00112258 0.41331932 - layer.2.v_cache 0.00000113 0.00076818 - layer.3.k_cache 0.00136851 0.44265674 - layer.3.v_cache 0.00000223 0.00127979 - layer.4.k_cache 0.00328884 0.82321664 - layer.4.v_cache 0.00000300 0.00213629 - layer.4.output 0.00019035 0.08751845 - ------------------------------------------------------------------------------------- - TOTAL 0.00264321 0.84047037 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 315868 -BPFP 1.9805 bits/point -EBPFP 3.9610 equivalent bits/point -MSE 0.840470 ----------------------- -------------------------------------------------------- -Time: 3.140s Load: 0.006s, Pack+Encode: 1.794s, Decode+Unpack: 1.340s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8405 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample154-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst (18/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 104, 128) -Output shape: (1, 104, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,268B, BPFP=0.4709 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,528B, BPFP=2.5186 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,080B, BPFP=1.5835 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,464B, BPFP=2.6641 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,100B, BPFP=2.0358 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,760B, BPFP=2.6863 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,924B, BPFP=2.2479 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,316B, BPFP=2.6529 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,776B, BPFP=1.7109 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,820B, BPFP=2.6908 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 79,344B, BPFP=1.4901 -⌛️ [2/4] FRONTEND: Frontend time: 1.800s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.output: torch.Size([1, 104, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.352s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.output: torch.Size([1, 104, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02512770 7.92997272 - layer.0.v_cache 0.00000027 0.00015801 - layer.1.k_cache 0.00323274 0.70129959 - layer.1.v_cache 0.00000093 0.00058854 - layer.2.k_cache 0.00116098 0.37341389 - layer.2.v_cache 0.00000116 0.00082885 - layer.3.k_cache 0.00137668 0.43880477 - layer.3.v_cache 0.00000213 0.00132584 - layer.4.k_cache 0.00339989 0.84809025 - layer.4.v_cache 0.00000305 0.00223262 - layer.4.output 0.00020049 0.07605808 - ------------------------------------------------------------------------------------- - TOTAL 0.00250768 0.75721053 - (elements=1,490,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1490944 -Total Bytes 362380 -BPFP 1.9444 bits/point -EBPFP 3.8889 equivalent bits/point -MSE 0.757211 ----------------------- -------------------------------------------------------- -Time: 3.160s Load: 0.007s, Pack+Encode: 1.800s, Decode+Unpack: 1.352s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7572 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample157-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst (19/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 82, 128) -Output shape: (1, 82, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.output: torch.Size([1, 82, 4096]) -> torch.Size([1, 1, 82, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,072B, BPFP=0.4832 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,040B, BPFP=2.9573 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,868B, BPFP=1.5118 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 32,040B, BPFP=3.0526 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,236B, BPFP=1.8327 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 32,856B, BPFP=3.1303 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,396B, BPFP=2.0385 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 31,916B, BPFP=3.0408 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,444B, BPFP=1.4714 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 32,848B, BPFP=3.1296 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 53,748B, BPFP=1.2802 -⌛️ [2/4] FRONTEND: Frontend time: 1.778s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.330s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02805163 7.74641232 - layer.0.v_cache 0.00000027 0.00015539 - layer.1.k_cache 0.00355746 0.76440160 - layer.1.v_cache 0.00000072 0.00053359 - layer.2.k_cache 0.00112546 0.41996607 - layer.2.v_cache 0.00000104 0.00075051 - layer.3.k_cache 0.00143209 0.48012375 - layer.3.v_cache 0.00000205 0.00125027 - layer.4.k_cache 0.00323843 0.92340274 - layer.4.v_cache 0.00000284 0.00204770 - layer.4.output 0.00022045 0.09775038 - ------------------------------------------------------------------------------------- - TOTAL 0.00273527 0.76643182 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 291464 -BPFP 1.9835 bits/point -EBPFP 3.9670 equivalent bits/point -MSE 0.766432 ----------------------- -------------------------------------------------------- -Time: 3.113s Load: 0.005s, Pack+Encode: 1.778s, Decode+Unpack: 1.330s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7664 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample159-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst (20/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 109, 128) -Output shape: (1, 109, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,216B, BPFP=0.4455 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,336B, BPFP=2.3893 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,304B, BPFP=1.6703 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,504B, BPFP=2.5447 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,008B, BPFP=2.0791 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,096B, BPFP=2.5872 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,860B, BPFP=2.1402 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,584B, BPFP=2.5505 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 24,484B, BPFP=1.7549 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,412B, BPFP=2.6098 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 74,568B, BPFP=1.3362 -⌛️ [2/4] FRONTEND: Frontend time: 1.812s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.321s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02442908 7.29896531 - layer.0.v_cache 0.00000028 0.00015716 - layer.1.k_cache 0.00329736 0.76066582 - layer.1.v_cache 0.00000085 0.00053633 - layer.2.k_cache 0.00114487 0.36793571 - layer.2.v_cache 0.00000123 0.00078052 - layer.3.k_cache 0.00134342 0.41887994 - layer.3.v_cache 0.00000206 0.00125645 - layer.4.k_cache 0.00348010 0.91149167 - layer.4.v_cache 0.00000308 0.00211838 - layer.4.output 0.00017639 0.08128333 - ------------------------------------------------------------------------------------- - TOTAL 0.00245771 0.72056576 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 364372 -BPFP 1.8654 bits/point -EBPFP 3.7309 equivalent bits/point -MSE 0.720566 ----------------------- -------------------------------------------------------- -Time: 3.140s Load: 0.007s, Pack+Encode: 1.812s, Decode+Unpack: 1.321s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7206 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample165-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst (21/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,580B, BPFP=0.4589 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,172B, BPFP=2.6457 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,784B, BPFP=1.6270 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,168B, BPFP=2.8099 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,420B, BPFP=2.0082 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,692B, BPFP=2.8530 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,668B, BPFP=2.3576 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,708B, BPFP=2.8543 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,264B, BPFP=1.5020 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,264B, BPFP=2.9000 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,712B, BPFP=1.4949 -⌛️ [2/4] FRONTEND: Frontend time: 1.782s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.327s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02733303 8.01725175 - layer.0.v_cache 0.00000028 0.00015735 - layer.1.k_cache 0.00367247 0.82787419 - layer.1.v_cache 0.00000082 0.00057174 - layer.2.k_cache 0.00112740 0.38605732 - layer.2.v_cache 0.00000120 0.00085070 - layer.3.k_cache 0.00137809 0.43058532 - layer.3.v_cache 0.00000227 0.00134322 - layer.4.k_cache 0.00319271 0.82352215 - layer.4.v_cache 0.00000332 0.00225553 - layer.4.output 0.00022674 0.08213444 - ------------------------------------------------------------------------------------- - TOTAL 0.00268704 0.77278621 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 340432 -BPFP 1.9997 bits/point -EBPFP 3.9994 equivalent bits/point -MSE 0.772786 ----------------------- -------------------------------------------------------- -Time: 3.116s Load: 0.007s, Pack+Encode: 1.782s, Decode+Unpack: 1.327s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7728 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample167-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst (22/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,148B, BPFP=0.4663 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,152B, BPFP=2.5146 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,296B, BPFP=1.6153 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,128B, BPFP=2.6644 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 25,756B, BPFP=1.9536 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,516B, BPFP=2.6939 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,252B, BPFP=2.2188 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,044B, BPFP=2.6581 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,420B, BPFP=1.6247 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,708B, BPFP=2.7084 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,052B, BPFP=1.3852 -⌛️ [2/4] FRONTEND: Frontend time: 1.820s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.356s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03002706 8.00436964 - layer.0.v_cache 0.00000028 0.00015311 - layer.1.k_cache 0.00329593 0.76478888 - layer.1.v_cache 0.00000087 0.00054649 - layer.2.k_cache 0.00114743 0.38945715 - layer.2.v_cache 0.00000126 0.00076596 - layer.3.k_cache 0.00133553 0.42067274 - layer.3.v_cache 0.00000217 0.00130349 - layer.4.k_cache 0.00338515 0.84379889 - layer.4.v_cache 0.00000318 0.00217952 - layer.4.output 0.00017408 0.07173288 - ------------------------------------------------------------------------------------- - TOTAL 0.00284966 0.76535481 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 351472 -BPFP 1.9042 bits/point -EBPFP 3.8084 equivalent bits/point -MSE 0.765355 ----------------------- -------------------------------------------------------- -Time: 3.182s Load: 0.006s, Pack+Encode: 1.820s, Decode+Unpack: 1.356s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7654 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample168-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst (23/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,260B, BPFP=0.4748 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,416B, BPFP=2.5346 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,912B, BPFP=1.5862 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,968B, BPFP=2.6523 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,048B, BPFP=2.0516 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,460B, BPFP=2.6896 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,356B, BPFP=2.2266 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,316B, BPFP=2.6787 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,464B, BPFP=1.7039 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,876B, BPFP=2.7212 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 77,864B, BPFP=1.4765 -⌛️ [2/4] FRONTEND: Frontend time: 1.764s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.333s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02636531 7.48982846 - layer.0.v_cache 0.00000028 0.00015595 - layer.1.k_cache 0.00340973 0.84659080 - layer.1.v_cache 0.00000088 0.00054901 - layer.2.k_cache 0.00113950 0.37869396 - layer.2.v_cache 0.00000120 0.00079905 - layer.3.k_cache 0.00133761 0.43022696 - layer.3.v_cache 0.00000227 0.00130223 - layer.4.k_cache 0.00339299 0.83537841 - layer.4.v_cache 0.00000316 0.00219901 - layer.4.output 0.00023648 0.07900389 - ------------------------------------------------------------------------------------- - TOTAL 0.00261420 0.73583853 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 358940 -BPFP 1.9447 bits/point -EBPFP 3.8893 equivalent bits/point -MSE 0.735839 ----------------------- -------------------------------------------------------- -Time: 3.103s Load: 0.006s, Pack+Encode: 1.764s, Decode+Unpack: 1.333s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7358 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample174-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst (24/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,660B, BPFP=0.4806 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,040B, BPFP=2.8057 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,372B, BPFP=1.5601 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,956B, BPFP=2.9684 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 23,116B, BPFP=1.9630 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,140B, BPFP=2.9840 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,192B, BPFP=2.3091 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,224B, BPFP=2.9912 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,380B, BPFP=1.6457 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,624B, BPFP=3.0251 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,928B, BPFP=1.5270 -⌛️ [2/4] FRONTEND: Frontend time: 1.811s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.339s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02458628 7.29408928 - layer.0.v_cache 0.00000028 0.00015525 - layer.1.k_cache 0.00347829 0.74519224 - layer.1.v_cache 0.00000086 0.00055621 - layer.2.k_cache 0.00113727 0.39619218 - layer.2.v_cache 0.00000116 0.00075410 - layer.3.k_cache 0.00132770 0.42661994 - layer.3.v_cache 0.00000243 0.00129047 - layer.4.k_cache 0.00339865 0.85028317 - layer.4.v_cache 0.00000297 0.00203746 - layer.4.output 0.00018871 0.08125876 - ------------------------------------------------------------------------------------- - TOTAL 0.00247791 0.71730038 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 339632 -BPFP 2.0601 bits/point -EBPFP 4.1201 equivalent bits/point -MSE 0.717300 ----------------------- -------------------------------------------------------- -Time: 3.156s Load: 0.006s, Pack+Encode: 1.811s, Decode+Unpack: 1.339s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7173 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample182-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst (25/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,836B, BPFP=0.4749 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,524B, BPFP=2.6468 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,728B, BPFP=1.5241 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,452B, BPFP=2.8037 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 23,656B, BPFP=1.9251 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,936B, BPFP=2.8431 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,940B, BPFP=2.2738 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,248B, BPFP=2.8685 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,300B, BPFP=1.5706 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,372B, BPFP=2.8786 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 79,544B, BPFP=1.6183 -⌛️ [2/4] FRONTEND: Frontend time: 1.824s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.364s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02634023 7.86820857 - layer.0.v_cache 0.00000027 0.00015844 - layer.1.k_cache 0.00351230 0.86243844 - layer.1.v_cache 0.00000075 0.00053274 - layer.2.k_cache 0.00114531 0.42296418 - layer.2.v_cache 0.00000110 0.00075406 - layer.3.k_cache 0.00136477 0.43231281 - layer.3.v_cache 0.00000217 0.00132582 - layer.4.k_cache 0.00356451 0.89603917 - layer.4.v_cache 0.00000291 0.00208279 - layer.4.output 0.00018809 0.09168525 - ------------------------------------------------------------------------------------- - TOTAL 0.00262048 0.77525414 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 347536 -BPFP 2.0202 bits/point -EBPFP 4.0404 equivalent bits/point -MSE 0.775254 ----------------------- -------------------------------------------------------- -Time: 3.194s Load: 0.006s, Pack+Encode: 1.824s, Decode+Unpack: 1.364s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7753 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample185-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst (26/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,740B, BPFP=0.4822 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,972B, BPFP=2.7698 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,396B, BPFP=1.5454 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,844B, BPFP=2.9271 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,292B, BPFP=2.0407 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,388B, BPFP=2.9728 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,804B, BPFP=2.3357 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,948B, BPFP=2.9358 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,628B, BPFP=1.4808 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,724B, BPFP=3.0010 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 70,724B, BPFP=1.4853 -⌛️ [2/4] FRONTEND: Frontend time: 1.821s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.363s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02618607 7.49297111 - layer.0.v_cache 0.00000028 0.00015988 - layer.1.k_cache 0.00358459 0.80241558 - layer.1.v_cache 0.00000078 0.00055046 - layer.2.k_cache 0.00113483 0.38515899 - layer.2.v_cache 0.00000114 0.00079237 - layer.3.k_cache 0.00136506 0.43264914 - layer.3.v_cache 0.00000214 0.00131037 - layer.4.k_cache 0.00334622 0.83709233 - layer.4.v_cache 0.00000303 0.00213802 - layer.4.output 0.00022383 0.08551900 - ------------------------------------------------------------------------------------- - TOTAL 0.00260854 0.73552245 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 338460 -BPFP 2.0309 bits/point -EBPFP 4.0618 equivalent bits/point -MSE 0.735522 ----------------------- -------------------------------------------------------- -Time: 3.190s Load: 0.006s, Pack+Encode: 1.821s, Decode+Unpack: 1.363s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7355 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample191-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst (27/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,868B, BPFP=0.4775 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,316B, BPFP=2.6299 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,092B, BPFP=1.5537 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,356B, BPFP=2.7959 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,144B, BPFP=2.2090 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,432B, BPFP=2.8835 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,712B, BPFP=2.2552 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,548B, BPFP=2.8115 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,612B, BPFP=1.5960 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,216B, BPFP=2.8659 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 77,484B, BPFP=1.5764 -⌛️ [2/4] FRONTEND: Frontend time: 1.785s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02641018 7.72756004 - layer.0.v_cache 0.00000028 0.00015708 - layer.1.k_cache 0.00334712 0.74977279 - layer.1.v_cache 0.00000079 0.00055884 - layer.2.k_cache 0.00118939 0.36663179 - layer.2.v_cache 0.00000116 0.00081316 - layer.3.k_cache 0.00138481 0.42397734 - layer.3.v_cache 0.00000224 0.00134499 - layer.4.k_cache 0.00336247 0.82192389 - layer.4.v_cache 0.00000319 0.00228074 - layer.4.output 0.00019453 0.06508581 - ------------------------------------------------------------------------------------- - TOTAL 0.00260570 0.73966885 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 348780 -BPFP 2.0274 bits/point -EBPFP 4.0548 equivalent bits/point -MSE 0.739669 ----------------------- -------------------------------------------------------- -Time: 3.146s Load: 0.007s, Pack+Encode: 1.785s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7397 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample196-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst (28/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,848B, BPFP=0.4966 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,476B, BPFP=2.7578 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,728B, BPFP=1.5054 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,520B, BPFP=2.9314 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,088B, BPFP=2.0455 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,928B, BPFP=2.9660 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,016B, BPFP=2.2942 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,808B, BPFP=2.9558 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,792B, BPFP=1.5958 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,324B, BPFP=2.9997 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,076B, BPFP=1.4452 -⌛️ [2/4] FRONTEND: Frontend time: 1.795s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.341s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02542763 7.83760602 - layer.0.v_cache 0.00000028 0.00016257 - layer.1.k_cache 0.00330708 0.71279716 - layer.1.v_cache 0.00000075 0.00056765 - layer.2.k_cache 0.00113337 0.39755278 - layer.2.v_cache 0.00000121 0.00083787 - layer.3.k_cache 0.00137180 0.43961683 - layer.3.v_cache 0.00000212 0.00132514 - layer.4.k_cache 0.00335567 0.83810665 - layer.4.v_cache 0.00000315 0.00222540 - layer.4.output 0.00020664 0.07397255 - ------------------------------------------------------------------------------------- - TOTAL 0.00253069 0.75190631 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 333604 -BPFP 2.0235 bits/point -EBPFP 4.0470 equivalent bits/point -MSE 0.751906 ----------------------- -------------------------------------------------------- -Time: 3.142s Load: 0.006s, Pack+Encode: 1.795s, Decode+Unpack: 1.341s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7519 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample197-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst (29/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 178, 128) -Output shape: (1, 178, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,836B, BPFP=0.4317 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 49,856B, BPFP=2.1882 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 36,448B, BPFP=1.5997 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 52,440B, BPFP=2.3016 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 44,260B, BPFP=1.9426 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,244B, BPFP=2.3369 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,700B, BPFP=1.9619 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,012B, BPFP=2.3267 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 36,916B, BPFP=1.6203 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 53,896B, BPFP=2.3655 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 113,380B, BPFP=1.2441 -⌛️ [2/4] FRONTEND: Frontend time: 1.986s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.535s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02409235 7.22418007 - layer.0.v_cache 0.00000026 0.00014034 - layer.1.k_cache 0.00309449 0.81332106 - layer.1.v_cache 0.00000082 0.00049680 - layer.2.k_cache 0.00118451 0.36645264 - layer.2.v_cache 0.00000118 0.00070167 - layer.3.k_cache 0.00136933 0.44186950 - layer.3.v_cache 0.00000226 0.00124728 - layer.4.k_cache 0.00340187 0.84466270 - layer.4.v_cache 0.00000300 0.00194357 - layer.4.output 0.00017401 0.06772061 - ------------------------------------------------------------------------------------- - TOTAL 0.00241758 0.71184986 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 547988 -BPFP 1.7180 bits/point -EBPFP 3.4359 equivalent bits/point -MSE 0.711850 ----------------------- -------------------------------------------------------- -Time: 3.532s Load: 0.010s, Pack+Encode: 1.986s, Decode+Unpack: 1.535s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7118 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample2-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst (30/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,588B, BPFP=0.4595 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,972B, BPFP=2.6293 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,932B, BPFP=1.5569 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,724B, BPFP=2.7734 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,460B, BPFP=2.1760 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,608B, BPFP=2.8461 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,932B, BPFP=2.3793 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,132B, BPFP=2.8069 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,680B, BPFP=1.5362 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,592B, BPFP=2.8447 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,868B, BPFP=1.3542 -⌛️ [2/4] FRONTEND: Frontend time: 1.823s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.323s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02560628 7.33522307 - layer.0.v_cache 0.00000028 0.00015851 - layer.1.k_cache 0.00332260 0.77614923 - layer.1.v_cache 0.00000077 0.00055587 - layer.2.k_cache 0.00114542 0.39452675 - layer.2.v_cache 0.00000120 0.00080280 - layer.3.k_cache 0.00143781 0.45221919 - layer.3.v_cache 0.00000216 0.00128687 - layer.4.k_cache 0.00324414 0.89695924 - layer.4.v_cache 0.00000298 0.00210819 - layer.4.output 0.00027467 0.09941102 - ------------------------------------------------------------------------------------- - TOTAL 0.00256159 0.73268813 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 333488 -BPFP 1.9589 bits/point -EBPFP 3.9179 equivalent bits/point -MSE 0.732688 ----------------------- -------------------------------------------------------- -Time: 3.152s Load: 0.006s, Pack+Encode: 1.823s, Decode+Unpack: 1.323s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7327 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample201-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst (31/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,520B, BPFP=0.4588 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,712B, BPFP=2.7188 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,728B, BPFP=1.6396 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,228B, BPFP=2.8447 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,928B, BPFP=2.0718 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,924B, BPFP=2.9026 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,748B, BPFP=2.3893 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,420B, BPFP=2.8607 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,992B, BPFP=1.5785 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,184B, BPFP=2.9242 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 63,400B, BPFP=1.3173 -⌛️ [2/4] FRONTEND: Frontend time: 1.827s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.312s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02535593 7.26972669 - layer.0.v_cache 0.00000029 0.00015430 - layer.1.k_cache 0.00347189 0.75275957 - layer.1.v_cache 0.00000073 0.00051590 - layer.2.k_cache 0.00115370 0.38671506 - layer.2.v_cache 0.00000109 0.00076399 - layer.3.k_cache 0.00136415 0.42001891 - layer.3.v_cache 0.00000218 0.00125181 - layer.4.k_cache 0.00350239 0.81337081 - layer.4.v_cache 0.00000301 0.00214853 - layer.4.output 0.00017689 0.08254960 - ------------------------------------------------------------------------------------- - TOTAL 0.00254021 0.71268743 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 332784 -BPFP 1.9756 bits/point -EBPFP 3.9512 equivalent bits/point -MSE 0.712687 ----------------------- -------------------------------------------------------- -Time: 3.145s Load: 0.006s, Pack+Encode: 1.827s, Decode+Unpack: 1.312s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7127 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample212-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst (32/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 99, 128) -Output shape: (1, 99, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,684B, BPFP=0.4485 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,496B, BPFP=2.5644 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,116B, BPFP=1.5874 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,328B, BPFP=2.7090 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,312B, BPFP=2.0764 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,120B, BPFP=2.7715 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,056B, BPFP=2.2929 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,576B, BPFP=2.7285 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,972B, BPFP=1.5761 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,388B, BPFP=2.7926 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,848B, BPFP=1.3583 -⌛️ [2/4] FRONTEND: Frontend time: 1.821s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.235s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02440176 7.96586470 - layer.0.v_cache 0.00000028 0.00016033 - layer.1.k_cache 0.00326434 0.73818107 - layer.1.v_cache 0.00000081 0.00051778 - layer.2.k_cache 0.00115982 0.38637593 - layer.2.v_cache 0.00000131 0.00077538 - layer.3.k_cache 0.00139121 0.43483846 - layer.3.v_cache 0.00000205 0.00124898 - layer.4.k_cache 0.00335893 0.82131049 - layer.4.v_cache 0.00000299 0.00213590 - layer.4.output 0.00020438 0.08076260 - ------------------------------------------------------------------------------------- - TOTAL 0.00245722 0.76246139 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 341896 -BPFP 1.9272 bits/point -EBPFP 3.8543 equivalent bits/point -MSE 0.762461 ----------------------- -------------------------------------------------------- -Time: 3.064s Load: 0.008s, Pack+Encode: 1.821s, Decode+Unpack: 1.235s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7625 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample214-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst (33/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,024B, BPFP=0.4569 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,848B, BPFP=2.4915 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,488B, BPFP=1.6299 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,532B, BPFP=2.6192 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,324B, BPFP=1.9967 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,144B, BPFP=2.6657 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,764B, BPFP=2.2576 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,892B, BPFP=2.6465 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,704B, BPFP=1.7221 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,592B, BPFP=2.6996 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,920B, BPFP=1.3827 -⌛️ [2/4] FRONTEND: Frontend time: 1.709s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.210s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02603974 8.08642578 - layer.0.v_cache 0.00000026 0.00015473 - layer.1.k_cache 0.00325535 0.81244215 - layer.1.v_cache 0.00000093 0.00054893 - layer.2.k_cache 0.00112553 0.39048482 - layer.2.v_cache 0.00000114 0.00074569 - layer.3.k_cache 0.00141218 0.47024314 - layer.3.v_cache 0.00000204 0.00125684 - layer.4.k_cache 0.00324713 0.90317195 - layer.4.v_cache 0.00000302 0.00210541 - layer.4.output 0.00020093 0.09278028 - ------------------------------------------------------------------------------------- - TOTAL 0.00256365 0.78847861 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 352232 -BPFP 1.9083 bits/point -EBPFP 3.8167 equivalent bits/point -MSE 0.788479 ----------------------- -------------------------------------------------------- -Time: 2.926s Load: 0.006s, Pack+Encode: 1.709s, Decode+Unpack: 1.210s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7885 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample224-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst (34/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,248B, BPFP=0.5085 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,784B, BPFP=2.6680 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,316B, BPFP=1.5719 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,620B, BPFP=2.8174 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 25,072B, BPFP=2.0404 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,480B, BPFP=2.8874 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,324B, BPFP=2.2236 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,068B, BPFP=2.8538 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,744B, BPFP=1.6068 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,588B, BPFP=2.8962 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,636B, BPFP=1.4778 -⌛️ [2/4] FRONTEND: Frontend time: 1.805s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.380s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02591230 7.37385432 - layer.0.v_cache 0.00000028 0.00015571 - layer.1.k_cache 0.00321835 0.77699542 - layer.1.v_cache 0.00000081 0.00058792 - layer.2.k_cache 0.00115341 0.39443278 - layer.2.v_cache 0.00000115 0.00078501 - layer.3.k_cache 0.00138578 0.42282085 - layer.3.v_cache 0.00000239 0.00137633 - layer.4.k_cache 0.00338578 0.84484148 - layer.4.v_cache 0.00000322 0.00227719 - layer.4.output 0.00018605 0.07412261 - ------------------------------------------------------------------------------------- - TOTAL 0.00255769 0.72247268 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 343880 -BPFP 1.9989 bits/point -EBPFP 3.9979 equivalent bits/point -MSE 0.722473 ----------------------- -------------------------------------------------------- -Time: 3.191s Load: 0.006s, Pack+Encode: 1.805s, Decode+Unpack: 1.380s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7225 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample227-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst (35/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,720B, BPFP=0.5257 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,228B, BPFP=2.8702 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,520B, BPFP=1.5184 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,616B, BPFP=3.0897 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,088B, BPFP=1.9382 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,472B, BPFP=3.1684 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 22,964B, BPFP=2.1107 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 32,544B, BPFP=2.9912 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,876B, BPFP=1.5511 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,008B, BPFP=3.1257 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 77,512B, BPFP=1.7811 -⌛️ [2/4] FRONTEND: Frontend time: 1.815s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.371s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02725663 7.96004064 - layer.0.v_cache 0.00000027 0.00016603 - layer.1.k_cache 0.00328410 0.78055770 - layer.1.v_cache 0.00000085 0.00059561 - layer.2.k_cache 0.00114159 0.43096780 - layer.2.v_cache 0.00000125 0.00085538 - layer.3.k_cache 0.00133976 0.46441507 - layer.3.v_cache 0.00000266 0.00144615 - layer.4.k_cache 0.00332377 0.85366983 - layer.4.v_cache 0.00000336 0.00240054 - layer.4.output 0.00023926 0.07697583 - ------------------------------------------------------------------------------------- - TOTAL 0.00266509 0.77164415 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 326548 -BPFP 2.1438 bits/point -EBPFP 4.2877 equivalent bits/point -MSE 0.771644 ----------------------- -------------------------------------------------------- -Time: 3.191s Load: 0.005s, Pack+Encode: 1.815s, Decode+Unpack: 1.371s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7716 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample233-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst (36/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,560B, BPFP=0.4478 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,312B, BPFP=2.6024 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,016B, BPFP=1.5316 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,208B, BPFP=2.7552 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,068B, BPFP=2.0995 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,048B, BPFP=2.8228 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,816B, BPFP=2.3209 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,760B, BPFP=2.7996 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,576B, BPFP=1.6572 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,584B, BPFP=2.8660 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 66,976B, BPFP=1.3486 -⌛️ [2/4] FRONTEND: Frontend time: 1.778s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.353s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02553191 7.67870024 - layer.0.v_cache 0.00000029 0.00016699 - layer.1.k_cache 0.00322990 0.76485215 - layer.1.v_cache 0.00000089 0.00057245 - layer.2.k_cache 0.00117012 0.39262756 - layer.2.v_cache 0.00000108 0.00079435 - layer.3.k_cache 0.00135684 0.42665808 - layer.3.v_cache 0.00000209 0.00129653 - layer.4.k_cache 0.00331928 0.91516326 - layer.4.v_cache 0.00000299 0.00219191 - layer.4.output 0.00021425 0.08038381 - ------------------------------------------------------------------------------------- - TOTAL 0.00253374 0.75032563 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 338924 -BPFP 1.9498 bits/point -EBPFP 3.8996 equivalent bits/point -MSE 0.750326 ----------------------- -------------------------------------------------------- -Time: 3.138s Load: 0.006s, Pack+Encode: 1.778s, Decode+Unpack: 1.353s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7503 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample241-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst (37/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,916B, BPFP=0.4765 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,980B, BPFP=2.5757 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 19,292B, BPFP=1.5538 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,396B, BPFP=2.7703 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 25,164B, BPFP=2.0267 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,096B, BPFP=2.8267 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,228B, BPFP=2.1930 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,868B, BPFP=2.8083 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 19,288B, BPFP=1.5535 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,432B, BPFP=2.8537 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 95,684B, BPFP=1.9266 -⌛️ [2/4] FRONTEND: Frontend time: 1.823s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02408928 8.27055249 - layer.0.v_cache 0.00000028 0.00016815 - layer.1.k_cache 0.00330002 0.86506330 - layer.1.v_cache 0.00000090 0.00056395 - layer.2.k_cache 0.00113900 0.37575562 - layer.2.v_cache 0.00000130 0.00080966 - layer.3.k_cache 0.00136322 0.44616982 - layer.3.v_cache 0.00000251 0.00142064 - layer.4.k_cache 0.00342626 0.86993235 - layer.4.v_cache 0.00000319 0.00224171 - layer.4.output 0.00021399 0.07526391 - ------------------------------------------------------------------------------------- - TOTAL 0.00244156 0.79526667 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 364344 -BPFP 2.0961 bits/point -EBPFP 4.1921 equivalent bits/point -MSE 0.795267 ----------------------- -------------------------------------------------------- -Time: 3.183s Load: 0.006s, Pack+Encode: 1.823s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7953 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample250-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst (38/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 99, 128) -Output shape: (1, 99, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,508B, BPFP=0.4347 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,052B, BPFP=2.5294 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,712B, BPFP=1.6345 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,076B, BPFP=2.6891 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,468B, BPFP=1.9309 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,596B, BPFP=2.7301 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,756B, BPFP=2.2693 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,184B, BPFP=2.6976 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,856B, BPFP=1.6458 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,808B, BPFP=2.7468 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,904B, BPFP=1.3594 -⌛️ [2/4] FRONTEND: Frontend time: 1.802s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.343s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02490362 7.36836751 - layer.0.v_cache 0.00000029 0.00016026 - layer.1.k_cache 0.00335574 0.81208516 - layer.1.v_cache 0.00000081 0.00052112 - layer.2.k_cache 0.00112037 0.37502246 - layer.2.v_cache 0.00000110 0.00073210 - layer.3.k_cache 0.00138480 0.42530761 - layer.3.v_cache 0.00000194 0.00118594 - layer.4.k_cache 0.00346640 0.85957907 - layer.4.v_cache 0.00000296 0.00201440 - layer.4.output 0.00019387 0.06798984 - ------------------------------------------------------------------------------------- - TOTAL 0.00250096 0.72263821 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 338920 -BPFP 1.9104 bits/point -EBPFP 3.8208 equivalent bits/point -MSE 0.722638 ----------------------- -------------------------------------------------------- -Time: 3.150s Load: 0.006s, Pack+Encode: 1.802s, Decode+Unpack: 1.343s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7226 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample251-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst (39/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 86, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 86, 128) -Output shape: (1, 86, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.output: torch.Size([1, 86, 4096]) -> torch.Size([1, 1, 86, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,800B, BPFP=0.5269 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,076B, BPFP=2.9139 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,120B, BPFP=1.5552 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,148B, BPFP=3.0113 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,308B, BPFP=1.9357 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 33,460B, BPFP=3.0396 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 24,460B, BPFP=2.2220 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 33,972B, BPFP=3.0861 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,136B, BPFP=1.5567 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,852B, BPFP=3.0752 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,176B, BPFP=1.6392 -⌛️ [2/4] FRONTEND: Frontend time: 1.794s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.output: torch.Size([1, 86, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.374s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.output: torch.Size([1, 86, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02609516 7.67679117 - layer.0.v_cache 0.00000027 0.00015784 - layer.1.k_cache 0.00348465 0.84178339 - layer.1.v_cache 0.00000080 0.00056259 - layer.2.k_cache 0.00118276 0.40986660 - layer.2.v_cache 0.00000133 0.00088241 - layer.3.k_cache 0.00137135 0.45256681 - layer.3.v_cache 0.00000258 0.00139674 - layer.4.k_cache 0.00338397 0.87162213 - layer.4.v_cache 0.00000307 0.00221703 - layer.4.output 0.00019889 0.08762606 - ------------------------------------------------------------------------------------- - TOTAL 0.00259439 0.75773935 - (elements=1,232,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1232896 -Total Bytes 324508 -BPFP 2.1057 bits/point -EBPFP 4.2113 equivalent bits/point -MSE 0.757739 ----------------------- -------------------------------------------------------- -Time: 3.173s Load: 0.005s, Pack+Encode: 1.794s, Decode+Unpack: 1.374s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 86, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7577 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample257-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst (40/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,772B, BPFP=0.5067 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,544B, BPFP=2.7690 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,368B, BPFP=1.5246 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,644B, BPFP=3.0411 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,996B, BPFP=1.9308 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,252B, BPFP=3.0067 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 26,472B, BPFP=2.3237 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,936B, BPFP=3.0667 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,644B, BPFP=1.5488 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,496B, BPFP=3.0281 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 80,572B, BPFP=1.7682 -⌛️ [2/4] FRONTEND: Frontend time: 1.818s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.332s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02681794 8.29899477 - layer.0.v_cache 0.00000027 0.00016460 - layer.1.k_cache 0.00344681 0.80151624 - layer.1.v_cache 0.00000079 0.00058182 - layer.2.k_cache 0.00114729 0.43245200 - layer.2.v_cache 0.00000121 0.00084133 - layer.3.k_cache 0.00138721 0.44576161 - layer.3.v_cache 0.00000216 0.00138424 - layer.4.k_cache 0.00341441 0.85657724 - layer.4.v_cache 0.00000316 0.00237083 - layer.4.output 0.00022172 0.09207324 - ------------------------------------------------------------------------------------- - TOTAL 0.00265058 0.80063840 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 339696 -BPFP 2.1299 bits/point -EBPFP 4.2598 equivalent bits/point -MSE 0.800638 ----------------------- -------------------------------------------------------- -Time: 3.157s Load: 0.006s, Pack+Encode: 1.818s, Decode+Unpack: 1.332s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8006 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample258-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst (41/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,704B, BPFP=0.5007 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,852B, BPFP=2.8838 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,128B, BPFP=1.5913 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,820B, BPFP=3.0565 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 23,196B, BPFP=2.0362 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,176B, BPFP=3.0878 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 26,148B, BPFP=2.2953 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,972B, BPFP=3.0699 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,032B, BPFP=1.5829 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,448B, BPFP=3.1117 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,100B, BPFP=1.5603 -⌛️ [2/4] FRONTEND: Frontend time: 1.781s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.341s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02424761 7.76409432 - layer.0.v_cache 0.00000029 0.00016537 - layer.1.k_cache 0.00343373 0.76340339 - layer.1.v_cache 0.00000078 0.00056703 - layer.2.k_cache 0.00114421 0.42211305 - layer.2.v_cache 0.00000115 0.00079574 - layer.3.k_cache 0.00135741 0.42237340 - layer.3.v_cache 0.00000236 0.00134622 - layer.4.k_cache 0.00332665 0.82689847 - layer.4.v_cache 0.00000319 0.00223508 - layer.4.output 0.00018046 0.08004886 - ------------------------------------------------------------------------------------- - TOTAL 0.00244566 0.75172768 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 335576 -BPFP 2.1041 bits/point -EBPFP 4.2082 equivalent bits/point -MSE 0.751728 ----------------------- -------------------------------------------------------- -Time: 3.127s Load: 0.005s, Pack+Encode: 1.781s, Decode+Unpack: 1.341s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7517 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample263-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst (42/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,516B, BPFP=0.4684 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,016B, BPFP=2.7188 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,112B, BPFP=1.5380 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,992B, BPFP=2.8865 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 23,860B, BPFP=2.0262 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,680B, BPFP=2.9450 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,720B, BPFP=2.3539 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,096B, BPFP=2.8954 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,392B, BPFP=1.5618 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,016B, BPFP=2.9735 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 61,268B, BPFP=1.3007 -⌛️ [2/4] FRONTEND: Frontend time: 1.759s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.349s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02551763 7.79653466 - layer.0.v_cache 0.00000027 0.00015825 - layer.1.k_cache 0.00347722 0.73145128 - layer.1.v_cache 0.00000076 0.00051996 - layer.2.k_cache 0.00117662 0.39038559 - layer.2.v_cache 0.00000108 0.00075533 - layer.3.k_cache 0.00141157 0.42943556 - layer.3.v_cache 0.00000195 0.00120699 - layer.4.k_cache 0.00338827 0.83097193 - layer.4.v_cache 0.00000294 0.00215133 - layer.4.output 0.00021021 0.07992583 - ------------------------------------------------------------------------------------- - TOTAL 0.00255851 0.75023387 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 324668 -BPFP 1.9693 bits/point -EBPFP 3.9386 equivalent bits/point -MSE 0.750234 ----------------------- -------------------------------------------------------- -Time: 3.114s Load: 0.006s, Pack+Encode: 1.759s, Decode+Unpack: 1.349s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7502 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample274-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst (43/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 87, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 87, 128) -Output shape: (1, 87, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.output: torch.Size([1, 87, 4096]) -> torch.Size([1, 1, 87, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,644B, BPFP=0.5068 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,760B, BPFP=2.8520 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,308B, BPFP=1.5542 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,012B, BPFP=3.0542 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,452B, BPFP=1.9264 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,796B, BPFP=3.1246 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 23,440B, BPFP=2.1049 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,720B, BPFP=3.1178 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,128B, BPFP=1.5381 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,612B, BPFP=3.0183 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,144B, BPFP=1.4625 -⌛️ [2/4] FRONTEND: Frontend time: 1.808s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.output: torch.Size([1, 87, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.336s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.output: torch.Size([1, 87, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02591714 7.62411534 - layer.0.v_cache 0.00000026 0.00016309 - layer.1.k_cache 0.00374749 0.79540297 - layer.1.v_cache 0.00000077 0.00055619 - layer.2.k_cache 0.00116905 0.41417067 - layer.2.v_cache 0.00000114 0.00080135 - layer.3.k_cache 0.00135864 0.44303657 - layer.3.v_cache 0.00000233 0.00136232 - layer.4.k_cache 0.00339113 0.85779957 - layer.4.v_cache 0.00000304 0.00218134 - layer.4.output 0.00030438 0.08257654 - ------------------------------------------------------------------------------------- - TOTAL 0.00262918 0.74784968 - (elements=1,247,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1247232 -Total Bytes 319016 -BPFP 2.0462 bits/point -EBPFP 4.0925 equivalent bits/point -MSE 0.747850 ----------------------- -------------------------------------------------------- -Time: 3.151s Load: 0.007s, Pack+Encode: 1.808s, Decode+Unpack: 1.336s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 87, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7478 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample282-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst (44/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,540B, BPFP=0.4604 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,448B, BPFP=2.6137 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,916B, BPFP=1.5721 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,204B, BPFP=2.7596 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,820B, BPFP=2.0628 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,692B, BPFP=2.8833 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,596B, BPFP=2.3767 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,508B, BPFP=2.8680 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,404B, BPFP=1.5296 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,032B, BPFP=2.9116 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,380B, BPFP=1.3377 -⌛️ [2/4] FRONTEND: Frontend time: 1.804s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423150 7.70997003 - layer.0.v_cache 0.00000028 0.00016136 - layer.1.k_cache 0.00344140 0.79252146 - layer.1.v_cache 0.00000074 0.00052932 - layer.2.k_cache 0.00110852 0.37402798 - layer.2.v_cache 0.00000113 0.00076299 - layer.3.k_cache 0.00135732 0.42928201 - layer.3.v_cache 0.00000223 0.00128298 - layer.4.k_cache 0.00337875 0.83077906 - layer.4.v_cache 0.00000288 0.00200074 - layer.4.output 0.00018208 0.08137863 - ------------------------------------------------------------------------------------- - TOTAL 0.00244665 0.74763089 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 329540 -BPFP 1.9563 bits/point -EBPFP 3.9127 equivalent bits/point -MSE 0.747631 ----------------------- -------------------------------------------------------- -Time: 3.164s Load: 0.006s, Pack+Encode: 1.804s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7476 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample290-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst (45/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 149, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 149, 128) -Output shape: (1, 149, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.output: torch.Size([1, 149, 4096]) -> torch.Size([1, 1, 149, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,680B, BPFP=0.4551 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,172B, BPFP=2.6307 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 31,472B, BPFP=1.6502 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,224B, BPFP=2.7907 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 41,008B, BPFP=2.1502 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,256B, BPFP=2.7924 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 43,432B, BPFP=2.2773 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,136B, BPFP=2.7861 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 32,628B, BPFP=1.7108 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,156B, BPFP=2.8396 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 117,100B, BPFP=1.5350 -⌛️ [2/4] FRONTEND: Frontend time: 1.931s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.output: torch.Size([1, 149, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.546s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.output: torch.Size([1, 149, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02473939 7.06125062 - layer.0.v_cache 0.00000028 0.00015308 - layer.1.k_cache 0.00314236 0.81149517 - layer.1.v_cache 0.00000089 0.00053135 - layer.2.k_cache 0.00115430 0.38625996 - layer.2.v_cache 0.00000109 0.00072646 - layer.3.k_cache 0.00134690 0.40303221 - layer.3.v_cache 0.00000207 0.00120734 - layer.4.k_cache 0.00351049 0.75052873 - layer.4.v_cache 0.00000318 0.00212168 - layer.4.output 0.00015300 0.06073301 - ------------------------------------------------------------------------------------- - TOTAL 0.00246521 0.69001705 - (elements=2,136,064) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2136064 -Total Bytes 538264 -BPFP 2.0159 bits/point -EBPFP 4.0318 equivalent bits/point -MSE 0.690017 ----------------------- -------------------------------------------------------- -Time: 3.485s Load: 0.008s, Pack+Encode: 1.931s, Decode+Unpack: 1.546s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 149, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6900 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample3-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst (46/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,524B, BPFP=0.4640 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,124B, BPFP=2.6986 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,548B, BPFP=1.5581 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,424B, BPFP=2.8918 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,968B, BPFP=1.8454 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,792B, BPFP=2.9227 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,692B, BPFP=2.3263 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,508B, BPFP=2.8989 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,556B, BPFP=1.4748 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,068B, BPFP=2.9459 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 69,904B, BPFP=1.4681 -⌛️ [2/4] FRONTEND: Frontend time: 1.817s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.339s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02417885 7.35415288 - layer.0.v_cache 0.00000029 0.00016246 - layer.1.k_cache 0.00337012 0.73993264 - layer.1.v_cache 0.00000087 0.00055233 - layer.2.k_cache 0.00111931 0.40173451 - layer.2.v_cache 0.00000117 0.00076635 - layer.3.k_cache 0.00133665 0.41783437 - layer.3.v_cache 0.00000243 0.00128946 - layer.4.k_cache 0.00331438 0.80373104 - layer.4.v_cache 0.00000316 0.00214044 - layer.4.output 0.00021472 0.08164859 - ------------------------------------------------------------------------------------- - TOTAL 0.00244186 0.71777792 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 332108 -BPFP 1.9928 bits/point -EBPFP 3.9856 equivalent bits/point -MSE 0.717778 ----------------------- -------------------------------------------------------- -Time: 3.162s Load: 0.006s, Pack+Encode: 1.817s, Decode+Unpack: 1.339s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7178 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample307-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst (47/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 81, 128) -Output shape: (1, 81, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.output: torch.Size([1, 81, 4096]) -> torch.Size([1, 1, 81, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,300B, BPFP=0.5112 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 28,964B, BPFP=2.7936 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,056B, BPFP=1.5486 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,356B, BPFP=3.2172 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,712B, BPFP=1.9012 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 33,780B, BPFP=3.2581 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,644B, BPFP=2.0876 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 33,432B, BPFP=3.2245 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,660B, BPFP=1.5104 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,712B, BPFP=3.2515 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,424B, BPFP=1.5052 -⌛️ [2/4] FRONTEND: Frontend time: 1.799s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.309s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02598516 8.11177798 - layer.0.v_cache 0.00000026 0.00016029 - layer.1.k_cache 0.00346748 0.81631112 - layer.1.v_cache 0.00000092 0.00058574 - layer.2.k_cache 0.00113221 0.40347309 - layer.2.v_cache 0.00000143 0.00084451 - layer.3.k_cache 0.00133539 0.43579299 - layer.3.v_cache 0.00000229 0.00135482 - layer.4.k_cache 0.00342405 0.84739195 - layer.4.v_cache 0.00000310 0.00226128 - layer.4.output 0.00020696 0.06813214 - ------------------------------------------------------------------------------------- - TOTAL 0.00258429 0.77803445 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 304040 -BPFP 2.0946 bits/point -EBPFP 4.1893 equivalent bits/point -MSE 0.778034 ----------------------- -------------------------------------------------------- -Time: 3.114s Load: 0.005s, Pack+Encode: 1.799s, Decode+Unpack: 1.309s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7780 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample313-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst (48/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,592B, BPFP=0.4698 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,012B, BPFP=2.6892 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,960B, BPFP=1.5927 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,976B, BPFP=2.8542 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,096B, BPFP=2.0242 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,120B, BPFP=2.9503 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 28,644B, BPFP=2.4062 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,992B, BPFP=2.9395 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,280B, BPFP=1.5356 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,648B, BPFP=2.9946 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 63,068B, BPFP=1.3245 -⌛️ [2/4] FRONTEND: Frontend time: 1.826s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.342s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02450277 7.93115366 - layer.0.v_cache 0.00000028 0.00016545 - layer.1.k_cache 0.00336978 0.82981659 - layer.1.v_cache 0.00000078 0.00056892 - layer.2.k_cache 0.00117610 0.38595671 - layer.2.v_cache 0.00000111 0.00080124 - layer.3.k_cache 0.00133121 0.45169937 - layer.3.v_cache 0.00000213 0.00131579 - layer.4.k_cache 0.00346230 0.88444404 - layer.4.v_cache 0.00000315 0.00225724 - layer.4.output 0.00018648 0.07502203 - ------------------------------------------------------------------------------------- - TOTAL 0.00247111 0.77059051 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 330388 -BPFP 1.9825 bits/point -EBPFP 3.9649 equivalent bits/point -MSE 0.770591 ----------------------- -------------------------------------------------------- -Time: 3.175s Load: 0.007s, Pack+Encode: 1.826s, Decode+Unpack: 1.342s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7706 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample319-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst (49/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 90, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 90, 128) -Output shape: (1, 90, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.output: torch.Size([1, 90, 4096]) -> torch.Size([1, 1, 90, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,556B, BPFP=0.4823 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,632B, BPFP=2.8326 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,832B, BPFP=1.4611 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,116B, BPFP=2.9615 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,028B, BPFP=1.8253 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,268B, BPFP=2.9747 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 26,672B, BPFP=2.3153 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,160B, BPFP=3.0521 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,972B, BPFP=1.4733 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,752B, BPFP=3.0167 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 58,152B, BPFP=1.2620 -⌛️ [2/4] FRONTEND: Frontend time: 1.816s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.output: torch.Size([1, 90, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.output: torch.Size([1, 90, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02906363 8.13335368 - layer.0.v_cache 0.00000028 0.00016212 - layer.1.k_cache 0.00333766 0.68825183 - layer.1.v_cache 0.00000075 0.00052678 - layer.2.k_cache 0.00114903 0.42228118 - layer.2.v_cache 0.00000112 0.00074166 - layer.3.k_cache 0.00131913 0.44065501 - layer.3.v_cache 0.00000216 0.00125457 - layer.4.k_cache 0.00329656 0.80351851 - layer.4.v_cache 0.00000300 0.00208010 - layer.4.output 0.00024019 0.08799070 - ------------------------------------------------------------------------------------- - TOTAL 0.00279529 0.77462773 - (elements=1,290,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1290240 -Total Bytes 316140 -BPFP 1.9602 bits/point -EBPFP 3.9204 equivalent bits/point -MSE 0.774628 ----------------------- -------------------------------------------------------- -Time: 3.168s Load: 0.006s, Pack+Encode: 1.816s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 90, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7746 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample333-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst (50/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 119, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 119, 128) -Output shape: (1, 119, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.output: torch.Size([1, 119, 4096]) -> torch.Size([1, 1, 119, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,236B, BPFP=0.4751 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,912B, BPFP=2.2264 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 24,204B, BPFP=1.5890 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 36,120B, BPFP=2.3713 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 30,156B, BPFP=1.9798 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,620B, BPFP=2.4041 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,604B, BPFP=2.0092 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 36,144B, BPFP=2.3729 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,756B, BPFP=1.6909 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,840B, BPFP=2.4186 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 100,836B, BPFP=1.6550 -⌛️ [2/4] FRONTEND: Frontend time: 1.799s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.output: torch.Size([1, 119, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.343s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.output: torch.Size([1, 119, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02458241 7.60041360 - layer.0.v_cache 0.00000028 0.00015748 - layer.1.k_cache 0.00331476 0.84355933 - layer.1.v_cache 0.00000080 0.00057751 - layer.2.k_cache 0.00114230 0.36070934 - layer.2.v_cache 0.00000115 0.00083004 - layer.3.k_cache 0.00138709 0.44160461 - layer.3.v_cache 0.00000225 0.00138558 - layer.4.k_cache 0.00333808 0.85076032 - layer.4.v_cache 0.00000314 0.00226688 - layer.4.output 0.00020645 0.07215230 - ------------------------------------------------------------------------------------- - TOTAL 0.00247129 0.74220528 - (elements=1,705,984) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1705984 -Total Bytes 398428 -BPFP 1.8684 bits/point -EBPFP 3.7368 equivalent bits/point -MSE 0.742205 ----------------------- -------------------------------------------------------- -Time: 3.149s Load: 0.007s, Pack+Encode: 1.799s, Decode+Unpack: 1.343s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 119, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7422 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample36-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst (51/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 82, 128) -Output shape: (1, 82, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.output: torch.Size([1, 82, 4096]) -> torch.Size([1, 1, 82, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,048B, BPFP=0.4809 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 30,728B, BPFP=2.9276 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,856B, BPFP=1.5107 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 31,940B, BPFP=3.0431 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,772B, BPFP=1.7885 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 32,216B, BPFP=3.0694 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,876B, BPFP=1.9889 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 31,204B, BPFP=2.9729 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,684B, BPFP=1.4943 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 31,704B, BPFP=3.0206 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 56,704B, BPFP=1.3506 -⌛️ [2/4] FRONTEND: Frontend time: 1.792s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.342s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02610582 8.08852852 - layer.0.v_cache 0.00000027 0.00015975 - layer.1.k_cache 0.00344214 0.82464972 - layer.1.v_cache 0.00000078 0.00057357 - layer.2.k_cache 0.00114699 0.41631285 - layer.2.v_cache 0.00000119 0.00080693 - layer.3.k_cache 0.00140111 0.43111103 - layer.3.v_cache 0.00000200 0.00126643 - layer.4.k_cache 0.00332288 0.87647312 - layer.4.v_cache 0.00000299 0.00209167 - layer.4.output 0.00018752 0.08911669 - ------------------------------------------------------------------------------------- - TOTAL 0.00258402 0.78560288 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 290732 -BPFP 1.9785 bits/point -EBPFP 3.9570 equivalent bits/point -MSE 0.785603 ----------------------- -------------------------------------------------------- -Time: 3.140s Load: 0.006s, Pack+Encode: 1.792s, Decode+Unpack: 1.342s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7856 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample365-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst (52/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 124, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 124, 128) -Output shape: (1, 124, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.output: torch.Size([1, 124, 4096]) -> torch.Size([1, 1, 124, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,236B, BPFP=0.4559 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,488B, BPFP=2.1099 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,016B, BPFP=1.5761 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,580B, BPFP=2.2417 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,504B, BPFP=1.8589 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,984B, BPFP=2.2671 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,224B, BPFP=1.9042 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,548B, BPFP=2.2397 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,308B, BPFP=1.5945 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,080B, BPFP=2.2732 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 83,996B, BPFP=1.3230 -⌛️ [2/4] FRONTEND: Frontend time: 1.829s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.output: torch.Size([1, 124, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.352s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.output: torch.Size([1, 124, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02574280 7.54030388 - layer.0.v_cache 0.00000027 0.00015564 - layer.1.k_cache 0.00324291 0.91857221 - layer.1.v_cache 0.00000080 0.00052137 - layer.2.k_cache 0.00117169 0.38781009 - layer.2.v_cache 0.00000108 0.00073245 - layer.3.k_cache 0.00137525 0.46673510 - layer.3.v_cache 0.00000203 0.00115407 - layer.4.k_cache 0.00340243 0.87138545 - layer.4.v_cache 0.00000299 0.00200403 - layer.4.output 0.00019780 0.07925823 - ------------------------------------------------------------------------------------- - TOTAL 0.00255239 0.75045766 - (elements=1,777,664) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1777664 -Total Bytes 377964 -BPFP 1.7009 bits/point -EBPFP 3.4019 equivalent bits/point -MSE 0.750458 ----------------------- -------------------------------------------------------- -Time: 3.188s Load: 0.007s, Pack+Encode: 1.829s, Decode+Unpack: 1.352s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 124, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7505 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample38-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst (53/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,228B, BPFP=0.4805 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,644B, BPFP=2.9085 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,044B, BPFP=1.4746 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 31,756B, BPFP=2.9188 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,712B, BPFP=1.8118 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 33,536B, BPFP=3.0824 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 22,616B, BPFP=2.0787 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 31,564B, BPFP=2.9011 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,624B, BPFP=1.5279 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 32,020B, BPFP=2.9430 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 56,040B, BPFP=1.2877 -⌛️ [2/4] FRONTEND: Frontend time: 1.815s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.362s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02690007 8.36402013 - layer.0.v_cache 0.00000028 0.00017218 - layer.1.k_cache 0.00337091 0.77416911 - layer.1.v_cache 0.00000080 0.00052982 - layer.2.k_cache 0.00114287 0.41767946 - layer.2.v_cache 0.00000105 0.00074391 - layer.3.k_cache 0.00139664 0.45914042 - layer.3.v_cache 0.00000196 0.00121285 - layer.4.k_cache 0.00332559 0.90551327 - layer.4.v_cache 0.00000294 0.00213484 - layer.4.output 0.00025832 0.09638633 - ------------------------------------------------------------------------------------- - TOTAL 0.00265546 0.80791866 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 296784 -BPFP 1.9484 bits/point -EBPFP 3.8968 equivalent bits/point -MSE 0.807919 ----------------------- -------------------------------------------------------- -Time: 3.182s Load: 0.005s, Pack+Encode: 1.815s, Decode+Unpack: 1.362s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8079 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample388-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst (54/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 74, 128) -Output shape: (1, 74, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.output: torch.Size([1, 74, 4096]) -> torch.Size([1, 1, 74, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,920B, BPFP=0.5194 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,620B, BPFP=2.8104 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,596B, BPFP=1.5410 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,272B, BPFP=3.1959 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,088B, BPFP=1.9096 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,756B, BPFP=3.2470 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,092B, BPFP=2.0156 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,424B, BPFP=3.1064 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,468B, BPFP=1.5274 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 30,172B, BPFP=3.1854 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 58,736B, BPFP=1.5503 -⌛️ [2/4] FRONTEND: Frontend time: 1.843s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.359s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02896725 8.20242062 - layer.0.v_cache 0.00000027 0.00017174 - layer.1.k_cache 0.00342263 0.76554953 - layer.1.v_cache 0.00000079 0.00061307 - layer.2.k_cache 0.00118243 0.43329765 - layer.2.v_cache 0.00000111 0.00082733 - layer.3.k_cache 0.00141981 0.45367689 - layer.3.v_cache 0.00000233 0.00132478 - layer.4.k_cache 0.00326747 0.91119137 - layer.4.v_cache 0.00000299 0.00219339 - layer.4.output 0.00023783 0.08814671 - ------------------------------------------------------------------------------------- - TOTAL 0.00280131 0.79456094 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 277144 -BPFP 2.0899 bits/point -EBPFP 4.1799 equivalent bits/point -MSE 0.794561 ----------------------- -------------------------------------------------------- -Time: 3.208s Load: 0.006s, Pack+Encode: 1.843s, Decode+Unpack: 1.359s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7946 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample390-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst (55/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 126, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 126, 128) -Output shape: (1, 126, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.output: torch.Size([1, 126, 4096]) -> torch.Size([1, 1, 126, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,460B, BPFP=0.4625 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,964B, BPFP=2.1059 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,676B, BPFP=1.5920 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 36,176B, BPFP=2.2431 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 30,300B, BPFP=1.8787 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,320B, BPFP=2.2520 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,144B, BPFP=1.8690 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 36,188B, BPFP=2.2438 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 24,964B, BPFP=1.5479 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,620B, BPFP=2.2706 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 78,532B, BPFP=1.2173 -⌛️ [2/4] FRONTEND: Frontend time: 1.818s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.output: torch.Size([1, 126, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.output: torch.Size([1, 126, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02675629 7.31692844 - layer.0.v_cache 0.00000026 0.00014898 - layer.1.k_cache 0.00313964 0.99109438 - layer.1.v_cache 0.00000077 0.00050472 - layer.2.k_cache 0.00115655 0.39010844 - layer.2.v_cache 0.00000107 0.00070643 - layer.3.k_cache 0.00144242 0.49027319 - layer.3.v_cache 0.00000205 0.00117286 - layer.4.k_cache 0.00340494 0.94433570 - layer.4.v_cache 0.00000295 0.00198925 - layer.4.output 0.00023638 0.08623045 - ------------------------------------------------------------------------------------- - TOTAL 0.00263232 0.74872744 - (elements=1,806,336) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1806336 -Total Bytes 376344 -BPFP 1.6668 bits/point -EBPFP 3.3335 equivalent bits/point -MSE 0.748727 ----------------------- -------------------------------------------------------- -Time: 3.172s Load: 0.008s, Pack+Encode: 1.818s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 126, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7487 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample40-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst (56/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 76, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 76, 128) -Output shape: (1, 76, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.output: torch.Size([1, 76, 4096]) -> torch.Size([1, 1, 76, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,832B, BPFP=0.4967 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,628B, BPFP=2.7373 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,800B, BPFP=1.5214 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 29,680B, BPFP=3.0510 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,896B, BPFP=1.8396 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 29,564B, BPFP=3.0391 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,020B, BPFP=1.9552 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,184B, BPFP=2.7944 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,664B, BPFP=1.5074 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,836B, BPFP=2.9642 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 55,996B, BPFP=1.4390 -⌛️ [2/4] FRONTEND: Frontend time: 1.781s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.output: torch.Size([1, 76, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.344s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.output: torch.Size([1, 76, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02706527 8.36983932 - layer.0.v_cache 0.00000027 0.00016241 - layer.1.k_cache 0.00360181 0.72716211 - layer.1.v_cache 0.00000073 0.00052204 - layer.2.k_cache 0.00113352 0.43707075 - layer.2.v_cache 0.00000108 0.00077022 - layer.3.k_cache 0.00139474 0.48268976 - layer.3.v_cache 0.00000224 0.00129167 - layer.4.k_cache 0.00318908 0.92384218 - layer.4.v_cache 0.00000296 0.00218268 - layer.4.output 0.00022408 0.10163020 - ------------------------------------------------------------------------------------- - TOTAL 0.00266343 0.81086100 - (elements=1,089,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1089536 -Total Bytes 269100 -BPFP 1.9759 bits/point -EBPFP 3.9518 equivalent bits/point -MSE 0.810861 ----------------------- -------------------------------------------------------- -Time: 3.130s Load: 0.005s, Pack+Encode: 1.781s, Decode+Unpack: 1.344s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 76, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8109 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample412-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst (57/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,256B, BPFP=0.4831 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,284B, BPFP=2.8754 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,244B, BPFP=1.4930 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,892B, BPFP=3.1151 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,400B, BPFP=1.7831 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 33,632B, BPFP=3.0912 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,904B, BPFP=2.0132 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 32,976B, BPFP=3.0309 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,316B, BPFP=1.4996 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,356B, BPFP=3.0658 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,788B, BPFP=1.3968 -⌛️ [2/4] FRONTEND: Frontend time: 1.828s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.367s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461138 7.02491527 - layer.0.v_cache 0.00000028 0.00015270 - layer.1.k_cache 0.00340719 0.79077759 - layer.1.v_cache 0.00000073 0.00050201 - layer.2.k_cache 0.00115796 0.45861233 - layer.2.v_cache 0.00000109 0.00071606 - layer.3.k_cache 0.00135314 0.43767934 - layer.3.v_cache 0.00000196 0.00115482 - layer.4.k_cache 0.00328730 0.84404162 - layer.4.v_cache 0.00000286 0.00202052 - layer.4.output 0.00022769 0.08304235 - ------------------------------------------------------------------------------------- - TOTAL 0.00248105 0.70662440 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 305048 -BPFP 2.0027 bits/point -EBPFP 4.0054 equivalent bits/point -MSE 0.706624 ----------------------- -------------------------------------------------------- -Time: 3.200s Load: 0.006s, Pack+Encode: 1.828s, Decode+Unpack: 1.367s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7066 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample414-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst (58/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 134, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 134, 128) -Output shape: (1, 134, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.output: torch.Size([1, 134, 4096]) -> torch.Size([1, 1, 134, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,652B, BPFP=0.4461 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 47,508B, BPFP=2.7698 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 27,496B, BPFP=1.6031 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 49,216B, BPFP=2.8694 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 35,416B, BPFP=2.0648 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 50,180B, BPFP=2.9256 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 40,376B, BPFP=2.3540 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 50,220B, BPFP=2.9279 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 29,056B, BPFP=1.6940 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 50,948B, BPFP=2.9704 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 83,400B, BPFP=1.2156 -⌛️ [2/4] FRONTEND: Frontend time: 1.951s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.output: torch.Size([1, 134, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.553s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.output: torch.Size([1, 134, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02623714 7.61057475 - layer.0.v_cache 0.00000029 0.00016102 - layer.1.k_cache 0.00318200 0.78813029 - layer.1.v_cache 0.00000078 0.00048611 - layer.2.k_cache 0.00115661 0.39397003 - layer.2.v_cache 0.00000102 0.00066461 - layer.3.k_cache 0.00142778 0.45950545 - layer.3.v_cache 0.00000209 0.00118884 - layer.4.k_cache 0.00344207 0.85945585 - layer.4.v_cache 0.00000292 0.00198169 - layer.4.output 0.00019652 0.07865279 - ------------------------------------------------------------------------------------- - TOTAL 0.00258848 0.74505213 - (elements=1,921,024) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1921024 -Total Bytes 471468 -BPFP 1.9634 bits/point -EBPFP 3.9268 equivalent bits/point -MSE 0.745052 ----------------------- -------------------------------------------------------- -Time: 3.513s Load: 0.009s, Pack+Encode: 1.951s, Decode+Unpack: 1.553s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 134, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7451 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample42-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst (59/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 132, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 132, 128) -Output shape: (1, 132, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.output: torch.Size([1, 132, 4096]) -> torch.Size([1, 1, 132, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,600B, BPFP=0.4498 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 47,348B, BPFP=2.8023 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 28,000B, BPFP=1.6572 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 50,096B, BPFP=2.9650 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 35,160B, BPFP=2.0810 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 50,684B, BPFP=2.9998 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 39,428B, BPFP=2.3336 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 50,496B, BPFP=2.9886 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 28,828B, BPFP=1.7062 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 50,648B, BPFP=2.9976 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 100,788B, BPFP=1.4913 -⌛️ [2/4] FRONTEND: Frontend time: 1.948s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.output: torch.Size([1, 132, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.524s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.output: torch.Size([1, 132, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02542288 8.04690737 - layer.0.v_cache 0.00000027 0.00015267 - layer.1.k_cache 0.00323066 0.82069732 - layer.1.v_cache 0.00000090 0.00052786 - layer.2.k_cache 0.00115984 0.41854928 - layer.2.v_cache 0.00000123 0.00077718 - layer.3.k_cache 0.00138390 0.44272619 - layer.3.v_cache 0.00000215 0.00128777 - layer.4.k_cache 0.00364779 0.85984953 - layer.4.v_cache 0.00000297 0.00202960 - layer.4.output 0.00020533 0.08610114 - ------------------------------------------------------------------------------------- - TOTAL 0.00254814 0.78127924 - (elements=1,892,352) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1892352 -Total Bytes 489076 -BPFP 2.0676 bits/point -EBPFP 4.1352 equivalent bits/point -MSE 0.781279 ----------------------- -------------------------------------------------------- -Time: 3.480s Load: 0.008s, Pack+Encode: 1.948s, Decode+Unpack: 1.524s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 132, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7813 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample44-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst (60/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 77, 128) -Output shape: (1, 77, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.output: torch.Size([1, 77, 4096]) -> torch.Size([1, 1, 77, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,276B, BPFP=0.5353 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 27,152B, BPFP=2.7549 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,160B, BPFP=1.5381 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 29,856B, BPFP=3.0292 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,672B, BPFP=1.8945 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,620B, BPFP=3.1067 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,028B, BPFP=2.0321 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,140B, BPFP=2.9566 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,128B, BPFP=1.5349 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 30,208B, BPFP=3.0649 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,656B, BPFP=1.6400 -⌛️ [2/4] FRONTEND: Frontend time: 1.800s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02625101 8.22272759 - layer.0.v_cache 0.00000028 0.00017266 - layer.1.k_cache 0.00355533 0.81971998 - layer.1.v_cache 0.00000093 0.00058780 - layer.2.k_cache 0.00115642 0.42947769 - layer.2.v_cache 0.00000119 0.00084035 - layer.3.k_cache 0.00136722 0.45288988 - layer.3.v_cache 0.00000230 0.00139376 - layer.4.k_cache 0.00337541 0.80573600 - layer.4.v_cache 0.00000304 0.00228899 - layer.4.output 0.00023573 0.07780065 - ------------------------------------------------------------------------------------- - TOTAL 0.00261829 0.78907409 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 285896 -BPFP 2.0720 bits/point -EBPFP 4.1439 equivalent bits/point -MSE 0.789074 ----------------------- -------------------------------------------------------- -Time: 3.152s Load: 0.005s, Pack+Encode: 1.800s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7891 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample454-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst (61/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,620B, BPFP=0.4671 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,332B, BPFP=2.6872 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,724B, BPFP=1.5562 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,924B, BPFP=2.8195 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 21,868B, BPFP=1.8175 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,460B, BPFP=2.8640 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,656B, BPFP=2.2985 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,644B, BPFP=2.8793 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,924B, BPFP=1.4897 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,220B, BPFP=2.9272 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 66,440B, BPFP=1.3805 -⌛️ [2/4] FRONTEND: Frontend time: 1.822s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.338s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02258782 7.83646214 - layer.0.v_cache 0.00000029 0.00016654 - layer.1.k_cache 0.00339276 0.79993828 - layer.1.v_cache 0.00000074 0.00049052 - layer.2.k_cache 0.00116230 0.38601120 - layer.2.v_cache 0.00000115 0.00071336 - layer.3.k_cache 0.00135573 0.42345623 - layer.3.v_cache 0.00000225 0.00121847 - layer.4.k_cache 0.00337629 0.91560396 - layer.4.v_cache 0.00000290 0.00188611 - layer.4.output 0.00017695 0.07165140 - ------------------------------------------------------------------------------------- - TOTAL 0.00232786 0.76089660 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 328812 -BPFP 1.9520 bits/point -EBPFP 3.9040 equivalent bits/point -MSE 0.760897 ----------------------- -------------------------------------------------------- -Time: 3.166s Load: 0.007s, Pack+Encode: 1.822s, Decode+Unpack: 1.338s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7609 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample464-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst (62/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 77, 128) -Output shape: (1, 77, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.output: torch.Size([1, 77, 4096]) -> torch.Size([1, 1, 77, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,128B, BPFP=0.5203 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 27,400B, BPFP=2.7800 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,136B, BPFP=1.5357 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,348B, BPFP=3.0791 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,864B, BPFP=1.9140 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 31,060B, BPFP=3.1514 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,544B, BPFP=1.9830 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,484B, BPFP=2.9915 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,028B, BPFP=1.5248 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 30,432B, BPFP=3.0877 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 54,828B, BPFP=1.3907 -⌛️ [2/4] FRONTEND: Frontend time: 1.784s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.349s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02707502 8.52761246 - layer.0.v_cache 0.00000027 0.00017021 - layer.1.k_cache 0.00339772 0.84075353 - layer.1.v_cache 0.00000076 0.00058181 - layer.2.k_cache 0.00114556 0.44565889 - layer.2.v_cache 0.00000112 0.00080758 - layer.3.k_cache 0.00146787 0.47262212 - layer.3.v_cache 0.00000204 0.00133411 - layer.4.k_cache 0.00318992 0.90465635 - layer.4.v_cache 0.00000300 0.00226135 - layer.4.output 0.00024419 0.10953767 - ------------------------------------------------------------------------------------- - TOTAL 0.00266143 0.83104351 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 277252 -BPFP 2.0093 bits/point -EBPFP 4.0186 equivalent bits/point -MSE 0.831044 ----------------------- -------------------------------------------------------- -Time: 3.140s Load: 0.007s, Pack+Encode: 1.784s, Decode+Unpack: 1.349s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8310 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample478-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst (63/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 112, 128) -Output shape: (1, 112, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.output: torch.Size([1, 112, 4096]) -> torch.Size([1, 1, 112, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,532B, BPFP=0.4556 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,064B, BPFP=2.3064 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,660B, BPFP=1.5806 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,904B, BPFP=2.4347 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 28,312B, BPFP=1.9749 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,396B, BPFP=2.4690 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,052B, BPFP=2.0963 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,016B, BPFP=2.4425 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,644B, BPFP=1.6493 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,800B, BPFP=2.4972 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 72,228B, BPFP=1.2596 -⌛️ [2/4] FRONTEND: Frontend time: 1.809s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.334s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02677278 7.10968835 - layer.0.v_cache 0.00000026 0.00015629 - layer.1.k_cache 0.00327242 0.76502466 - layer.1.v_cache 0.00000072 0.00053293 - layer.2.k_cache 0.00117792 0.38558831 - layer.2.v_cache 0.00000108 0.00073918 - layer.3.k_cache 0.00137547 0.43609524 - layer.3.v_cache 0.00000205 0.00124347 - layer.4.k_cache 0.00331726 0.89762960 - layer.4.v_cache 0.00000297 0.00208339 - layer.4.output 0.00024922 0.09563027 - ------------------------------------------------------------------------------------- - TOTAL 0.00263713 0.71295018 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 357608 -BPFP 1.7818 bits/point -EBPFP 3.5635 equivalent bits/point -MSE 0.712950 ----------------------- -------------------------------------------------------- -Time: 3.150s Load: 0.007s, Pack+Encode: 1.809s, Decode+Unpack: 1.334s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7130 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample48-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst (64/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 78, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 78, 128) -Output shape: (1, 78, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.output: torch.Size([1, 78, 4096]) -> torch.Size([1, 1, 78, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,224B, BPFP=0.5232 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 28,428B, BPFP=2.8474 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,380B, BPFP=1.5405 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 31,680B, BPFP=3.1731 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,516B, BPFP=1.9547 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 32,468B, BPFP=3.2520 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,648B, BPFP=2.0681 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 32,692B, BPFP=3.2744 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,372B, BPFP=1.5397 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 32,564B, BPFP=3.2616 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,524B, BPFP=1.5656 -⌛️ [2/4] FRONTEND: Frontend time: 1.802s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.output: torch.Size([1, 78, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.360s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.output: torch.Size([1, 78, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02543071 7.42342748 - layer.0.v_cache 0.00000030 0.00016268 - layer.1.k_cache 0.00350441 0.75985062 - layer.1.v_cache 0.00000082 0.00058654 - layer.2.k_cache 0.00114279 0.40468519 - layer.2.v_cache 0.00000132 0.00087855 - layer.3.k_cache 0.00135780 0.45158132 - layer.3.v_cache 0.00000231 0.00140386 - layer.4.k_cache 0.00328533 0.80056303 - layer.4.v_cache 0.00000329 0.00240172 - layer.4.output 0.00022272 0.09112309 - ------------------------------------------------------------------------------------- - TOTAL 0.00254428 0.72928809 - (elements=1,118,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1118208 -Total Bytes 296496 -BPFP 2.1212 bits/point -EBPFP 4.2424 equivalent bits/point -MSE 0.729288 ----------------------- -------------------------------------------------------- -Time: 3.166s Load: 0.005s, Pack+Encode: 1.802s, Decode+Unpack: 1.360s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 78, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7293 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample485-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst (65/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 79, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 79, 128) -Output shape: (1, 79, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.output: torch.Size([1, 79, 4096]) -> torch.Size([1, 1, 79, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,912B, BPFP=0.4858 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 27,268B, BPFP=2.6966 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,440B, BPFP=1.5269 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 31,416B, BPFP=3.1068 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,676B, BPFP=1.8469 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,980B, BPFP=3.0637 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,076B, BPFP=2.0843 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 30,868B, BPFP=3.0526 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,220B, BPFP=1.5051 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 31,868B, BPFP=3.1515 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 50,904B, BPFP=1.2585 -⌛️ [2/4] FRONTEND: Frontend time: 1.763s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.output: torch.Size([1, 79, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.356s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.output: torch.Size([1, 79, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02536859 8.31915979 - layer.0.v_cache 0.00000029 0.00016742 - layer.1.k_cache 0.00331158 0.85197903 - layer.1.v_cache 0.00000077 0.00056287 - layer.2.k_cache 0.00112976 0.40438913 - layer.2.v_cache 0.00000111 0.00077298 - layer.3.k_cache 0.00133964 0.43809562 - layer.3.v_cache 0.00000223 0.00133178 - layer.4.k_cache 0.00332594 0.84798885 - layer.4.v_cache 0.00000299 0.00209177 - layer.4.output 0.00020400 0.07818903 - ------------------------------------------------------------------------------------- - TOTAL 0.00252135 0.79852110 - (elements=1,132,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1132544 -Total Bytes 278628 -BPFP 1.9682 bits/point -EBPFP 3.9363 equivalent bits/point -MSE 0.798521 ----------------------- -------------------------------------------------------- -Time: 3.124s Load: 0.006s, Pack+Encode: 1.763s, Decode+Unpack: 1.356s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 79, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7985 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample487-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst (66/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 81, 128) -Output shape: (1, 81, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.output: torch.Size([1, 81, 4096]) -> torch.Size([1, 1, 81, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,344B, BPFP=0.5154 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 29,724B, BPFP=2.8669 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,452B, BPFP=1.5868 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,768B, BPFP=3.2569 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,760B, BPFP=1.9059 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 33,628B, BPFP=3.2434 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,436B, BPFP=2.0675 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 32,864B, BPFP=3.1698 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,160B, BPFP=1.5586 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,684B, BPFP=3.2488 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,064B, BPFP=1.7618 -⌛️ [2/4] FRONTEND: Frontend time: 1.782s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.334s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02658518 8.01049503 - layer.0.v_cache 0.00000029 0.00017391 - layer.1.k_cache 0.00348817 0.78936829 - layer.1.v_cache 0.00000087 0.00059692 - layer.2.k_cache 0.00114702 0.41238695 - layer.2.v_cache 0.00000128 0.00089945 - layer.3.k_cache 0.00135614 0.44542256 - layer.3.v_cache 0.00000252 0.00144154 - layer.4.k_cache 0.00328397 0.83626170 - layer.4.v_cache 0.00000321 0.00233130 - layer.4.output 0.00022006 0.11070918 - ------------------------------------------------------------------------------------- - TOTAL 0.00262492 0.78158674 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 315884 -BPFP 2.1762 bits/point -EBPFP 4.3525 equivalent bits/point -MSE 0.781587 ----------------------- -------------------------------------------------------- -Time: 3.122s Load: 0.005s, Pack+Encode: 1.782s, Decode+Unpack: 1.334s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7816 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample495-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst (67/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 117, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 117, 128) -Output shape: (1, 117, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.output: torch.Size([1, 117, 4096]) -> torch.Size([1, 1, 117, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,796B, BPFP=0.4538 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,884B, BPFP=2.2626 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,832B, BPFP=1.5913 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,980B, BPFP=2.4025 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,972B, BPFP=2.0013 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,564B, BPFP=2.4415 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,644B, BPFP=2.0462 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 36,120B, BPFP=2.4119 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,780B, BPFP=1.7214 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,736B, BPFP=2.4530 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 79,192B, BPFP=1.3220 -⌛️ [2/4] FRONTEND: Frontend time: 1.814s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.output: torch.Size([1, 117, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.343s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.output: torch.Size([1, 117, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02657393 7.56335084 - layer.0.v_cache 0.00000026 0.00015149 - layer.1.k_cache 0.00326448 0.81412969 - layer.1.v_cache 0.00000079 0.00055041 - layer.2.k_cache 0.00117087 0.39000122 - layer.2.v_cache 0.00000122 0.00080356 - layer.3.k_cache 0.00141366 0.45786103 - layer.3.v_cache 0.00000241 0.00131448 - layer.4.k_cache 0.00339367 0.92668973 - layer.4.v_cache 0.00000311 0.00213289 - layer.4.output 0.00025967 0.09168077 - ------------------------------------------------------------------------------------- - TOTAL 0.00263308 0.75169346 - (elements=1,677,312) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1677312 -Total Bytes 375500 -BPFP 1.7910 bits/point -EBPFP 3.5819 equivalent bits/point -MSE 0.751693 ----------------------- -------------------------------------------------------- -Time: 3.165s Load: 0.008s, Pack+Encode: 1.814s, Decode+Unpack: 1.343s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 117, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7517 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample50-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst (68/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 84, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 84, 128) -Output shape: (1, 84, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.output: torch.Size([1, 84, 4096]) -> torch.Size([1, 1, 84, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 5,176B, BPFP=0.4814 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,556B, BPFP=2.9349 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,484B, BPFP=1.5331 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,356B, BPFP=3.1023 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,120B, BPFP=1.8713 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 32,908B, BPFP=3.0606 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 22,244B, BPFP=2.0688 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 31,664B, BPFP=2.9449 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,604B, BPFP=1.5443 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 33,788B, BPFP=3.1425 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,424B, BPFP=1.4049 -⌛️ [2/4] FRONTEND: Frontend time: 1.747s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.output: torch.Size([1, 84, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.348s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.output: torch.Size([1, 84, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02585370 8.10583351 - layer.0.v_cache 0.00000028 0.00016429 - layer.1.k_cache 0.00342466 0.76678894 - layer.1.v_cache 0.00000080 0.00055430 - layer.2.k_cache 0.00116354 0.41395923 - layer.2.v_cache 0.00000112 0.00076730 - layer.3.k_cache 0.00133515 0.43237109 - layer.3.v_cache 0.00000214 0.00127618 - layer.4.k_cache 0.00358854 0.85030374 - layer.4.v_cache 0.00000313 0.00213675 - layer.4.output 0.00016584 0.07699415 - ------------------------------------------------------------------------------------- - TOTAL 0.00257403 0.77729514 - (elements=1,204,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1204224 -Total Bytes 304324 -BPFP 2.0217 bits/point -EBPFP 4.0434 equivalent bits/point -MSE 0.777295 ----------------------- -------------------------------------------------------- -Time: 3.101s Load: 0.006s, Pack+Encode: 1.747s, Decode+Unpack: 1.348s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 84, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7773 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample516-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst (69/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 133, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 133, 128) -Output shape: (1, 133, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.output: torch.Size([1, 133, 4096]) -> torch.Size([1, 1, 133, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,620B, BPFP=0.4476 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 47,284B, BPFP=2.7775 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 27,900B, BPFP=1.6389 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 50,428B, BPFP=2.9622 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 36,684B, BPFP=2.1548 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 51,004B, BPFP=2.9960 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 37,880B, BPFP=2.2251 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 50,012B, BPFP=2.9377 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 28,920B, BPFP=1.6988 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 51,484B, BPFP=3.0242 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 96,184B, BPFP=1.4125 -⌛️ [2/4] FRONTEND: Frontend time: 1.972s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.output: torch.Size([1, 133, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.566s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.output: torch.Size([1, 133, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02254625 7.84427224 - layer.0.v_cache 0.00000028 0.00015359 - layer.1.k_cache 0.00324298 0.86218067 - layer.1.v_cache 0.00000087 0.00050435 - layer.2.k_cache 0.00114197 0.40294022 - layer.2.v_cache 0.00000110 0.00074432 - layer.3.k_cache 0.00134409 0.41136766 - layer.3.v_cache 0.00000211 0.00121722 - layer.4.k_cache 0.00375619 0.83209510 - layer.4.v_cache 0.00000306 0.00207942 - layer.4.output 0.00014515 0.05985183 - ------------------------------------------------------------------------------------- - TOTAL 0.00232996 0.75692586 - (elements=1,906,688) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1906688 -Total Bytes 485400 -BPFP 2.0366 bits/point -EBPFP 4.0732 equivalent bits/point -MSE 0.756926 ----------------------- -------------------------------------------------------- -Time: 3.546s Load: 0.008s, Pack+Encode: 1.972s, Decode+Unpack: 1.566s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 133, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7569 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample52-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst (70/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 125, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 125, 128) -Output shape: (1, 125, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.output: torch.Size([1, 125, 4096]) -> torch.Size([1, 1, 125, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,096B, BPFP=0.4435 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,292B, BPFP=2.0808 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 24,448B, BPFP=1.5280 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,520B, BPFP=2.2200 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,700B, BPFP=1.8562 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,972B, BPFP=2.2483 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,992B, BPFP=1.8745 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,756B, BPFP=2.2348 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,384B, BPFP=1.5865 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,048B, BPFP=2.2530 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 80,876B, BPFP=1.2637 -⌛️ [2/4] FRONTEND: Frontend time: 1.814s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.output: torch.Size([1, 125, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.362s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.output: torch.Size([1, 125, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02456201 7.36643555 - layer.0.v_cache 0.00000027 0.00014321 - layer.1.k_cache 0.00318185 0.86075311 - layer.1.v_cache 0.00000072 0.00046546 - layer.2.k_cache 0.00113729 0.38717206 - layer.2.v_cache 0.00000107 0.00066629 - layer.3.k_cache 0.00136705 0.46270981 - layer.3.v_cache 0.00000203 0.00114514 - layer.4.k_cache 0.00364986 0.89775220 - layer.4.v_cache 0.00000297 0.00183538 - layer.4.output 0.00019454 0.07508241 - ------------------------------------------------------------------------------------- - TOTAL 0.00247738 0.73424342 - (elements=1,792,000) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1792000 -Total Bytes 374084 -BPFP 1.6700 bits/point -EBPFP 3.3400 equivalent bits/point -MSE 0.734243 ----------------------- -------------------------------------------------------- -Time: 3.183s Load: 0.007s, Pack+Encode: 1.814s, Decode+Unpack: 1.362s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 125, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7342 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample53-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst (71/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 71, 128) -Output shape: (1, 71, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.output: torch.Size([1, 71, 4096]) -> torch.Size([1, 1, 71, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,732B, BPFP=0.5207 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,968B, BPFP=2.8574 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,396B, BPFP=1.5841 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,220B, BPFP=3.1052 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,032B, BPFP=1.8741 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 29,760B, BPFP=3.2746 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,020B, BPFP=1.9828 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 28,992B, BPFP=3.1901 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,088B, BPFP=1.5502 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,880B, BPFP=3.0678 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,684B, BPFP=1.6693 -⌛️ [2/4] FRONTEND: Frontend time: 1.812s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.350s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02849298 8.34100686 - layer.0.v_cache 0.00000028 0.00016991 - layer.1.k_cache 0.00376352 0.85139637 - layer.1.v_cache 0.00000080 0.00060098 - layer.2.k_cache 0.00117123 0.43509336 - layer.2.v_cache 0.00000118 0.00081214 - layer.3.k_cache 0.00138233 0.44647558 - layer.3.v_cache 0.00000223 0.00137290 - layer.4.k_cache 0.00337686 0.88129296 - layer.4.v_cache 0.00000300 0.00223816 - layer.4.output 0.00019204 0.07433429 - ------------------------------------------------------------------------------------- - TOTAL 0.00278304 0.80412831 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 269772 -BPFP 2.1203 bits/point -EBPFP 4.2406 equivalent bits/point -MSE 0.804128 ----------------------- -------------------------------------------------------- -Time: 3.167s Load: 0.005s, Pack+Encode: 1.812s, Decode+Unpack: 1.350s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8041 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample543-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst (72/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 68, 128) -Output shape: (1, 68, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.output: torch.Size([1, 68, 4096]) -> torch.Size([1, 1, 68, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,724B, BPFP=0.5427 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,520B, BPFP=2.5873 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,668B, BPFP=1.5703 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,372B, BPFP=3.1448 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,416B, BPFP=2.0009 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 26,232B, BPFP=3.0138 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,704B, BPFP=2.0340 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 25,768B, BPFP=2.9605 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,704B, BPFP=1.5744 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,140B, BPFP=3.1181 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,968B, BPFP=1.8086 -⌛️ [2/4] FRONTEND: Frontend time: 1.799s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.349s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02799314 8.72005238 - layer.0.v_cache 0.00000027 0.00016736 - layer.1.k_cache 0.00364027 0.77472771 - layer.1.v_cache 0.00000080 0.00061685 - layer.2.k_cache 0.00115812 0.43579475 - layer.2.v_cache 0.00000117 0.00088070 - layer.3.k_cache 0.00137295 0.46485357 - layer.3.v_cache 0.00000227 0.00148384 - layer.4.k_cache 0.00327663 0.86021479 - layer.4.v_cache 0.00000341 0.00254846 - layer.4.output 0.00018782 0.07360957 - ------------------------------------------------------------------------------------- - TOTAL 0.00272859 0.82541276 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 259216 -BPFP 2.1272 bits/point -EBPFP 4.2545 equivalent bits/point -MSE 0.825413 ----------------------- -------------------------------------------------------- -Time: 3.155s Load: 0.007s, Pack+Encode: 1.799s, Decode+Unpack: 1.349s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8254 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample561-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst (73/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 72, 128) -Output shape: (1, 72, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.output: torch.Size([1, 72, 4096]) -> torch.Size([1, 1, 72, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,712B, BPFP=0.5113 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 28,896B, BPFP=3.1354 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,280B, BPFP=1.5495 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,996B, BPFP=3.3633 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,700B, BPFP=1.9206 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,580B, BPFP=3.3181 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,508B, BPFP=2.0082 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,556B, BPFP=3.2070 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,344B, BPFP=1.5564 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 29,920B, BPFP=3.2465 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,588B, BPFP=1.4265 -⌛️ [2/4] FRONTEND: Frontend time: 1.791s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02695644 8.39369710 - layer.0.v_cache 0.00000028 0.00017220 - layer.1.k_cache 0.00362044 0.79513502 - layer.1.v_cache 0.00000088 0.00059647 - layer.2.k_cache 0.00116887 0.44408756 - layer.2.v_cache 0.00000122 0.00087755 - layer.3.k_cache 0.00138034 0.44428115 - layer.3.v_cache 0.00000238 0.00146166 - layer.4.k_cache 0.00335768 0.88343657 - layer.4.v_cache 0.00000294 0.00222349 - layer.4.output 0.00021991 0.08739786 - ------------------------------------------------------------------------------------- - TOTAL 0.00266936 0.80825430 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 272080 -BPFP 2.1088 bits/point -EBPFP 4.2175 equivalent bits/point -MSE 0.808254 ----------------------- -------------------------------------------------------- -Time: 3.150s Load: 0.005s, Pack+Encode: 1.791s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8083 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample570-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst (74/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,752B, BPFP=0.5086 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 30,300B, BPFP=3.2427 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,168B, BPFP=1.5163 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,080B, BPFP=3.2192 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,952B, BPFP=1.8142 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 31,232B, BPFP=3.3425 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,284B, BPFP=1.9568 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 30,424B, BPFP=3.2560 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,080B, BPFP=1.5068 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,644B, BPFP=3.0655 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 49,424B, BPFP=1.3223 -⌛️ [2/4] FRONTEND: Frontend time: 1.831s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.326s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02610407 8.46165905 - layer.0.v_cache 0.00000028 0.00016835 - layer.1.k_cache 0.00357364 0.76405962 - layer.1.v_cache 0.00000075 0.00056300 - layer.2.k_cache 0.00114681 0.42942332 - layer.2.v_cache 0.00000108 0.00078420 - layer.3.k_cache 0.00136272 0.44782142 - layer.3.v_cache 0.00000212 0.00126299 - layer.4.k_cache 0.00341161 0.88884975 - layer.4.v_cache 0.00000293 0.00215694 - layer.4.output 0.00018354 0.09569448 - ------------------------------------------------------------------------------------- - TOTAL 0.00259573 0.81282333 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 268340 -BPFP 2.0513 bits/point -EBPFP 4.1026 equivalent bits/point -MSE 0.812823 ----------------------- -------------------------------------------------------- -Time: 3.162s Load: 0.005s, Pack+Encode: 1.831s, Decode+Unpack: 1.326s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8128 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample581-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst (75/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 72, 128) -Output shape: (1, 72, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.output: torch.Size([1, 72, 4096]) -> torch.Size([1, 1, 72, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,568B, BPFP=0.4957 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 27,500B, BPFP=2.9839 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,348B, BPFP=1.5569 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,484B, BPFP=3.3077 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,972B, BPFP=1.8416 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,504B, BPFP=3.3099 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,276B, BPFP=1.9831 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 30,548B, BPFP=3.3147 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,856B, BPFP=1.5035 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 29,076B, BPFP=3.1549 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 50,152B, BPFP=1.3605 -⌛️ [2/4] FRONTEND: Frontend time: 1.776s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.315s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02727311 8.20708126 - layer.0.v_cache 0.00000029 0.00017741 - layer.1.k_cache 0.00335621 0.82119984 - layer.1.v_cache 0.00000072 0.00054278 - layer.2.k_cache 0.00111533 0.42067750 - layer.2.v_cache 0.00000112 0.00080135 - layer.3.k_cache 0.00138591 0.45846176 - layer.3.v_cache 0.00000206 0.00130904 - layer.4.k_cache 0.00323834 0.87947252 - layer.4.v_cache 0.00000293 0.00218925 - layer.4.output 0.00022510 0.08224686 - ------------------------------------------------------------------------------------- - TOTAL 0.00266260 0.79435001 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 266284 -BPFP 2.0638 bits/point -EBPFP 4.1277 equivalent bits/point -MSE 0.794350 ----------------------- -------------------------------------------------------- -Time: 3.096s Load: 0.005s, Pack+Encode: 1.776s, Decode+Unpack: 1.315s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7944 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample584-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst (76/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 115, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 115, 128) -Output shape: (1, 115, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.output: torch.Size([1, 115, 4096]) -> torch.Size([1, 1, 115, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,844B, BPFP=0.4649 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,812B, BPFP=2.2970 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 24,676B, BPFP=1.6764 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,900B, BPFP=2.4389 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,488B, BPFP=2.0033 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,172B, BPFP=2.4573 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,428B, BPFP=2.0671 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,668B, BPFP=2.4231 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,772B, BPFP=1.7508 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,404B, BPFP=2.4731 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 92,640B, BPFP=1.5734 -⌛️ [2/4] FRONTEND: Frontend time: 1.708s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.output: torch.Size([1, 115, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.211s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.output: torch.Size([1, 115, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02486876 7.58158861 - layer.0.v_cache 0.00000027 0.00015648 - layer.1.k_cache 0.00345256 0.75172955 - layer.1.v_cache 0.00000081 0.00055800 - layer.2.k_cache 0.00114556 0.36331960 - layer.2.v_cache 0.00000111 0.00076988 - layer.3.k_cache 0.00139466 0.42229024 - layer.3.v_cache 0.00000215 0.00131226 - layer.4.k_cache 0.00339940 0.84380400 - layer.4.v_cache 0.00000318 0.00231490 - layer.4.output 0.00018696 0.06970984 - ------------------------------------------------------------------------------------- - TOTAL 0.00250116 0.73190592 - (elements=1,648,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1648640 -Total Bytes 387804 -BPFP 1.8818 bits/point -EBPFP 3.7636 equivalent bits/point -MSE 0.731906 ----------------------- -------------------------------------------------------- -Time: 2.925s Load: 0.006s, Pack+Encode: 1.708s, Decode+Unpack: 1.211s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 115, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7319 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample60-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst (77/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 71, 128) -Output shape: (1, 71, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.output: torch.Size([1, 71, 4096]) -> torch.Size([1, 1, 71, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,932B, BPFP=0.5427 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,656B, BPFP=2.9331 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,212B, BPFP=1.5638 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,700B, BPFP=3.1580 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,292B, BPFP=1.9027 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 29,804B, BPFP=3.2795 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,192B, BPFP=2.0018 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,120B, BPFP=3.2042 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,068B, BPFP=1.5480 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,628B, BPFP=3.1501 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 54,712B, BPFP=1.5051 -⌛️ [2/4] FRONTEND: Frontend time: 1.711s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.362s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02663550 8.55804701 - layer.0.v_cache 0.00000027 0.00017154 - layer.1.k_cache 0.00355294 0.81368895 - layer.1.v_cache 0.00000089 0.00060822 - layer.2.k_cache 0.00117088 0.46686919 - layer.2.v_cache 0.00000116 0.00085014 - layer.3.k_cache 0.00139091 0.47051921 - layer.3.v_cache 0.00000218 0.00139973 - layer.4.k_cache 0.00325425 0.86177106 - layer.4.v_cache 0.00000306 0.00226345 - layer.4.output 0.00022008 0.09236529 - ------------------------------------------------------------------------------------- - TOTAL 0.00263517 0.82468926 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 266316 -BPFP 2.0932 bits/point -EBPFP 4.1863 equivalent bits/point -MSE 0.824689 ----------------------- -------------------------------------------------------- -Time: 3.078s Load: 0.005s, Pack+Encode: 1.711s, Decode+Unpack: 1.362s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8247 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample600-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst (78/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,744B, BPFP=0.5077 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 29,172B, BPFP=3.1220 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,672B, BPFP=1.5702 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,524B, BPFP=3.2667 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,784B, BPFP=1.9033 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 31,316B, BPFP=3.3515 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,472B, BPFP=1.9769 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 30,524B, BPFP=3.2667 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,148B, BPFP=1.5141 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 30,688B, BPFP=3.2842 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 56,064B, BPFP=1.5000 -⌛️ [2/4] FRONTEND: Frontend time: 1.811s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.343s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02450750 8.38720369 - layer.0.v_cache 0.00000028 0.00017066 - layer.1.k_cache 0.00347943 0.74071534 - layer.1.v_cache 0.00000079 0.00059758 - layer.2.k_cache 0.00114165 0.40722194 - layer.2.v_cache 0.00000118 0.00084857 - layer.3.k_cache 0.00136377 0.46819008 - layer.3.v_cache 0.00000212 0.00135931 - layer.4.k_cache 0.00321585 0.86432083 - layer.4.v_cache 0.00000299 0.00224727 - layer.4.output 0.00023923 0.09107652 - ------------------------------------------------------------------------------------- - TOTAL 0.00247661 0.80265581 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 278108 -BPFP 2.1259 bits/point -EBPFP 4.2519 equivalent bits/point -MSE 0.802656 ----------------------- -------------------------------------------------------- -Time: 3.158s Load: 0.005s, Pack+Encode: 1.811s, Decode+Unpack: 1.343s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8027 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample622-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst (79/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 112, 128) -Output shape: (1, 112, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.output: torch.Size([1, 112, 4096]) -> torch.Size([1, 1, 112, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,388B, BPFP=0.5153 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,704B, BPFP=2.3510 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,148B, BPFP=1.7542 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,924B, BPFP=2.5059 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,392B, BPFP=2.0502 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,232B, BPFP=2.5273 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,356B, BPFP=2.1175 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,912B, BPFP=2.5050 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,312B, BPFP=1.7656 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,396B, BPFP=2.5388 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 112,796B, BPFP=1.9670 -⌛️ [2/4] FRONTEND: Frontend time: 1.811s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02850304 7.67058236 - layer.0.v_cache 0.00000027 0.00015934 - layer.1.k_cache 0.00329348 0.84254783 - layer.1.v_cache 0.00000088 0.00061846 - layer.2.k_cache 0.00114370 0.36112751 - layer.2.v_cache 0.00000135 0.00081144 - layer.3.k_cache 0.00133332 0.42239404 - layer.3.v_cache 0.00000257 0.00141676 - layer.4.k_cache 0.00339413 0.80704594 - layer.4.v_cache 0.00000321 0.00220398 - layer.4.output 0.00019581 0.06178675 - ------------------------------------------------------------------------------------- - TOTAL 0.00274709 0.73971819 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 408560 -BPFP 2.0356 bits/point -EBPFP 4.0713 equivalent bits/point -MSE 0.739718 ----------------------- -------------------------------------------------------- -Time: 3.173s Load: 0.008s, Pack+Encode: 1.811s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7397 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample65-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst (80/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 68, 128) -Output shape: (1, 68, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.output: torch.Size([1, 68, 4096]) -> torch.Size([1, 1, 68, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,592B, BPFP=0.5276 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,800B, BPFP=2.6195 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,204B, BPFP=1.5170 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,640B, BPFP=2.8309 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,364B, BPFP=1.8801 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,264B, BPFP=2.9026 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,884B, BPFP=1.9398 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,992B, BPFP=2.8713 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,252B, BPFP=1.5225 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,600B, BPFP=2.9412 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 54,112B, BPFP=1.5542 -⌛️ [2/4] FRONTEND: Frontend time: 1.777s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02721363 8.30499447 - layer.0.v_cache 0.00000029 0.00016689 - layer.1.k_cache 0.00369932 0.78526839 - layer.1.v_cache 0.00000081 0.00059041 - layer.2.k_cache 0.00117793 0.40619486 - layer.2.v_cache 0.00000112 0.00080916 - layer.3.k_cache 0.00140028 0.46717212 - layer.3.v_cache 0.00000224 0.00137137 - layer.4.k_cache 0.00315349 0.88409884 - layer.4.v_cache 0.00000301 0.00222647 - layer.4.output 0.00021496 0.09223517 - ------------------------------------------------------------------------------------- - TOTAL 0.00267943 0.80155955 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 241704 -BPFP 1.9835 bits/point -EBPFP 3.9670 equivalent bits/point -MSE 0.801560 ----------------------- -------------------------------------------------------- -Time: 3.128s Load: 0.005s, Pack+Encode: 1.777s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8016 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample656-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst (81/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 136, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 136, 128) -Output shape: (1, 136, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.output: torch.Size([1, 136, 4096]) -> torch.Size([1, 1, 136, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,168B, BPFP=0.4692 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 48,696B, BPFP=2.7973 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 28,192B, BPFP=1.6195 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 50,764B, BPFP=2.9161 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 35,868B, BPFP=2.0604 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 51,600B, BPFP=2.9642 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 39,856B, BPFP=2.2895 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,344B, BPFP=2.9494 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 30,440B, BPFP=1.7486 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 52,040B, BPFP=2.9894 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 127,372B, BPFP=1.8292 -⌛️ [2/4] FRONTEND: Frontend time: 1.958s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.output: torch.Size([1, 136, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.587s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.output: torch.Size([1, 136, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02725438 7.56487140 - layer.0.v_cache 0.00000027 0.00015435 - layer.1.k_cache 0.00321745 0.84859859 - layer.1.v_cache 0.00000084 0.00052544 - layer.2.k_cache 0.00114365 0.39458348 - layer.2.v_cache 0.00000128 0.00074946 - layer.3.k_cache 0.00137918 0.42365228 - layer.3.v_cache 0.00000219 0.00126270 - layer.4.k_cache 0.00340721 0.87080058 - layer.4.v_cache 0.00000338 0.00214419 - layer.4.output 0.00019585 0.07381023 - ------------------------------------------------------------------------------------- - TOTAL 0.00265666 0.74304167 - (elements=1,949,696) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1949696 -Total Bytes 524340 -BPFP 2.1515 bits/point -EBPFP 4.3029 equivalent bits/point -MSE 0.743042 ----------------------- -------------------------------------------------------- -Time: 3.552s Load: 0.007s, Pack+Encode: 1.958s, Decode+Unpack: 1.587s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 136, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7430 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample66-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst (82/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 67, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 67, 128) -Output shape: (1, 67, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.output: torch.Size([1, 67, 4096]) -> torch.Size([1, 1, 67, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,364B, BPFP=0.5089 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,380B, BPFP=2.6096 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,360B, BPFP=1.5578 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 25,256B, BPFP=2.9450 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,744B, BPFP=1.9524 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,436B, BPFP=2.8493 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,272B, BPFP=2.0140 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,064B, BPFP=2.8060 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,160B, BPFP=1.5345 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,188B, BPFP=2.9370 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,588B, BPFP=1.8245 -⌛️ [2/4] FRONTEND: Frontend time: 1.784s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.output: torch.Size([1, 67, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.350s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.output: torch.Size([1, 67, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02769006 9.12912124 - layer.0.v_cache 0.00000028 0.00016929 - layer.1.k_cache 0.00368886 0.84113596 - layer.1.v_cache 0.00000112 0.00061707 - layer.2.k_cache 0.00116499 0.42672895 - layer.2.v_cache 0.00000120 0.00086413 - layer.3.k_cache 0.00139456 0.47420001 - layer.3.v_cache 0.00000232 0.00145356 - layer.4.k_cache 0.00321227 0.88387390 - layer.4.v_cache 0.00000307 0.00232484 - layer.4.output 0.00021890 0.08371989 - ------------------------------------------------------------------------------------- - TOTAL 0.00271674 0.86395489 - (elements=960,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 960512 -Total Bytes 248812 -BPFP 2.0723 bits/point -EBPFP 4.1447 equivalent bits/point -MSE 0.863955 ----------------------- -------------------------------------------------------- -Time: 3.139s Load: 0.004s, Pack+Encode: 1.784s, Decode+Unpack: 1.350s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 67, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8640 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample663-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst (83/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 123, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 123, 128) -Output shape: (1, 123, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.output: torch.Size([1, 123, 4096]) -> torch.Size([1, 1, 123, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,316B, BPFP=0.4647 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,492B, BPFP=2.1273 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 24,492B, BPFP=1.5556 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,368B, BPFP=2.2464 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,452B, BPFP=1.8707 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,188B, BPFP=2.2985 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,192B, BPFP=1.9177 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,780B, BPFP=2.2726 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,604B, BPFP=1.6263 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,644B, BPFP=2.3275 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 93,124B, BPFP=1.4787 -⌛️ [2/4] FRONTEND: Frontend time: 1.779s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.output: torch.Size([1, 123, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.382s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.output: torch.Size([1, 123, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02602563 7.86290114 - layer.0.v_cache 0.00000027 0.00015445 - layer.1.k_cache 0.00336899 0.85710572 - layer.1.v_cache 0.00000079 0.00051187 - layer.2.k_cache 0.00113470 0.40421221 - layer.2.v_cache 0.00000113 0.00073745 - layer.3.k_cache 0.00144113 0.47589815 - layer.3.v_cache 0.00000264 0.00134349 - layer.4.k_cache 0.00356668 0.90440790 - layer.4.v_cache 0.00000295 0.00203917 - layer.4.output 0.00019508 0.08742899 - ------------------------------------------------------------------------------------- - TOTAL 0.00259466 0.77564482 - (elements=1,763,328) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1763328 -Total Bytes 387652 -BPFP 1.7587 bits/point -EBPFP 3.5175 equivalent bits/point -MSE 0.775645 ----------------------- -------------------------------------------------------- -Time: 3.168s Load: 0.007s, Pack+Encode: 1.779s, Decode+Unpack: 1.382s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 123, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7756 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample68-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst (84/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.010s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 157, 128) -Output shape: (1, 157, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 9,080B, BPFP=0.4518 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 50,068B, BPFP=2.4914 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 33,060B, BPFP=1.6451 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 53,004B, BPFP=2.6375 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 42,056B, BPFP=2.0928 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 53,380B, BPFP=2.6562 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 44,084B, BPFP=2.1937 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 53,312B, BPFP=2.6529 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 35,304B, BPFP=1.7568 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 54,164B, BPFP=2.6953 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 130,176B, BPFP=1.6194 -⌛️ [2/4] FRONTEND: Frontend time: 1.968s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.551s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423319 7.55412147 - layer.0.v_cache 0.00000027 0.00014815 - layer.1.k_cache 0.00309445 0.86581878 - layer.1.v_cache 0.00000088 0.00052379 - layer.2.k_cache 0.00117908 0.36231246 - layer.2.v_cache 0.00000110 0.00073047 - layer.3.k_cache 0.00138553 0.41470181 - layer.3.v_cache 0.00000215 0.00124597 - layer.4.k_cache 0.00349636 0.79841653 - layer.4.v_cache 0.00000310 0.00204090 - layer.4.output 0.00020206 0.07055437 - ------------------------------------------------------------------------------------- - TOTAL 0.00244317 0.73444841 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 557688 -BPFP 1.9822 bits/point -EBPFP 3.9645 equivalent bits/point -MSE 0.734448 ----------------------- -------------------------------------------------------- -Time: 3.529s Load: 0.010s, Pack+Encode: 1.968s, Decode+Unpack: 1.551s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7344 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample7-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst (85/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 75, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 75, 128) -Output shape: (1, 75, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.output: torch.Size([1, 75, 4096]) -> torch.Size([1, 1, 75, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,976B, BPFP=0.5183 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 30,512B, BPFP=3.1783 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,872B, BPFP=1.5492 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,572B, BPFP=3.1846 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,164B, BPFP=1.8921 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 30,496B, BPFP=3.1767 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,872B, BPFP=1.9658 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,308B, BPFP=3.0529 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,772B, BPFP=1.5388 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 29,996B, BPFP=3.1246 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,936B, BPFP=1.6910 -⌛️ [2/4] FRONTEND: Frontend time: 1.768s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.output: torch.Size([1, 75, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.346s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.output: torch.Size([1, 75, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02489581 8.29596761 - layer.0.v_cache 0.00000028 0.00017049 - layer.1.k_cache 0.00341811 0.79769282 - layer.1.v_cache 0.00000080 0.00057991 - layer.2.k_cache 0.00110890 0.42293498 - layer.2.v_cache 0.00000112 0.00077378 - layer.3.k_cache 0.00134674 0.46222407 - layer.3.v_cache 0.00000209 0.00127744 - layer.4.k_cache 0.00330252 0.85403239 - layer.4.v_cache 0.00000296 0.00216648 - layer.4.output 0.00020017 0.07409002 - ------------------------------------------------------------------------------------- - TOTAL 0.00249143 0.79529857 - (elements=1,075,200) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1075200 -Total Bytes 287476 -BPFP 2.1390 bits/point -EBPFP 4.2779 equivalent bits/point -MSE 0.795299 ----------------------- -------------------------------------------------------- -Time: 3.118s Load: 0.005s, Pack+Encode: 1.768s, Decode+Unpack: 1.346s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 75, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7953 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample736-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst (86/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 122, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 122, 128) -Output shape: (1, 122, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.output: torch.Size([1, 122, 4096]) -> torch.Size([1, 1, 122, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 7,284B, BPFP=0.4664 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,632B, BPFP=2.1537 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,232B, BPFP=1.6158 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,840B, BPFP=2.2951 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,896B, BPFP=1.9144 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,384B, BPFP=2.3299 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,468B, BPFP=1.9511 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,920B, BPFP=2.3002 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 26,236B, BPFP=1.6801 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,672B, BPFP=2.3484 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 96,768B, BPFP=1.5492 -⌛️ [2/4] FRONTEND: Frontend time: 1.791s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.output: torch.Size([1, 122, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.326s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.output: torch.Size([1, 122, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02525679 7.19835225 - layer.0.v_cache 0.00000028 0.00015895 - layer.1.k_cache 0.00316206 0.86534006 - layer.1.v_cache 0.00000097 0.00055132 - layer.2.k_cache 0.00113900 0.37945582 - layer.2.v_cache 0.00000117 0.00076319 - layer.3.k_cache 0.00136642 0.45453228 - layer.3.v_cache 0.00000221 0.00125887 - layer.4.k_cache 0.00339647 0.85739286 - layer.4.v_cache 0.00000316 0.00207498 - layer.4.output 0.00022994 0.07299730 - ------------------------------------------------------------------------------------- - TOTAL 0.00251773 0.71799070 - (elements=1,748,992) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1748992 -Total Bytes 394332 -BPFP 1.8037 bits/point -EBPFP 3.6074 equivalent bits/point -MSE 0.717991 ----------------------- -------------------------------------------------------- -Time: 3.125s Load: 0.008s, Pack+Encode: 1.791s, Decode+Unpack: 1.326s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 122, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7180 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample74-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst (87/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 142, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 142, 128) -Output shape: (1, 142, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.output: torch.Size([1, 142, 4096]) -> torch.Size([1, 1, 142, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 8,356B, BPFP=0.4597 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 48,272B, BPFP=2.6558 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 30,684B, BPFP=1.6882 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 51,660B, BPFP=2.8422 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 38,364B, BPFP=2.1107 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 51,708B, BPFP=2.8449 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 41,848B, BPFP=2.3024 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 51,688B, BPFP=2.8438 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 31,060B, BPFP=1.7088 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 52,256B, BPFP=2.8750 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 143,408B, BPFP=1.9725 -⌛️ [2/4] FRONTEND: Frontend time: 1.963s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.output: torch.Size([1, 142, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.579s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.output: torch.Size([1, 142, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02301751 7.82861758 - layer.0.v_cache 0.00000029 0.00015251 - layer.1.k_cache 0.00315333 0.81685542 - layer.1.v_cache 0.00000086 0.00051345 - layer.2.k_cache 0.00112591 0.39029557 - layer.2.v_cache 0.00000130 0.00071678 - layer.3.k_cache 0.00130425 0.41540216 - layer.3.v_cache 0.00000237 0.00130992 - layer.4.k_cache 0.00348660 0.77365220 - layer.4.v_cache 0.00000314 0.00199670 - layer.4.output 0.00018111 0.07032462 - ------------------------------------------------------------------------------------- - TOTAL 0.00234429 0.75077220 - (elements=2,035,712) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2035712 -Total Bytes 549304 -BPFP 2.1587 bits/point -EBPFP 4.3173 equivalent bits/point -MSE 0.750772 ----------------------- -------------------------------------------------------- -Time: 3.551s Load: 0.009s, Pack+Encode: 1.963s, Decode+Unpack: 1.579s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 142, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7508 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample75-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst (88/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 74, 128) -Output shape: (1, 74, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.output: torch.Size([1, 74, 4096]) -> torch.Size([1, 1, 74, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,644B, BPFP=0.4903 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 29,780B, BPFP=3.1440 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,356B, BPFP=1.5156 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,636B, BPFP=3.2344 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,668B, BPFP=1.8653 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 31,024B, BPFP=3.2753 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,888B, BPFP=1.9941 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 29,208B, BPFP=3.0836 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,288B, BPFP=1.5084 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 31,336B, BPFP=3.3083 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 49,540B, BPFP=1.3075 -⌛️ [2/4] FRONTEND: Frontend time: 1.818s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02555943 8.27109342 - layer.0.v_cache 0.00000028 0.00016906 - layer.1.k_cache 0.00343224 0.72595648 - layer.1.v_cache 0.00000079 0.00057420 - layer.2.k_cache 0.00111603 0.41125756 - layer.2.v_cache 0.00000116 0.00080216 - layer.3.k_cache 0.00133047 0.44924566 - layer.3.v_cache 0.00000218 0.00128812 - layer.4.k_cache 0.00334566 0.87352691 - layer.4.v_cache 0.00000301 0.00215589 - layer.4.output 0.00024685 0.08302176 - ------------------------------------------------------------------------------------- - TOTAL 0.00255562 0.79058261 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 271368 -BPFP 2.0464 bits/point -EBPFP 4.0928 equivalent bits/point -MSE 0.790583 ----------------------- -------------------------------------------------------- -Time: 3.178s Load: 0.005s, Pack+Encode: 1.818s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7906 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample777-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst (89/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 61, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 61, 128) -Output shape: (1, 61, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.output: torch.Size([1, 61, 4096]) -> torch.Size([1, 1, 61, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,356B, BPFP=0.5579 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,404B, BPFP=2.2290 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,876B, BPFP=1.7772 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 18,816B, BPFP=2.4098 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,340B, BPFP=1.8366 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 18,992B, BPFP=2.4324 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,404B, BPFP=1.9728 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 18,796B, BPFP=2.4073 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,164B, BPFP=1.6860 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 18,984B, BPFP=2.4314 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 44,816B, BPFP=1.4349 -⌛️ [2/4] FRONTEND: Frontend time: 1.575s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.output: torch.Size([1, 61, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.150s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.output: torch.Size([1, 61, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02755715 8.64161957 - layer.0.v_cache 0.00000026 0.00015041 - layer.1.k_cache 0.00385674 1.03858910 - layer.1.v_cache 0.00000079 0.00053862 - layer.2.k_cache 0.00117960 0.44212685 - layer.2.v_cache 0.00000113 0.00079020 - layer.3.k_cache 0.00144752 0.54695217 - layer.3.v_cache 0.00000212 0.00123342 - layer.4.k_cache 0.00316257 1.04350531 - layer.4.v_cache 0.00000306 0.00202632 - layer.4.output 0.00025867 0.11460597 - ------------------------------------------------------------------------------------- - TOTAL 0.00273183 0.86971113 - (elements=874,496) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 874496 -Total Bytes 198948 -BPFP 1.8200 bits/point -EBPFP 3.6400 equivalent bits/point -MSE 0.869711 ----------------------- -------------------------------------------------------- -Time: 2.729s Load: 0.004s, Pack+Encode: 1.575s, Decode+Unpack: 1.150s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 61, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8697 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample778-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst (90/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 59, 128) -Output shape: (1, 59, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.output: torch.Size([1, 59, 4096]) -> torch.Size([1, 1, 59, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,488B, BPFP=0.5943 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,352B, BPFP=2.2977 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,748B, BPFP=1.8204 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 18,516B, BPFP=2.4518 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,800B, BPFP=1.9597 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 18,992B, BPFP=2.5148 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,388B, BPFP=2.0376 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 18,492B, BPFP=2.4486 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,580B, BPFP=1.7982 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 19,076B, BPFP=2.5260 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 59,852B, BPFP=1.9813 -⌛️ [2/4] FRONTEND: Frontend time: 1.564s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.175s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03100202 7.91141316 - layer.0.v_cache 0.00000029 0.00017845 - layer.1.k_cache 0.00372036 0.78293267 - layer.1.v_cache 0.00000085 0.00065278 - layer.2.k_cache 0.00115281 0.44282389 - layer.2.v_cache 0.00000145 0.00094268 - layer.3.k_cache 0.00140016 0.53777889 - layer.3.v_cache 0.00000257 0.00167136 - layer.4.k_cache 0.00323529 0.97168111 - layer.4.v_cache 0.00000321 0.00249461 - layer.4.output 0.00024217 0.08944083 - ------------------------------------------------------------------------------------- - TOTAL 0.00296341 0.78645235 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 214284 -BPFP 2.0267 bits/point -EBPFP 4.0535 equivalent bits/point -MSE 0.786452 ----------------------- -------------------------------------------------------- -Time: 2.743s Load: 0.004s, Pack+Encode: 1.564s, Decode+Unpack: 1.175s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7865 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample807-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst (91/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 66, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 66, 128) -Output shape: (1, 66, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.output: torch.Size([1, 66, 4096]) -> torch.Size([1, 1, 66, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,196B, BPFP=0.4967 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,620B, BPFP=2.5592 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,764B, BPFP=1.5109 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,584B, BPFP=2.7917 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,272B, BPFP=1.8078 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,664B, BPFP=2.8011 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,292B, BPFP=1.9285 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,936B, BPFP=2.7150 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,128B, BPFP=1.4356 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,544B, BPFP=2.7869 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 42,192B, BPFP=1.2486 -⌛️ [2/4] FRONTEND: Frontend time: 1.776s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.output: torch.Size([1, 66, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.340s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.output: torch.Size([1, 66, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02815026 8.32131773 - layer.0.v_cache 0.00000028 0.00017385 - layer.1.k_cache 0.00365239 0.75270815 - layer.1.v_cache 0.00000074 0.00057710 - layer.2.k_cache 0.00118996 0.43877538 - layer.2.v_cache 0.00000101 0.00076054 - layer.3.k_cache 0.00141080 0.47899217 - layer.3.v_cache 0.00000191 0.00122286 - layer.4.k_cache 0.00329622 0.89680868 - layer.4.v_cache 0.00000294 0.00212992 - layer.4.output 0.00019151 0.11046610 - ------------------------------------------------------------------------------------- - TOTAL 0.00274804 0.80966649 - (elements=946,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 946176 -Total Bytes 218192 -BPFP 1.8448 bits/point -EBPFP 3.6897 equivalent bits/point -MSE 0.809666 ----------------------- -------------------------------------------------------- -Time: 3.121s Load: 0.005s, Pack+Encode: 1.776s, Decode+Unpack: 1.340s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 66, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.8097 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample855-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst (92/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,532B, BPFP=0.4850 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 28,932B, BPFP=3.0963 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,368B, BPFP=1.5377 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 30,996B, BPFP=3.3172 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,988B, BPFP=1.8181 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 31,972B, BPFP=3.4217 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,776B, BPFP=2.0094 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 31,316B, BPFP=3.3515 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,840B, BPFP=1.4812 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 30,984B, BPFP=3.3159 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 53,948B, BPFP=1.4434 -⌛️ [2/4] FRONTEND: Frontend time: 1.779s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.371s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02483096 7.88156504 - layer.0.v_cache 0.00000028 0.00016680 - layer.1.k_cache 0.00355597 0.73643546 - layer.1.v_cache 0.00000080 0.00056119 - layer.2.k_cache 0.00115478 0.42799790 - layer.2.v_cache 0.00000105 0.00077599 - layer.3.k_cache 0.00135922 0.44622233 - layer.3.v_cache 0.00000200 0.00127138 - layer.4.k_cache 0.00333109 0.92606929 - layer.4.v_cache 0.00000279 0.00205641 - layer.4.output 0.00023199 0.07810727 - ------------------------------------------------------------------------------------- - TOTAL 0.00251192 0.76682506 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 276652 -BPFP 2.1148 bits/point -EBPFP 4.2296 equivalent bits/point -MSE 0.766825 ----------------------- -------------------------------------------------------- -Time: 3.154s Load: 0.004s, Pack+Encode: 1.779s, Decode+Unpack: 1.371s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7668 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample859-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst (93/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 121, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 121, 128) -Output shape: (1, 121, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.output: torch.Size([1, 121, 4096]) -> torch.Size([1, 1, 121, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,988B, BPFP=0.4512 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,332B, BPFP=2.1521 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,212B, BPFP=1.6278 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,484B, BPFP=2.2911 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,592B, BPFP=1.9106 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,872B, BPFP=2.3161 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,220B, BPFP=1.9512 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,640B, BPFP=2.3011 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 26,516B, BPFP=1.7120 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,284B, BPFP=2.3427 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 88,536B, BPFP=1.4291 -⌛️ [2/4] FRONTEND: Frontend time: 1.816s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.output: torch.Size([1, 121, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.389s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.output: torch.Size([1, 121, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02495229 7.58927173 - layer.0.v_cache 0.00000028 0.00015516 - layer.1.k_cache 0.00332455 0.88943267 - layer.1.v_cache 0.00000080 0.00053368 - layer.2.k_cache 0.00116246 0.38474974 - layer.2.v_cache 0.00000109 0.00071246 - layer.3.k_cache 0.00134746 0.42041243 - layer.3.v_cache 0.00000209 0.00121893 - layer.4.k_cache 0.00345892 0.86880619 - layer.4.v_cache 0.00000309 0.00207563 - layer.4.output 0.00018371 0.07832298 - ------------------------------------------------------------------------------------- - TOTAL 0.00249913 0.74790432 - (elements=1,734,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1734656 -Total Bytes 383676 -BPFP 1.7695 bits/point -EBPFP 3.5389 equivalent bits/point -MSE 0.747904 ----------------------- -------------------------------------------------------- -Time: 3.212s Load: 0.007s, Pack+Encode: 1.816s, Decode+Unpack: 1.389s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 121, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7479 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample86-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst (94/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 113, 128) -Output shape: (1, 113, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.output: torch.Size([1, 113, 4096]) -> torch.Size([1, 1, 113, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,768B, BPFP=0.4679 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,608B, BPFP=2.3236 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,664B, BPFP=1.6361 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,432B, BPFP=2.4497 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 27,924B, BPFP=1.9306 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,896B, BPFP=2.4817 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,324B, BPFP=2.0965 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,892B, BPFP=2.4815 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,852B, BPFP=1.6491 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,432B, BPFP=2.5188 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,452B, BPFP=1.2350 -⌛️ [2/4] FRONTEND: Frontend time: 1.797s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.368s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02530831 7.86332392 - layer.0.v_cache 0.00000029 0.00015876 - layer.1.k_cache 0.00330439 0.86241447 - layer.1.v_cache 0.00000076 0.00052509 - layer.2.k_cache 0.00114562 0.39329320 - layer.2.v_cache 0.00000109 0.00072278 - layer.3.k_cache 0.00136171 0.42770082 - layer.3.v_cache 0.00000227 0.00126999 - layer.4.k_cache 0.00342222 0.82856649 - layer.4.v_cache 0.00000313 0.00205579 - layer.4.output 0.00018801 0.07293995 - ------------------------------------------------------------------------------------- - TOTAL 0.00252156 0.76227079 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 361244 -BPFP 1.7840 bits/point -EBPFP 3.5679 equivalent bits/point -MSE 0.762271 ----------------------- -------------------------------------------------------- -Time: 3.173s Load: 0.008s, Pack+Encode: 1.797s, Decode+Unpack: 1.368s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7623 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample91-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst (95/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 110, 128) -Output shape: (1, 110, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.output: torch.Size([1, 110, 4096]) -> torch.Size([1, 1, 110, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,420B, BPFP=0.4560 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,388B, BPFP=2.3713 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,608B, BPFP=1.6057 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,500B, BPFP=2.5213 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,360B, BPFP=2.0852 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,088B, BPFP=2.5631 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,012B, BPFP=2.1315 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,456B, BPFP=2.5182 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 24,044B, BPFP=1.7077 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,296B, BPFP=2.5778 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,240B, BPFP=1.3004 -⌛️ [2/4] FRONTEND: Frontend time: 1.789s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.output: torch.Size([1, 110, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.391s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.output: torch.Size([1, 110, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02547969 7.10403775 - layer.0.v_cache 0.00000028 0.00015551 - layer.1.k_cache 0.00323236 0.81471093 - layer.1.v_cache 0.00000082 0.00052432 - layer.2.k_cache 0.00114874 0.37159327 - layer.2.v_cache 0.00000131 0.00076605 - layer.3.k_cache 0.00138397 0.43903087 - layer.3.v_cache 0.00000212 0.00124386 - layer.4.k_cache 0.00346958 0.90786570 - layer.4.v_cache 0.00000311 0.00209943 - layer.4.output 0.00021756 0.08370031 - ------------------------------------------------------------------------------------- - TOTAL 0.00254230 0.71263064 - (elements=1,576,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1576960 -Total Bytes 362412 -BPFP 1.8385 bits/point -EBPFP 3.6771 equivalent bits/point -MSE 0.712631 ----------------------- -------------------------------------------------------- -Time: 3.188s Load: 0.009s, Pack+Encode: 1.789s, Decode+Unpack: 1.391s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7126 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample92-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst (96/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 65, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 65, 128) -Output shape: (1, 65, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.output: torch.Size([1, 65, 4096]) -> torch.Size([1, 1, 65, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,200B, BPFP=0.5048 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,704B, BPFP=2.6087 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,020B, BPFP=1.4447 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,632B, BPFP=2.9606 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,164B, BPFP=1.7024 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,756B, BPFP=2.7351 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,968B, BPFP=1.9192 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,340B, BPFP=2.8053 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,996B, BPFP=1.4418 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,624B, BPFP=2.8394 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 43,144B, BPFP=1.2964 -⌛️ [2/4] FRONTEND: Frontend time: 1.797s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.output: torch.Size([1, 65, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.354s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.output: torch.Size([1, 65, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02613364 9.73713379 - layer.0.v_cache 0.00000028 0.00017669 - layer.1.k_cache 0.00351263 0.86550164 - layer.1.v_cache 0.00000086 0.00058782 - layer.2.k_cache 0.00113140 0.44078275 - layer.2.v_cache 0.00000103 0.00077079 - layer.3.k_cache 0.00134002 0.45098874 - layer.3.v_cache 0.00000212 0.00131812 - layer.4.k_cache 0.00331905 0.89600631 - layer.4.v_cache 0.00000293 0.00217928 - layer.4.output 0.00019456 0.08090611 - ------------------------------------------------------------------------------------- - TOTAL 0.00258730 0.90850503 - (elements=931,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 931840 -Total Bytes 217548 -BPFP 1.8677 bits/point -EBPFP 3.7354 equivalent bits/point -MSE 0.908505 ----------------------- -------------------------------------------------------- -Time: 3.156s Load: 0.006s, Pack+Encode: 1.797s, Decode+Unpack: 1.354s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 65, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.9085 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample925-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst (97/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 116, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 116, 128) -Output shape: (1, 116, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.output: torch.Size([1, 116, 4096]) -> torch.Size([1, 1, 116, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,844B, BPFP=0.4609 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,516B, BPFP=2.2573 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,780B, BPFP=1.6016 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,744B, BPFP=2.4073 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,164B, BPFP=1.9642 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,056B, BPFP=2.4283 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,148B, BPFP=2.0304 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,784B, BPFP=2.4100 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,968B, BPFP=1.7489 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,492B, BPFP=2.4577 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 82,952B, BPFP=1.3967 -⌛️ [2/4] FRONTEND: Frontend time: 1.849s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.output: torch.Size([1, 116, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.361s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.output: torch.Size([1, 116, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02494503 7.61799358 - layer.0.v_cache 0.00000027 0.00015623 - layer.1.k_cache 0.00326550 0.90765644 - layer.1.v_cache 0.00000084 0.00054581 - layer.2.k_cache 0.00114506 0.36613777 - layer.2.v_cache 0.00000107 0.00074424 - layer.3.k_cache 0.00140586 0.43717263 - layer.3.v_cache 0.00000237 0.00130764 - layer.4.k_cache 0.00329219 0.86159469 - layer.4.v_cache 0.00000320 0.00217818 - layer.4.output 0.00021339 0.07739288 - ------------------------------------------------------------------------------------- - TOTAL 0.00249392 0.75036134 - (elements=1,662,976) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1662976 -Total Bytes 376448 -BPFP 1.8110 bits/point -EBPFP 3.6219 equivalent bits/point -MSE 0.750361 ----------------------- -------------------------------------------------------- -Time: 3.217s Load: 0.007s, Pack+Encode: 1.849s, Decode+Unpack: 1.361s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 116, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7504 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample95-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst (98/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 59, 128) -Output shape: (1, 59, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.output: torch.Size([1, 59, 4096]) -> torch.Size([1, 1, 59, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,096B, BPFP=0.5424 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,988B, BPFP=2.2495 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,532B, BPFP=1.6594 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 18,292B, BPFP=2.4221 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,336B, BPFP=1.8983 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 18,536B, BPFP=2.4544 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,072B, BPFP=1.9958 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 17,912B, BPFP=2.3718 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,452B, BPFP=1.7812 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 18,584B, BPFP=2.4608 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 48,536B, BPFP=1.6067 -⌛️ [2/4] FRONTEND: Frontend time: 1.567s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.143s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02763205 7.95157429 - layer.0.v_cache 0.00000028 0.00017579 - layer.1.k_cache 0.00373147 0.80340046 - layer.1.v_cache 0.00000088 0.00061409 - layer.2.k_cache 0.00131257 0.43821509 - layer.2.v_cache 0.00000108 0.00080454 - layer.3.k_cache 0.00142120 0.51682424 - layer.3.v_cache 0.00000202 0.00135448 - layer.4.k_cache 0.00321287 1.00516930 - layer.4.v_cache 0.00000283 0.00209955 - layer.4.output 0.00027608 0.10518352 - ------------------------------------------------------------------------------------- - TOTAL 0.00274440 0.79578328 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 198336 -BPFP 1.8759 bits/point -EBPFP 3.7518 equivalent bits/point -MSE 0.795783 ----------------------- -------------------------------------------------------- -Time: 2.714s Load: 0.004s, Pack+Encode: 1.567s, Decode+Unpack: 1.143s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7958 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample967-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst (99/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 60, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 60, 128) -Output shape: (1, 60, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.output: torch.Size([1, 60, 4096]) -> torch.Size([1, 1, 60, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 4,260B, BPFP=0.5547 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,268B, BPFP=2.2484 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,640B, BPFP=1.7760 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 18,800B, BPFP=2.4479 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,980B, BPFP=1.8203 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 19,040B, BPFP=2.4792 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,520B, BPFP=2.0208 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 18,752B, BPFP=2.4417 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,664B, BPFP=1.7792 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 18,776B, BPFP=2.4448 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 48,132B, BPFP=1.5668 -⌛️ [2/4] FRONTEND: Frontend time: 1.597s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.output: torch.Size([1, 60, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.176s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.output: torch.Size([1, 60, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02844323 7.64265238 - layer.0.v_cache 0.00000027 0.00016986 - layer.1.k_cache 0.00360089 0.87657693 - layer.1.v_cache 0.00000087 0.00057923 - layer.2.k_cache 0.00116909 0.43805059 - layer.2.v_cache 0.00000119 0.00081948 - layer.3.k_cache 0.00139808 0.50993824 - layer.3.v_cache 0.00000212 0.00131359 - layer.4.k_cache 0.00324009 1.00902481 - layer.4.v_cache 0.00000310 0.00208676 - layer.4.output 0.00020032 0.08225494 - ------------------------------------------------------------------------------------- - TOTAL 0.00276144 0.77215940 - (elements=860,160) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 860160 -Total Bytes 201832 -BPFP 1.8772 bits/point -EBPFP 3.7543 equivalent bits/point -MSE 0.772159 ----------------------- -------------------------------------------------------- -Time: 2.779s Load: 0.006s, Pack+Encode: 1.597s, Decode+Unpack: 1.176s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 60, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7722 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample969-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst (100/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 114, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 114, 128) -Output shape: (1, 114, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.output: torch.Size([1, 114, 4096]) -> torch.Size([1, 1, 114, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 6,472B, BPFP=0.4435 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,308B, BPFP=2.2826 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 23,336B, BPFP=1.5992 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,476B, BPFP=2.4312 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,204B, BPFP=2.0014 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,840B, BPFP=2.4561 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,356B, BPFP=2.0803 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,524B, BPFP=2.4345 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 24,580B, BPFP=1.6845 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,256B, BPFP=2.4846 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 77,676B, BPFP=1.3308 -⌛️ [2/4] FRONTEND: Frontend time: 1.840s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.output: torch.Size([1, 114, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.372s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.output: torch.Size([1, 114, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435601 7.61603319 - layer.0.v_cache 0.00000028 0.00015952 - layer.1.k_cache 0.00337316 0.83020468 - layer.1.v_cache 0.00000076 0.00053722 - layer.2.k_cache 0.00114524 0.37212489 - layer.2.v_cache 0.00000110 0.00078139 - layer.3.k_cache 0.00139029 0.44121106 - layer.3.v_cache 0.00000222 0.00127320 - layer.4.k_cache 0.00339712 0.85869946 - layer.4.v_cache 0.00000302 0.00213202 - layer.4.output 0.00021100 0.07414947 - ------------------------------------------------------------------------------------- - TOTAL 0.00246523 0.74426818 - (elements=1,634,304) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1634304 -Total Bytes 368028 -BPFP 1.8015 bits/point -EBPFP 3.6030 equivalent bits/point -MSE 0.744268 ----------------------- -------------------------------------------------------- -Time: 3.220s Load: 0.008s, Pack+Encode: 1.840s, Decode+Unpack: 1.372s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 114, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7443 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_gsm8k/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 1.9693 bits/point -Avg EBPFP 3.9386 equivalent bits/point -Avg MSE 0.769041 -Avg Time 3.169s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:ca8324bbabb75d6eb5655e26298d02208b7368d608eb5c433dc004b8c4ec0abe +size 1117689 diff --git a/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/dtufc_elic-featurecoding_qwen_individual.log b/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/dtufc_elic-featurecoding_qwen_individual.log index a2cda721fba44acdaa7d7726fc9a7267311c0d3d..4c27c453a2058f92003177c2742bc1b9bc432982 100644 --- a/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/dtufc_elic-featurecoding_qwen_individual.log +++ b/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/dtufc_elic-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_elic-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/dtufc_elic-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: elic-featurecoding - handler: qwen - checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar - transform_type: kmeans - transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json - bit_depth: 8 - device: cuda:0 -Loading checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 506 -Loaded elic-featurecoding (1-channel) on cuda:0 -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json -Loaded per-key mappings: model=qwen - Keys: ['layer.0.k_cache', 'layer.0.v_cache', 'layer.1.k_cache', 'layer.1.v_cache', 'layer.2.k_cache', 'layer.2.v_cache', 'layer.3.k_cache', 'layer.3.v_cache', 'layer.4.k_cache', 'layer.4.v_cache', 'layer.4.output'] ----------------- ------------------------------------------------------------------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding -Checkpoint codec_weights/elic_hybrid/elic2022-official_lambda0.01_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar -Transform type kmeans -Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json -Input ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag -Output output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag ----------------- ------------------------------------------------------------------------------------------------------------------- -Files found: 100 ----------------------------------------------------------------------- - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst (1/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 371, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 371, 128) -Output shape: (1, 371, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.output: torch.Size([1, 371, 4096]) -> torch.Size([1, 1, 371, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,004B, BPFP=0.4002 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,228B, BPFP=2.0895 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 73,284B, BPFP=1.5432 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,036B, BPFP=2.2118 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,732B, BPFP=1.8685 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,796B, BPFP=2.2489 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,424B, BPFP=1.8831 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,724B, BPFP=2.2263 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 76,584B, BPFP=1.6127 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,844B, BPFP=2.2710 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 246,892B, BPFP=1.2998 -⌛️ [2/4] FRONTEND: Frontend time: 3.073s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.output: torch.Size([1, 371, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.301s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.output: torch.Size([1, 371, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02382561 6.49697950 - layer.0.v_cache 0.00000026 0.00013614 - layer.1.k_cache 0.00289382 0.80596685 - layer.1.v_cache 0.00000075 0.00048769 - layer.2.k_cache 0.00115144 0.35120032 - layer.2.v_cache 0.00000114 0.00071909 - layer.3.k_cache 0.00133317 0.42444259 - layer.3.v_cache 0.00000213 0.00116362 - layer.4.k_cache 0.00354181 0.72764156 - layer.4.v_cache 0.00000324 0.00200359 - layer.4.output 0.00015639 0.05818950 - ------------------------------------------------------------------------------------- - TOTAL 0.00238421 0.64596421 - (elements=5,318,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5318656 -Total Bytes 1118548 -BPFP 1.6825 bits/point -EBPFP 3.3649 equivalent bits/point -MSE 0.645964 ----------------------- -------------------------------------------------------- -Time: 5.391s Load: 0.017s, Pack+Encode: 3.073s, Decode+Unpack: 2.301s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 371, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6460 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample0-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst (2/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,824B, BPFP=0.4275 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,868B, BPFP=2.2454 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,616B, BPFP=1.5810 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,704B, BPFP=2.4006 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,068B, BPFP=2.0001 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,316B, BPFP=2.4372 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,968B, BPFP=1.9978 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,396B, BPFP=2.4163 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 71,112B, BPFP=1.6150 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,792B, BPFP=2.4480 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 267,480B, BPFP=1.5187 -⌛️ [2/4] FRONTEND: Frontend time: 2.580s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.261s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02409526 6.96032360 - layer.0.v_cache 0.00000026 0.00013817 - layer.1.k_cache 0.00287302 0.84415968 - layer.1.v_cache 0.00000078 0.00049917 - layer.2.k_cache 0.00123767 0.35291525 - layer.2.v_cache 0.00000117 0.00075340 - layer.3.k_cache 0.00130179 0.40837004 - layer.3.v_cache 0.00000226 0.00121872 - layer.4.k_cache 0.00355144 0.71768978 - layer.4.v_cache 0.00000325 0.00204822 - layer.4.output 0.00015063 0.06580902 - ------------------------------------------------------------------------------------- - TOTAL 0.00240496 0.68223944 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 1129144 -BPFP 1.8317 bits/point -EBPFP 3.6634 equivalent bits/point -MSE 0.682239 ----------------------- -------------------------------------------------------- -Time: 4.858s Load: 0.017s, Pack+Encode: 2.580s, Decode+Unpack: 2.261s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6822 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample1-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst (3/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 377, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 377, 128) -Output shape: (1, 377, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.output: torch.Size([1, 377, 4096]) -> torch.Size([1, 1, 377, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 20,224B, BPFP=0.4191 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,740B, BPFP=2.0669 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 73,676B, BPFP=1.5268 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 106,168B, BPFP=2.2001 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,164B, BPFP=1.8477 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,696B, BPFP=2.2318 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,292B, BPFP=1.8504 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,264B, BPFP=2.2021 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 76,496B, BPFP=1.5852 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,856B, BPFP=2.2558 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 264,124B, BPFP=1.3683 -⌛️ [2/4] FRONTEND: Frontend time: 2.588s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.output: torch.Size([1, 377, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.252s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.output: torch.Size([1, 377, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02400962 6.99759097 - layer.0.v_cache 0.00000027 0.00014131 - layer.1.k_cache 0.00291640 0.75641468 - layer.1.v_cache 0.00000080 0.00051967 - layer.2.k_cache 0.00116459 0.36006606 - layer.2.v_cache 0.00000114 0.00073162 - layer.3.k_cache 0.00131780 0.42623735 - layer.3.v_cache 0.00000210 0.00116291 - layer.4.k_cache 0.00355012 0.71487953 - layer.4.v_cache 0.00000332 0.00209601 - layer.4.output 0.00014433 0.05828713 - ------------------------------------------------------------------------------------- - TOTAL 0.00239596 0.67807062 - (elements=5,404,672) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5404672 -Total Bytes 1141700 -BPFP 1.6899 bits/point -EBPFP 3.3799 equivalent bits/point -MSE 0.678071 ----------------------- -------------------------------------------------------- -Time: 4.858s Load: 0.019s, Pack+Encode: 2.588s, Decode+Unpack: 2.252s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 377, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6781 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample10-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst (4/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 315, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 315, 128) -Output shape: (1, 315, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.output: torch.Size([1, 315, 4096]) -> torch.Size([1, 1, 315, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,140B, BPFP=0.4251 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,368B, BPFP=2.0677 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,140B, BPFP=1.5164 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 89,020B, BPFP=2.2078 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,356B, BPFP=1.8441 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 90,064B, BPFP=2.2337 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,512B, BPFP=1.8480 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,020B, BPFP=2.2078 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,092B, BPFP=1.5648 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 91,016B, BPFP=2.2573 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 235,272B, BPFP=1.4588 -⌛️ [2/4] FRONTEND: Frontend time: 2.590s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.output: torch.Size([1, 315, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.896s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.output: torch.Size([1, 315, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02468849 7.36195747 - layer.0.v_cache 0.00000027 0.00014118 - layer.1.k_cache 0.00284817 0.85101842 - layer.1.v_cache 0.00000082 0.00051835 - layer.2.k_cache 0.00122538 0.35036858 - layer.2.v_cache 0.00000116 0.00072802 - layer.3.k_cache 0.00131111 0.41201409 - layer.3.v_cache 0.00000216 0.00120803 - layer.4.k_cache 0.00356899 0.69294007 - layer.4.v_cache 0.00000330 0.00206128 - layer.4.output 0.00013845 0.05694077 - ------------------------------------------------------------------------------------- - TOTAL 0.00244312 0.70719418 - (elements=4,515,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4515840 -Total Bytes 968000 -BPFP 1.7149 bits/point -EBPFP 3.4297 equivalent bits/point -MSE 0.707194 ----------------------- -------------------------------------------------------- -Time: 4.501s Load: 0.016s, Pack+Encode: 2.590s, Decode+Unpack: 1.896s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 315, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7072 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample100-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst (5/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 362, 128) -Output shape: (1, 362, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.output: torch.Size([1, 362, 4096]) -> torch.Size([1, 1, 362, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,572B, BPFP=0.4008 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,428B, BPFP=2.1458 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,344B, BPFP=1.5613 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,836B, BPFP=2.2841 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,100B, BPFP=1.9229 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,188B, BPFP=2.3133 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,036B, BPFP=1.9215 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,212B, BPFP=2.2922 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,172B, BPFP=1.6007 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,216B, BPFP=2.3355 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 230,048B, BPFP=1.2412 -⌛️ [2/4] FRONTEND: Frontend time: 2.659s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.258s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02398593 6.98864611 - layer.0.v_cache 0.00000027 0.00013943 - layer.1.k_cache 0.00294127 0.77212390 - layer.1.v_cache 0.00000079 0.00052089 - layer.2.k_cache 0.00116280 0.35293347 - layer.2.v_cache 0.00000114 0.00073220 - layer.3.k_cache 0.00131002 0.40615019 - layer.3.v_cache 0.00000215 0.00118586 - layer.4.k_cache 0.00366934 0.73073801 - layer.4.v_cache 0.00000328 0.00209868 - layer.4.output 0.00013411 0.05898833 - ------------------------------------------------------------------------------------- - TOTAL 0.00240096 0.67794443 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 1100152 -BPFP 1.6959 bits/point -EBPFP 3.3918 equivalent bits/point -MSE 0.677944 ----------------------- -------------------------------------------------------- -Time: 4.935s Load: 0.018s, Pack+Encode: 2.659s, Decode+Unpack: 2.258s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6779 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample102-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst (6/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,064B, BPFP=0.4200 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,900B, BPFP=2.2531 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,432B, BPFP=1.5446 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,564B, BPFP=2.3848 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,540B, BPFP=1.9424 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,384B, BPFP=2.4503 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,404B, BPFP=2.0090 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,132B, BPFP=2.4212 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,528B, BPFP=1.5701 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,692B, BPFP=2.4575 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 229,552B, BPFP=1.3344 -⌛️ [2/4] FRONTEND: Frontend time: 2.593s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.223s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02466884 6.90184748 - layer.0.v_cache 0.00000026 0.00014114 - layer.1.k_cache 0.00293366 0.84706070 - layer.1.v_cache 0.00000078 0.00051193 - layer.2.k_cache 0.00117059 0.36376376 - layer.2.v_cache 0.00000113 0.00072616 - layer.3.k_cache 0.00132874 0.39424678 - layer.3.v_cache 0.00000215 0.00120046 - layer.4.k_cache 0.00360252 0.77183351 - layer.4.v_cache 0.00000314 0.00197724 - layer.4.output 0.00014135 0.05984185 - ------------------------------------------------------------------------------------- - TOTAL 0.00244837 0.68019118 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 1066192 -BPFP 1.7708 bits/point -EBPFP 3.5415 equivalent bits/point -MSE 0.680191 ----------------------- -------------------------------------------------------- -Time: 4.832s Load: 0.016s, Pack+Encode: 2.593s, Decode+Unpack: 2.223s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6802 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample105-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst (7/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 317, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 317, 128) -Output shape: (1, 317, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.output: torch.Size([1, 317, 4096]) -> torch.Size([1, 1, 317, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,820B, BPFP=0.4145 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,128B, BPFP=2.0487 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,364B, BPFP=1.5123 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,564B, BPFP=2.1827 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,104B, BPFP=1.8263 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,812B, BPFP=2.2134 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,556B, BPFP=1.8374 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,952B, BPFP=2.1922 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,544B, BPFP=1.5660 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,416B, BPFP=2.2283 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 236,052B, BPFP=1.4544 -⌛️ [2/4] FRONTEND: Frontend time: 2.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.output: torch.Size([1, 317, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.059s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.output: torch.Size([1, 317, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02499472 7.15556379 - layer.0.v_cache 0.00000027 0.00013807 - layer.1.k_cache 0.00294942 0.86779930 - layer.1.v_cache 0.00000081 0.00050432 - layer.2.k_cache 0.00116222 0.37589603 - layer.2.v_cache 0.00000115 0.00072479 - layer.3.k_cache 0.00131165 0.43151904 - layer.3.v_cache 0.00000215 0.00119102 - layer.4.k_cache 0.00349495 0.72411792 - layer.4.v_cache 0.00000332 0.00204717 - layer.4.output 0.00014092 0.06086446 - ------------------------------------------------------------------------------------- - TOTAL 0.00246317 0.70021138 - (elements=4,544,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4544512 -Total Bytes 967312 -BPFP 1.7028 bits/point -EBPFP 3.4056 equivalent bits/point -MSE 0.700211 ----------------------- -------------------------------------------------------- -Time: 4.481s Load: 0.016s, Pack+Encode: 2.406s, Decode+Unpack: 2.059s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 317, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7002 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample108-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst (8/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 353, 128) -Output shape: (1, 353, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.output: torch.Size([1, 353, 4096]) -> torch.Size([1, 1, 353, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,276B, BPFP=0.4045 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,956B, BPFP=2.1901 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,112B, BPFP=1.5960 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,888B, BPFP=2.3214 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,580B, BPFP=1.9604 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,816B, BPFP=2.3640 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,188B, BPFP=1.9739 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,528B, BPFP=2.3355 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,828B, BPFP=1.6118 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,520B, BPFP=2.3796 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 223,628B, BPFP=1.2373 -⌛️ [2/4] FRONTEND: Frontend time: 2.554s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.175s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02498756 7.31347795 - layer.0.v_cache 0.00000027 0.00014153 - layer.1.k_cache 0.00288490 0.79647931 - layer.1.v_cache 0.00000079 0.00052780 - layer.2.k_cache 0.00118233 0.35926201 - layer.2.v_cache 0.00000115 0.00075531 - layer.3.k_cache 0.00132883 0.40515197 - layer.3.v_cache 0.00000217 0.00117489 - layer.4.k_cache 0.00368366 0.74621772 - layer.4.v_cache 0.00000317 0.00205328 - layer.4.output 0.00013605 0.05902886 - ------------------------------------------------------------------------------------- - TOTAL 0.00247279 0.70438266 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 1088320 -BPFP 1.7205 bits/point -EBPFP 3.4409 equivalent bits/point -MSE 0.704383 ----------------------- -------------------------------------------------------- -Time: 4.747s Load: 0.018s, Pack+Encode: 2.554s, Decode+Unpack: 2.175s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7044 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample11-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst (9/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 318, 128) -Output shape: (1, 318, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.output: torch.Size([1, 318, 4096]) -> torch.Size([1, 1, 318, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,044B, BPFP=0.4187 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,424B, BPFP=2.0495 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,452B, BPFP=1.4852 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 87,480B, BPFP=2.1492 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 73,716B, BPFP=1.8110 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,048B, BPFP=2.1877 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 75,004B, BPFP=1.8427 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,476B, BPFP=2.1736 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,308B, BPFP=1.5553 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 89,960B, BPFP=2.2101 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 257,200B, BPFP=1.5797 -⌛️ [2/4] FRONTEND: Frontend time: 2.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.981s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461332 6.72001600 - layer.0.v_cache 0.00000025 0.00013446 - layer.1.k_cache 0.00290030 0.86362817 - layer.1.v_cache 0.00000077 0.00046051 - layer.2.k_cache 0.00116467 0.37469785 - layer.2.v_cache 0.00000108 0.00066081 - layer.3.k_cache 0.00132360 0.46123980 - layer.3.v_cache 0.00000220 0.00112844 - layer.4.k_cache 0.00367546 0.85513469 - layer.4.v_cache 0.00000314 0.00192966 - layer.4.output 0.00019099 0.06471643 - ------------------------------------------------------------------------------------- - TOTAL 0.00246062 0.68127829 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 985112 -BPFP 1.7287 bits/point -EBPFP 3.4574 equivalent bits/point -MSE 0.681278 ----------------------- -------------------------------------------------------- -Time: 4.398s Load: 0.016s, Pack+Encode: 2.401s, Decode+Unpack: 1.981s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6813 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample112-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst (10/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,812B, BPFP=0.4217 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,020B, BPFP=2.3205 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,216B, BPFP=1.5676 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,196B, BPFP=2.4668 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,080B, BPFP=1.9669 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,820B, BPFP=2.4579 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,256B, BPFP=2.0184 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,416B, BPFP=2.4483 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,300B, BPFP=1.5933 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,248B, BPFP=2.4917 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 253,288B, BPFP=1.4991 -⌛️ [2/4] FRONTEND: Frontend time: 2.588s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.210s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02377391 6.81328717 - layer.0.v_cache 0.00000027 0.00014292 - layer.1.k_cache 0.00289258 0.80397265 - layer.1.v_cache 0.00000083 0.00050787 - layer.2.k_cache 0.00117497 0.35363508 - layer.2.v_cache 0.00000126 0.00075963 - layer.3.k_cache 0.00131231 0.39046478 - layer.3.v_cache 0.00000230 0.00121887 - layer.4.k_cache 0.00357793 0.69361193 - layer.4.v_cache 0.00000347 0.00211767 - layer.4.output 0.00014070 0.05455338 - ------------------------------------------------------------------------------------- - TOTAL 0.00237876 0.66270944 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 1087652 -BPFP 1.8392 bits/point -EBPFP 3.6785 equivalent bits/point -MSE 0.662709 ----------------------- -------------------------------------------------------- -Time: 4.815s Load: 0.016s, Pack+Encode: 2.588s, Decode+Unpack: 2.210s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6627 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample114-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst (11/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 357, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 357, 128) -Output shape: (1, 357, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.output: torch.Size([1, 357, 4096]) -> torch.Size([1, 1, 357, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,480B, BPFP=0.4263 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,504B, BPFP=2.1775 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 71,756B, BPFP=1.5703 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,916B, BPFP=2.2960 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,992B, BPFP=1.9475 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,604B, BPFP=2.3329 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,920B, BPFP=1.9459 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,812B, BPFP=2.3156 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,448B, BPFP=1.6292 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,556B, BPFP=2.3537 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 317,492B, BPFP=1.7370 -⌛️ [2/4] FRONTEND: Frontend time: 2.586s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.output: torch.Size([1, 357, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.259s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.output: torch.Size([1, 357, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02468803 6.69285536 - layer.0.v_cache 0.00000027 0.00014427 - layer.1.k_cache 0.00290374 0.86098705 - layer.1.v_cache 0.00000082 0.00051472 - layer.2.k_cache 0.00112985 0.35360570 - layer.2.v_cache 0.00000115 0.00071581 - layer.3.k_cache 0.00131122 0.41863373 - layer.3.v_cache 0.00000220 0.00119161 - layer.4.k_cache 0.00345488 0.70350510 - layer.4.v_cache 0.00000325 0.00205292 - layer.4.output 0.00016267 0.05552307 - ------------------------------------------------------------------------------------- - TOTAL 0.00243901 0.66116418 - (elements=5,117,952) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5117952 -Total Bytes 1185480 -BPFP 1.8531 bits/point -EBPFP 3.7061 equivalent bits/point -MSE 0.661164 ----------------------- -------------------------------------------------------- -Time: 4.866s Load: 0.022s, Pack+Encode: 2.586s, Decode+Unpack: 2.259s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 357, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6612 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample119-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst (12/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 358, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 358, 128) -Output shape: (1, 358, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.output: torch.Size([1, 358, 4096]) -> torch.Size([1, 1, 358, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,080B, BPFP=0.4164 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,344B, BPFP=2.1679 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 71,452B, BPFP=1.5593 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,956B, BPFP=2.2904 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 90,184B, BPFP=1.9681 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,836B, BPFP=2.3314 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,532B, BPFP=1.9320 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,748B, BPFP=2.3077 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 73,392B, BPFP=1.6016 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,744B, BPFP=2.3513 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 277,096B, BPFP=1.5117 -⌛️ [2/4] FRONTEND: Frontend time: 2.642s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.output: torch.Size([1, 358, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.286s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.output: torch.Size([1, 358, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02418174 7.13174046 - layer.0.v_cache 0.00000026 0.00014056 - layer.1.k_cache 0.00288075 0.83820765 - layer.1.v_cache 0.00000077 0.00050435 - layer.2.k_cache 0.00118336 0.34074295 - layer.2.v_cache 0.00000117 0.00072950 - layer.3.k_cache 0.00130988 0.39259215 - layer.3.v_cache 0.00000211 0.00115610 - layer.4.k_cache 0.00353008 0.71516904 - layer.4.v_cache 0.00000317 0.00204358 - layer.4.output 0.00013834 0.05973558 - ------------------------------------------------------------------------------------- - TOTAL 0.00240333 0.69014062 - (elements=5,132,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5132288 -Total Bytes 1144364 -BPFP 1.7838 bits/point -EBPFP 3.5676 equivalent bits/point -MSE 0.690141 ----------------------- -------------------------------------------------------- -Time: 4.945s Load: 0.017s, Pack+Encode: 2.642s, Decode+Unpack: 2.286s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 358, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6901 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample12-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst (13/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,160B, BPFP=0.4222 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,588B, BPFP=2.2691 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,324B, BPFP=1.5421 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,264B, BPFP=2.4243 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,272B, BPFP=1.9362 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,540B, BPFP=2.4540 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,456B, BPFP=2.0335 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,140B, BPFP=2.4447 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,204B, BPFP=1.5858 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,100B, BPFP=2.4670 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 260,812B, BPFP=1.5161 -⌛️ [2/4] FRONTEND: Frontend time: 2.576s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.002s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02407829 6.52166821 - layer.0.v_cache 0.00000026 0.00013675 - layer.1.k_cache 0.00285290 0.80743735 - layer.1.v_cache 0.00000077 0.00050368 - layer.2.k_cache 0.00116461 0.35591462 - layer.2.v_cache 0.00000115 0.00075492 - layer.3.k_cache 0.00131253 0.42881071 - layer.3.v_cache 0.00000220 0.00121019 - layer.4.k_cache 0.00369396 0.73574420 - layer.4.v_cache 0.00000328 0.00210277 - layer.4.output 0.00015589 0.06092948 - ------------------------------------------------------------------------------------- - TOTAL 0.00240954 0.64985724 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 1102860 -BPFP 1.8317 bits/point -EBPFP 3.6633 equivalent bits/point -MSE 0.649857 ----------------------- -------------------------------------------------------- -Time: 4.594s Load: 0.016s, Pack+Encode: 2.576s, Decode+Unpack: 2.002s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6499 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample13-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst (14/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 346, 128) -Output shape: (1, 346, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.output: torch.Size([1, 346, 4096]) -> torch.Size([1, 1, 346, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,744B, BPFP=0.4232 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,668B, BPFP=2.2505 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,312B, BPFP=1.5876 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,540B, BPFP=2.3830 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,308B, BPFP=1.9939 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,000B, BPFP=2.4160 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,508B, BPFP=1.9985 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,016B, BPFP=2.3938 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 70,608B, BPFP=1.5943 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,544B, BPFP=2.4283 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 283,372B, BPFP=1.5996 -⌛️ [2/4] FRONTEND: Frontend time: 2.532s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.267s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02507911 6.89173563 - layer.0.v_cache 0.00000027 0.00014050 - layer.1.k_cache 0.00290669 0.84345320 - layer.1.v_cache 0.00000082 0.00051034 - layer.2.k_cache 0.00115107 0.36395711 - layer.2.v_cache 0.00000117 0.00072290 - layer.3.k_cache 0.00130922 0.40556499 - layer.3.v_cache 0.00000214 0.00117646 - layer.4.k_cache 0.00347357 0.73285432 - layer.4.v_cache 0.00000325 0.00202430 - layer.4.output 0.00015455 0.05686956 - ------------------------------------------------------------------------------------- - TOTAL 0.00246754 0.67640128 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 1145620 -BPFP 1.8477 bits/point -EBPFP 3.6954 equivalent bits/point -MSE 0.676401 ----------------------- -------------------------------------------------------- -Time: 4.817s Load: 0.018s, Pack+Encode: 2.532s, Decode+Unpack: 2.267s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6764 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample132-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst (15/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,504B, BPFP=0.3987 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,432B, BPFP=2.2192 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,784B, BPFP=1.5667 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,772B, BPFP=2.3864 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,448B, BPFP=1.9918 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,748B, BPFP=2.4086 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,740B, BPFP=1.9985 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,916B, BPFP=2.3897 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,532B, BPFP=1.5610 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,324B, BPFP=2.4445 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 220,428B, BPFP=1.2552 -⌛️ [2/4] FRONTEND: Frontend time: 2.598s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.238s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02309080 6.43066193 - layer.0.v_cache 0.00000026 0.00013499 - layer.1.k_cache 0.00292563 0.78704176 - layer.1.v_cache 0.00000076 0.00050689 - layer.2.k_cache 0.00117950 0.34856837 - layer.2.v_cache 0.00000116 0.00071791 - layer.3.k_cache 0.00131686 0.40574188 - layer.3.v_cache 0.00000208 0.00114657 - layer.4.k_cache 0.00357410 0.73652854 - layer.4.v_cache 0.00000334 0.00201649 - layer.4.output 0.00016206 0.05867207 - ------------------------------------------------------------------------------------- - TOTAL 0.00233876 0.63912526 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 1070628 -BPFP 1.7418 bits/point -EBPFP 3.4837 equivalent bits/point -MSE 0.639125 ----------------------- -------------------------------------------------------- -Time: 4.852s Load: 0.016s, Pack+Encode: 2.598s, Decode+Unpack: 2.238s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6391 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample135-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst (16/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 359, 128) -Output shape: (1, 359, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.output: torch.Size([1, 359, 4096]) -> torch.Size([1, 1, 359, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,796B, BPFP=0.4090 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,056B, BPFP=2.1556 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 73,108B, BPFP=1.5910 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,956B, BPFP=2.3058 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,760B, BPFP=1.9316 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,232B, BPFP=2.3336 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,164B, BPFP=1.9404 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,188B, BPFP=2.3108 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 75,456B, BPFP=1.6421 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,828B, BPFP=2.3465 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 249,832B, BPFP=1.3592 -⌛️ [2/4] FRONTEND: Frontend time: 2.581s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.205s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02394788 6.62817995 - layer.0.v_cache 0.00000027 0.00013671 - layer.1.k_cache 0.00287336 0.74936780 - layer.1.v_cache 0.00000081 0.00053687 - layer.2.k_cache 0.00118844 0.34611713 - layer.2.v_cache 0.00000120 0.00076927 - layer.3.k_cache 0.00133454 0.41135188 - layer.3.v_cache 0.00000220 0.00121848 - layer.4.k_cache 0.00356152 0.72271176 - layer.4.v_cache 0.00000363 0.00214465 - layer.4.output 0.00015158 0.05708065 - ------------------------------------------------------------------------------------- - TOTAL 0.00239430 0.64934694 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 1121376 -BPFP 1.7431 bits/point -EBPFP 3.4862 equivalent bits/point -MSE 0.649347 ----------------------- -------------------------------------------------------- -Time: 4.808s Load: 0.022s, Pack+Encode: 2.581s, Decode+Unpack: 2.205s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6493 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample14-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst (17/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 359, 128) -Output shape: (1, 359, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.output: torch.Size([1, 359, 4096]) -> torch.Size([1, 1, 359, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,200B, BPFP=0.4178 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,400B, BPFP=2.1631 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,960B, BPFP=1.5877 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,328B, BPFP=2.2921 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,268B, BPFP=1.9426 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,044B, BPFP=2.3295 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,600B, BPFP=1.9499 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,864B, BPFP=2.3038 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 73,532B, BPFP=1.6002 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,840B, BPFP=2.3468 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 277,816B, BPFP=1.5114 -⌛️ [2/4] FRONTEND: Frontend time: 2.600s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02338462 6.67945288 - layer.0.v_cache 0.00000027 0.00014050 - layer.1.k_cache 0.00293092 0.83879111 - layer.1.v_cache 0.00000080 0.00052483 - layer.2.k_cache 0.00116855 0.33944292 - layer.2.v_cache 0.00000117 0.00074350 - layer.3.k_cache 0.00131991 0.41575580 - layer.3.v_cache 0.00000231 0.00125319 - layer.4.k_cache 0.00358756 0.71652162 - layer.4.v_cache 0.00000356 0.00216455 - layer.4.output 0.00016308 0.05732424 - ------------------------------------------------------------------------------------- - TOTAL 0.00236086 0.65886342 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 1147852 -BPFP 1.7842 bits/point -EBPFP 3.5685 equivalent bits/point -MSE 0.658863 ----------------------- -------------------------------------------------------- -Time: 4.899s Load: 0.018s, Pack+Encode: 2.600s, Decode+Unpack: 2.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6589 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample15-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst (18/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 347, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 347, 128) -Output shape: (1, 347, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.output: torch.Size([1, 347, 4096]) -> torch.Size([1, 1, 347, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,304B, BPFP=0.4121 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,600B, BPFP=2.2424 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,968B, BPFP=1.5978 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,916B, BPFP=2.3846 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,560B, BPFP=1.9939 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,708B, BPFP=2.4025 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,932B, BPFP=2.0023 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,604B, BPFP=2.3776 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,512B, BPFP=1.6326 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,108B, BPFP=2.4340 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 241,572B, BPFP=1.3597 -⌛️ [2/4] FRONTEND: Frontend time: 2.612s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.output: torch.Size([1, 347, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.262s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.output: torch.Size([1, 347, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02341881 6.78517032 - layer.0.v_cache 0.00000026 0.00013996 - layer.1.k_cache 0.00293388 0.82181718 - layer.1.v_cache 0.00000079 0.00052159 - layer.2.k_cache 0.00115427 0.35829077 - layer.2.v_cache 0.00000115 0.00072558 - layer.3.k_cache 0.00129772 0.40078933 - layer.3.v_cache 0.00000217 0.00117570 - layer.4.k_cache 0.00353122 0.69286274 - layer.4.v_cache 0.00000331 0.00209869 - layer.4.output 0.00014441 0.05909520 - ------------------------------------------------------------------------------------- - TOTAL 0.00235152 0.66428376 - (elements=4,974,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4974592 -Total Bytes 1106784 -BPFP 1.7799 bits/point -EBPFP 3.5598 equivalent bits/point -MSE 0.664284 ----------------------- -------------------------------------------------------- -Time: 4.892s Load: 0.018s, Pack+Encode: 2.612s, Decode+Unpack: 2.262s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 347, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6643 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample16-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst (19/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 337, 128) -Output shape: (1, 337, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.output: torch.Size([1, 337, 4096]) -> torch.Size([1, 1, 337, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,124B, BPFP=0.3970 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,248B, BPFP=2.2545 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,036B, BPFP=1.5541 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,668B, BPFP=2.3801 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 85,772B, BPFP=1.9884 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,492B, BPFP=2.4224 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,292B, BPFP=2.0236 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,032B, BPFP=2.4117 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,972B, BPFP=1.5989 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,544B, BPFP=2.4468 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 255,760B, BPFP=1.4823 -⌛️ [2/4] FRONTEND: Frontend time: 2.615s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.238s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02403486 6.85083950 - layer.0.v_cache 0.00000026 0.00014129 - layer.1.k_cache 0.00293885 0.83089875 - layer.1.v_cache 0.00000076 0.00046753 - layer.2.k_cache 0.00113696 0.35443867 - layer.2.v_cache 0.00000113 0.00067445 - layer.3.k_cache 0.00132482 0.43155844 - layer.3.v_cache 0.00000208 0.00112171 - layer.4.k_cache 0.00376518 0.80515577 - layer.4.v_cache 0.00000304 0.00194093 - layer.4.output 0.00014776 0.06166519 - ------------------------------------------------------------------------------------- - TOTAL 0.00241421 0.68027841 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 1095940 -BPFP 1.8148 bits/point -EBPFP 3.6295 equivalent bits/point -MSE 0.680278 ----------------------- -------------------------------------------------------- -Time: 4.869s Load: 0.016s, Pack+Encode: 2.615s, Decode+Unpack: 2.238s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6803 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample165-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst (20/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,148B, BPFP=0.4039 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,828B, BPFP=2.1997 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,520B, BPFP=1.5474 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,372B, BPFP=2.3231 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,608B, BPFP=1.9722 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,824B, BPFP=2.3554 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,180B, BPFP=1.9627 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,732B, BPFP=2.3311 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,892B, BPFP=1.6224 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,776B, BPFP=2.3766 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 239,008B, BPFP=1.3300 -⌛️ [2/4] FRONTEND: Frontend time: 2.547s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.316s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02526469 6.84691061 - layer.0.v_cache 0.00000026 0.00014036 - layer.1.k_cache 0.00294286 0.83994704 - layer.1.v_cache 0.00000076 0.00050338 - layer.2.k_cache 0.00116394 0.37475634 - layer.2.v_cache 0.00000112 0.00071548 - layer.3.k_cache 0.00133268 0.41057115 - layer.3.v_cache 0.00000207 0.00114856 - layer.4.k_cache 0.00360177 0.77582187 - layer.4.v_cache 0.00000315 0.00202826 - layer.4.output 0.00019540 0.06072864 - ------------------------------------------------------------------------------------- - TOTAL 0.00250678 0.67824697 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 1096888 -BPFP 1.7439 bits/point -EBPFP 3.4878 equivalent bits/point -MSE 0.678247 ----------------------- -------------------------------------------------------- -Time: 4.879s Load: 0.016s, Pack+Encode: 2.547s, Decode+Unpack: 2.316s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6782 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample17-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst (21/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 346, 128) -Output shape: (1, 346, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.output: torch.Size([1, 346, 4096]) -> torch.Size([1, 1, 346, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,172B, BPFP=0.4103 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,488B, BPFP=2.2464 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,044B, BPFP=1.5816 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,496B, BPFP=2.3820 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,056B, BPFP=1.9883 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,568B, BPFP=2.4288 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,024B, BPFP=1.9875 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,300B, BPFP=2.4002 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,572B, BPFP=1.6386 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,096B, BPFP=2.4408 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 264,636B, BPFP=1.4938 -⌛️ [2/4] FRONTEND: Frontend time: 2.612s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.189s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02293222 6.84492131 - layer.0.v_cache 0.00000026 0.00013595 - layer.1.k_cache 0.00295457 0.75815119 - layer.1.v_cache 0.00000081 0.00052799 - layer.2.k_cache 0.00115565 0.35100412 - layer.2.v_cache 0.00000119 0.00075605 - layer.3.k_cache 0.00131728 0.39150331 - layer.3.v_cache 0.00000219 0.00119522 - layer.4.k_cache 0.00358690 0.71667009 - layer.4.v_cache 0.00000349 0.00208636 - layer.4.output 0.00015659 0.05478282 - ------------------------------------------------------------------------------------- - TOTAL 0.00232721 0.66329163 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 1128452 -BPFP 1.8200 bits/point -EBPFP 3.6400 equivalent bits/point -MSE 0.663292 ----------------------- -------------------------------------------------------- -Time: 4.818s Load: 0.017s, Pack+Encode: 2.612s, Decode+Unpack: 2.189s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6633 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample18-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst (22/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 340, 128) -Output shape: (1, 340, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.output: torch.Size([1, 340, 4096]) -> torch.Size([1, 1, 340, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,244B, BPFP=0.4192 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,152B, BPFP=2.2553 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,984B, BPFP=1.6081 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,784B, BPFP=2.4077 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,044B, BPFP=2.0001 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,500B, BPFP=2.4472 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,820B, BPFP=2.0179 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,880B, BPFP=2.4099 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 71,828B, BPFP=1.6505 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,772B, BPFP=2.4534 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 245,488B, BPFP=1.4102 -⌛️ [2/4] FRONTEND: Frontend time: 2.580s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.172s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02361420 6.74267865 - layer.0.v_cache 0.00000026 0.00014070 - layer.1.k_cache 0.00281368 0.76992448 - layer.1.v_cache 0.00000080 0.00052342 - layer.2.k_cache 0.00120688 0.35270009 - layer.2.v_cache 0.00000119 0.00076861 - layer.3.k_cache 0.00132000 0.40094497 - layer.3.v_cache 0.00000216 0.00121101 - layer.4.k_cache 0.00354241 0.70096382 - layer.4.v_cache 0.00000378 0.00216076 - layer.4.output 0.00015030 0.05786711 - ------------------------------------------------------------------------------------- - TOTAL 0.00236476 0.65739178 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 1101496 -BPFP 1.8079 bits/point -EBPFP 3.6157 equivalent bits/point -MSE 0.657392 ----------------------- -------------------------------------------------------- -Time: 4.768s Load: 0.016s, Pack+Encode: 2.580s, Decode+Unpack: 2.172s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6574 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample19-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst (23/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,636B, BPFP=0.4232 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,496B, BPFP=2.2369 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,888B, BPFP=1.5872 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,388B, BPFP=2.3934 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,012B, BPFP=1.9761 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,408B, BPFP=2.4166 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,400B, BPFP=2.0076 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,920B, BPFP=2.4055 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 70,472B, BPFP=1.6005 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,600B, BPFP=2.4437 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 267,036B, BPFP=1.5161 -⌛️ [2/4] FRONTEND: Frontend time: 2.583s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.182s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02321494 7.04644846 - layer.0.v_cache 0.00000027 0.00013625 - layer.1.k_cache 0.00289516 0.82582385 - layer.1.v_cache 0.00000077 0.00049612 - layer.2.k_cache 0.00116570 0.34631991 - layer.2.v_cache 0.00000112 0.00072180 - layer.3.k_cache 0.00132982 0.39844766 - layer.3.v_cache 0.00000218 0.00119210 - layer.4.k_cache 0.00351457 0.70288702 - layer.4.v_cache 0.00000367 0.00217221 - layer.4.output 0.00015182 0.05643834 - ------------------------------------------------------------------------------------- - TOTAL 0.00233825 0.68217134 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 1125256 -BPFP 1.8254 bits/point -EBPFP 3.6508 equivalent bits/point -MSE 0.682171 ----------------------- -------------------------------------------------------- -Time: 4.783s Load: 0.018s, Pack+Encode: 2.583s, Decode+Unpack: 2.182s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6822 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample2-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst (24/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 337, 128) -Output shape: (1, 337, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.output: torch.Size([1, 337, 4096]) -> torch.Size([1, 1, 337, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,800B, BPFP=0.4126 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,900B, BPFP=2.2696 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,524B, BPFP=1.5422 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,628B, BPFP=2.4024 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,168B, BPFP=1.9976 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,864B, BPFP=2.4542 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,916B, BPFP=2.0149 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,464B, BPFP=2.4217 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,452B, BPFP=1.5637 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,240B, BPFP=2.4629 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 229,344B, BPFP=1.3292 -⌛️ [2/4] FRONTEND: Frontend time: 2.580s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.237s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435875 7.23912741 - layer.0.v_cache 0.00000026 0.00014082 - layer.1.k_cache 0.00286757 0.84235851 - layer.1.v_cache 0.00000078 0.00051906 - layer.2.k_cache 0.00117745 0.35202203 - layer.2.v_cache 0.00000112 0.00071784 - layer.3.k_cache 0.00130673 0.40256367 - layer.3.v_cache 0.00000209 0.00115574 - layer.4.k_cache 0.00356837 0.72063344 - layer.4.v_cache 0.00000322 0.00205395 - layer.4.output 0.00014638 0.05569294 - ------------------------------------------------------------------------------------- - TOTAL 0.00241942 0.69886173 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 1072300 -BPFP 1.7756 bits/point -EBPFP 3.5512 equivalent bits/point -MSE 0.698862 ----------------------- -------------------------------------------------------- -Time: 4.833s Load: 0.016s, Pack+Encode: 2.580s, Decode+Unpack: 2.237s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6989 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample20-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst (25/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 395, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.025s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 395, 128) -Output shape: (1, 395, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.output: torch.Size([1, 395, 4096]) -> torch.Size([1, 1, 395, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 20,480B, BPFP=0.4051 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 113,504B, BPFP=2.2449 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 78,720B, BPFP=1.5570 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 119,640B, BPFP=2.3663 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 96,348B, BPFP=1.9056 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 120,872B, BPFP=2.3907 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 100,688B, BPFP=1.9915 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 120,456B, BPFP=2.3824 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 78,428B, BPFP=1.5512 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 122,104B, BPFP=2.4150 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 301,528B, BPFP=1.4909 -⌛️ [2/4] FRONTEND: Frontend time: 2.971s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.output: torch.Size([1, 395, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.437s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.output: torch.Size([1, 395, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02333523 6.98126174 - layer.0.v_cache 0.00000027 0.00014295 - layer.1.k_cache 0.00291865 0.88639758 - layer.1.v_cache 0.00000080 0.00050972 - layer.2.k_cache 0.00114101 0.36095616 - layer.2.v_cache 0.00000113 0.00069460 - layer.3.k_cache 0.00133612 0.41124978 - layer.3.v_cache 0.00000221 0.00114963 - layer.4.k_cache 0.00365959 0.77239774 - layer.4.v_cache 0.00000309 0.00197135 - layer.4.output 0.00017803 0.06155861 - ------------------------------------------------------------------------------------- - TOTAL 0.00236502 0.69021184 - (elements=5,662,720) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5662720 -Total Bytes 1272768 -BPFP 1.7981 bits/point -EBPFP 3.5962 equivalent bits/point -MSE 0.690212 ----------------------- -------------------------------------------------------- -Time: 5.433s Load: 0.025s, Pack+Encode: 2.971s, Decode+Unpack: 2.437s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 395, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6902 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample21-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst (26/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 348, 128) -Output shape: (1, 348, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.output: torch.Size([1, 348, 4096]) -> torch.Size([1, 1, 348, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,192B, BPFP=0.4084 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,676B, BPFP=2.2377 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,624B, BPFP=1.5855 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,200B, BPFP=2.3617 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,384B, BPFP=2.0066 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,676B, BPFP=2.3948 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,012B, BPFP=1.9983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,476B, BPFP=2.3679 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,336B, BPFP=1.6239 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,680B, BPFP=2.4174 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 224,704B, BPFP=1.2611 -⌛️ [2/4] FRONTEND: Frontend time: 2.568s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.226s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02579716 7.05841766 - layer.0.v_cache 0.00000026 0.00014191 - layer.1.k_cache 0.00289751 0.75264968 - layer.1.v_cache 0.00000077 0.00051624 - layer.2.k_cache 0.00114660 0.35081530 - layer.2.v_cache 0.00000117 0.00074238 - layer.3.k_cache 0.00133042 0.40999156 - layer.3.v_cache 0.00000210 0.00115967 - layer.4.k_cache 0.00358099 0.74246076 - layer.4.v_cache 0.00000335 0.00210123 - layer.4.output 0.00013932 0.05784966 - ------------------------------------------------------------------------------------- - TOTAL 0.00252269 0.68217107 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 1088960 -BPFP 1.7462 bits/point -EBPFP 3.4924 equivalent bits/point -MSE 0.682171 ----------------------- -------------------------------------------------------- -Time: 4.813s Load: 0.019s, Pack+Encode: 2.568s, Decode+Unpack: 2.226s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6822 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample22-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst (27/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 316, 128) -Output shape: (1, 316, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.output: torch.Size([1, 316, 4096]) -> torch.Size([1, 1, 316, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,148B, BPFP=0.4240 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,084B, BPFP=2.0541 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,840B, BPFP=1.5289 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,204B, BPFP=2.1807 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,724B, BPFP=1.8474 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,336B, BPFP=2.2087 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,404B, BPFP=1.8395 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,776B, BPFP=2.1948 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,472B, BPFP=1.5445 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,192B, BPFP=2.2298 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 230,604B, BPFP=1.4253 -⌛️ [2/4] FRONTEND: Frontend time: 2.386s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.916s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02512236 6.84216618 - layer.0.v_cache 0.00000027 0.00013801 - layer.1.k_cache 0.00293329 0.74739191 - layer.1.v_cache 0.00000080 0.00049147 - layer.2.k_cache 0.00117467 0.36838020 - layer.2.v_cache 0.00000111 0.00070566 - layer.3.k_cache 0.00131319 0.43508042 - layer.3.v_cache 0.00000216 0.00116272 - layer.4.k_cache 0.00359311 0.75364753 - layer.4.v_cache 0.00000344 0.00202547 - layer.4.output 0.00015753 0.05683376 - ------------------------------------------------------------------------------------- - TOTAL 0.00248390 0.66989461 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 960784 -BPFP 1.6967 bits/point -EBPFP 3.3934 equivalent bits/point -MSE 0.669895 ----------------------- -------------------------------------------------------- -Time: 4.319s Load: 0.017s, Pack+Encode: 2.386s, Decode+Unpack: 1.916s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6699 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample23-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst (28/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 348, 128) -Output shape: (1, 348, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.output: torch.Size([1, 348, 4096]) -> torch.Size([1, 1, 348, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,164B, BPFP=0.4078 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,208B, BPFP=2.2272 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,924B, BPFP=1.5698 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,528B, BPFP=2.3691 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,124B, BPFP=2.0008 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,384B, BPFP=2.4107 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,084B, BPFP=1.9999 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,084B, BPFP=2.3816 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,100B, BPFP=1.6186 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,996B, BPFP=2.4245 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 249,988B, BPFP=1.4030 -⌛️ [2/4] FRONTEND: Frontend time: 2.600s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.170s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02368600 7.05519472 - layer.0.v_cache 0.00000027 0.00014080 - layer.1.k_cache 0.00288004 0.77680452 - layer.1.v_cache 0.00000079 0.00052235 - layer.2.k_cache 0.00117070 0.35410820 - layer.2.v_cache 0.00000116 0.00076144 - layer.3.k_cache 0.00131678 0.41304626 - layer.3.v_cache 0.00000214 0.00120927 - layer.4.k_cache 0.00357349 0.72397973 - layer.4.v_cache 0.00000347 0.00210801 - layer.4.output 0.00013697 0.05610911 - ------------------------------------------------------------------------------------- - TOTAL 0.00237019 0.68230798 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 1114584 -BPFP 1.7873 bits/point -EBPFP 3.5746 equivalent bits/point -MSE 0.682308 ----------------------- -------------------------------------------------------- -Time: 4.788s Load: 0.018s, Pack+Encode: 2.600s, Decode+Unpack: 2.170s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6823 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample24-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst (29/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.020s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 345, 128) -Output shape: (1, 345, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.output: torch.Size([1, 345, 4096]) -> torch.Size([1, 1, 345, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,864B, BPFP=0.4045 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,844B, BPFP=2.2383 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,336B, BPFP=1.5701 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,968B, BPFP=2.3996 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,060B, BPFP=1.9941 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,336B, BPFP=2.4306 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,712B, BPFP=2.0089 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,260B, BPFP=2.4062 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 70,740B, BPFP=1.6019 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,028B, BPFP=2.4463 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 246,980B, BPFP=1.3982 -⌛️ [2/4] FRONTEND: Frontend time: 2.651s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.261s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02357864 6.82285015 - layer.0.v_cache 0.00000026 0.00013668 - layer.1.k_cache 0.00288937 0.76303163 - layer.1.v_cache 0.00000077 0.00051062 - layer.2.k_cache 0.00116899 0.34020698 - layer.2.v_cache 0.00000115 0.00074744 - layer.3.k_cache 0.00131620 0.39873330 - layer.3.v_cache 0.00000216 0.00119220 - layer.4.k_cache 0.00353981 0.72173727 - layer.4.v_cache 0.00000338 0.00211140 - layer.4.output 0.00014567 0.05609593 - ------------------------------------------------------------------------------------- - TOTAL 0.00236310 0.66254581 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 1108128 -BPFP 1.7924 bits/point -EBPFP 3.5848 equivalent bits/point -MSE 0.662546 ----------------------- -------------------------------------------------------- -Time: 4.932s Load: 0.020s, Pack+Encode: 2.651s, Decode+Unpack: 2.261s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6625 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample25-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst (30/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,164B, BPFP=0.4300 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,520B, BPFP=2.3087 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,396B, BPFP=1.5719 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,052B, BPFP=2.4634 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,236B, BPFP=1.9705 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,808B, BPFP=2.4813 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,136B, BPFP=2.0155 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,636B, BPFP=2.4772 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,968B, BPFP=1.6091 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,756B, BPFP=2.5037 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 271,316B, BPFP=1.6058 -⌛️ [2/4] FRONTEND: Frontend time: 2.571s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.224s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02434325 6.89808978 - layer.0.v_cache 0.00000027 0.00014289 - layer.1.k_cache 0.00287412 0.85958511 - layer.1.v_cache 0.00000081 0.00051648 - layer.2.k_cache 0.00119273 0.34149602 - layer.2.v_cache 0.00000117 0.00075593 - layer.3.k_cache 0.00130297 0.38633182 - layer.3.v_cache 0.00000221 0.00121707 - layer.4.k_cache 0.00352071 0.69310978 - layer.4.v_cache 0.00000330 0.00215976 - layer.4.output 0.00014370 0.05225649 - ------------------------------------------------------------------------------------- - TOTAL 0.00241545 0.67088790 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 1108988 -BPFP 1.8753 bits/point -EBPFP 3.7506 equivalent bits/point -MSE 0.670888 ----------------------- -------------------------------------------------------- -Time: 4.813s Load: 0.018s, Pack+Encode: 2.571s, Decode+Unpack: 2.224s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6709 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample26-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst (31/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 334, 128) -Output shape: (1, 334, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.output: torch.Size([1, 334, 4096]) -> torch.Size([1, 1, 334, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,204B, BPFP=0.4258 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,300B, BPFP=2.2759 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,340B, BPFP=1.5751 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,184B, BPFP=2.4135 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,460B, BPFP=1.9756 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,324B, BPFP=2.4402 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,284B, BPFP=2.0182 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,560B, BPFP=2.4457 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,688B, BPFP=1.6067 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,672B, BPFP=2.4717 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 267,836B, BPFP=1.5662 -⌛️ [2/4] FRONTEND: Frontend time: 2.535s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.226s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02363260 7.02747532 - layer.0.v_cache 0.00000026 0.00014066 - layer.1.k_cache 0.00290221 0.83520910 - layer.1.v_cache 0.00000079 0.00051332 - layer.2.k_cache 0.00116170 0.36400449 - layer.2.v_cache 0.00000115 0.00074367 - layer.3.k_cache 0.00132769 0.40021867 - layer.3.v_cache 0.00000211 0.00116973 - layer.4.k_cache 0.00362300 0.72351367 - layer.4.v_cache 0.00000352 0.00208218 - layer.4.output 0.00014243 0.06096746 - ------------------------------------------------------------------------------------- - TOTAL 0.00237320 0.68563862 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 1107852 -BPFP 1.8510 bits/point -EBPFP 3.7019 equivalent bits/point -MSE 0.685639 ----------------------- -------------------------------------------------------- -Time: 4.778s Load: 0.017s, Pack+Encode: 2.535s, Decode+Unpack: 2.226s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6856 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample27-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst (32/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,912B, BPFP=0.4241 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,948B, BPFP=2.2952 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,440B, BPFP=1.5729 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,616B, BPFP=2.4530 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 85,476B, BPFP=2.0236 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,756B, BPFP=2.4563 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,576B, BPFP=2.0259 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,964B, BPFP=2.4613 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,404B, BPFP=1.5721 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,396B, BPFP=2.4952 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 275,280B, BPFP=1.6293 -⌛️ [2/4] FRONTEND: Frontend time: 2.571s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.248s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02410887 6.96771351 - layer.0.v_cache 0.00000026 0.00014345 - layer.1.k_cache 0.00288475 0.82714853 - layer.1.v_cache 0.00000079 0.00051153 - layer.2.k_cache 0.00119186 0.34004332 - layer.2.v_cache 0.00000115 0.00073878 - layer.3.k_cache 0.00132245 0.40396576 - layer.3.v_cache 0.00000226 0.00122942 - layer.4.k_cache 0.00352035 0.69196856 - layer.4.v_cache 0.00000339 0.00207642 - layer.4.output 0.00014978 0.05928616 - ------------------------------------------------------------------------------------- - TOTAL 0.00240252 0.67662028 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 1110768 -BPFP 1.8783 bits/point -EBPFP 3.7567 equivalent bits/point -MSE 0.676620 ----------------------- -------------------------------------------------------- -Time: 4.841s Load: 0.022s, Pack+Encode: 2.571s, Decode+Unpack: 2.248s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6766 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample28-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst (33/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,144B, BPFP=0.4309 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,584B, BPFP=2.3172 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,004B, BPFP=1.5673 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,488B, BPFP=2.4574 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,832B, BPFP=1.9907 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,332B, BPFP=2.4775 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 84,508B, BPFP=2.0067 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,896B, BPFP=2.4671 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,840B, BPFP=1.5397 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,124B, BPFP=2.5200 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 264,672B, BPFP=1.5712 -⌛️ [2/4] FRONTEND: Frontend time: 2.582s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.288s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02426959 6.99855742 - layer.0.v_cache 0.00000027 0.00014094 - layer.1.k_cache 0.00291730 0.88080261 - layer.1.v_cache 0.00000079 0.00050005 - layer.2.k_cache 0.00116725 0.35450167 - layer.2.v_cache 0.00000113 0.00071310 - layer.3.k_cache 0.00130327 0.39310319 - layer.3.v_cache 0.00000218 0.00116465 - layer.4.k_cache 0.00359671 0.70075738 - layer.4.v_cache 0.00000327 0.00211203 - layer.4.output 0.00014627 0.05523620 - ------------------------------------------------------------------------------------- - TOTAL 0.00241763 0.68237842 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 1097424 -BPFP 1.8614 bits/point -EBPFP 3.7228 equivalent bits/point -MSE 0.682378 ----------------------- -------------------------------------------------------- -Time: 4.888s Load: 0.017s, Pack+Encode: 2.582s, Decode+Unpack: 2.288s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6824 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample29-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst (34/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 365, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.026s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 365, 128) -Output shape: (1, 365, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.output: torch.Size([1, 365, 4096]) -> torch.Size([1, 1, 365, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,476B, BPFP=0.4169 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,820B, BPFP=2.1366 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 73,084B, BPFP=1.5643 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 106,244B, BPFP=2.2741 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,760B, BPFP=1.9212 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 108,056B, BPFP=2.3128 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,736B, BPFP=1.9207 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,788B, BPFP=2.2857 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,980B, BPFP=1.6049 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,752B, BPFP=2.3277 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 293,352B, BPFP=1.5697 -⌛️ [2/4] FRONTEND: Frontend time: 2.608s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.output: torch.Size([1, 365, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.232s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.output: torch.Size([1, 365, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02487199 7.12003692 - layer.0.v_cache 0.00000027 0.00013958 - layer.1.k_cache 0.00290279 0.72563560 - layer.1.v_cache 0.00000079 0.00052066 - layer.2.k_cache 0.00116885 0.35597614 - layer.2.v_cache 0.00000114 0.00075773 - layer.3.k_cache 0.00131452 0.41987622 - layer.3.v_cache 0.00000216 0.00120868 - layer.4.k_cache 0.00360833 0.72024386 - layer.4.v_cache 0.00000353 0.00209684 - layer.4.output 0.00014951 0.05512628 - ------------------------------------------------------------------------------------- - TOTAL 0.00246232 0.68335695 - (elements=5,232,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5232640 -Total Bytes 1170048 -BPFP 1.7888 bits/point -EBPFP 3.5777 equivalent bits/point -MSE 0.683357 ----------------------- -------------------------------------------------------- -Time: 4.867s Load: 0.026s, Pack+Encode: 2.608s, Decode+Unpack: 2.232s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 365, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6834 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample3-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst (35/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.023s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,468B, BPFP=0.4074 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,108B, BPFP=2.2880 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,240B, BPFP=1.5914 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,604B, BPFP=2.4395 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,508B, BPFP=2.0174 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,484B, BPFP=2.4367 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,292B, BPFP=2.0124 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,176B, BPFP=2.4295 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,988B, BPFP=1.6089 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,792B, BPFP=2.4672 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 243,056B, BPFP=1.4171 -⌛️ [2/4] FRONTEND: Frontend time: 2.593s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.300s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02462482 6.65592569 - layer.0.v_cache 0.00000027 0.00014046 - layer.1.k_cache 0.00287179 0.75542625 - layer.1.v_cache 0.00000079 0.00052709 - layer.2.k_cache 0.00118919 0.35170826 - layer.2.v_cache 0.00000114 0.00074619 - layer.3.k_cache 0.00130718 0.39530435 - layer.3.v_cache 0.00000215 0.00119536 - layer.4.k_cache 0.00355403 0.69238996 - layer.4.v_cache 0.00000335 0.00208779 - layer.4.output 0.00013864 0.05722230 - ------------------------------------------------------------------------------------- - TOTAL 0.00243638 0.64888147 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 1087716 -BPFP 1.8119 bits/point -EBPFP 3.6238 equivalent bits/point -MSE 0.648881 ----------------------- -------------------------------------------------------- -Time: 4.916s Load: 0.023s, Pack+Encode: 2.593s, Decode+Unpack: 2.300s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6489 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample30-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst (36/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,800B, BPFP=0.4151 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,220B, BPFP=2.2673 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,468B, BPFP=1.5734 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,032B, BPFP=2.4028 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 85,692B, BPFP=1.9984 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,500B, BPFP=2.4370 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,244B, BPFP=2.0113 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,920B, BPFP=2.4468 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,200B, BPFP=1.5672 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,700B, BPFP=2.4650 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 248,852B, BPFP=1.4509 -⌛️ [2/4] FRONTEND: Frontend time: 2.573s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.225s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02394044 6.56193082 - layer.0.v_cache 0.00000026 0.00013767 - layer.1.k_cache 0.00289851 0.80702314 - layer.1.v_cache 0.00000079 0.00051333 - layer.2.k_cache 0.00115963 0.34447199 - layer.2.v_cache 0.00000116 0.00076297 - layer.3.k_cache 0.00131431 0.39645454 - layer.3.v_cache 0.00000219 0.00122058 - layer.4.k_cache 0.00369786 0.74114585 - layer.4.v_cache 0.00000335 0.00209741 - layer.4.output 0.00014933 0.05661584 - ------------------------------------------------------------------------------------- - TOTAL 0.00240113 0.64873012 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 1088628 -BPFP 1.8134 bits/point -EBPFP 3.6268 equivalent bits/point -MSE 0.648730 ----------------------- -------------------------------------------------------- -Time: 4.819s Load: 0.022s, Pack+Encode: 2.573s, Decode+Unpack: 2.225s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6487 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample31-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst (37/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,088B, BPFP=0.4169 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,320B, BPFP=2.2659 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,472B, BPFP=1.5780 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,148B, BPFP=2.4002 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 85,876B, BPFP=1.9791 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,376B, BPFP=2.4515 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,792B, BPFP=2.0232 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,956B, BPFP=2.4188 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,148B, BPFP=1.5936 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,108B, BPFP=2.4684 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 255,464B, BPFP=1.4718 -⌛️ [2/4] FRONTEND: Frontend time: 2.565s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.251s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02382689 6.77071018 - layer.0.v_cache 0.00000027 0.00014293 - layer.1.k_cache 0.00292607 0.81484553 - layer.1.v_cache 0.00000081 0.00051123 - layer.2.k_cache 0.00125336 0.35761378 - layer.2.v_cache 0.00000116 0.00075133 - layer.3.k_cache 0.00131849 0.39801588 - layer.3.v_cache 0.00000220 0.00121379 - layer.4.k_cache 0.00356446 0.72492765 - layer.4.v_cache 0.00000343 0.00210048 - layer.4.output 0.00016600 0.06054505 - ------------------------------------------------------------------------------------- - TOTAL 0.00239722 0.66521521 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 1105748 -BPFP 1.8202 bits/point -EBPFP 3.6404 equivalent bits/point -MSE 0.665215 ----------------------- -------------------------------------------------------- -Time: 4.834s Load: 0.018s, Pack+Encode: 2.565s, Decode+Unpack: 2.251s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6652 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample32-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst (38/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.025s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 362, 128) -Output shape: (1, 362, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.output: torch.Size([1, 362, 4096]) -> torch.Size([1, 1, 362, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,764B, BPFP=0.4050 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,636B, BPFP=2.1503 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 74,208B, BPFP=1.6015 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,968B, BPFP=2.2869 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 90,060B, BPFP=1.9436 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,484B, BPFP=2.3197 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,036B, BPFP=1.9215 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,144B, BPFP=2.2907 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 75,620B, BPFP=1.6320 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,356B, BPFP=2.3385 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 244,164B, BPFP=1.3174 -⌛️ [2/4] FRONTEND: Frontend time: 2.598s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.300s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02338883 6.96582368 - layer.0.v_cache 0.00000028 0.00014366 - layer.1.k_cache 0.00287711 0.79265607 - layer.1.v_cache 0.00000077 0.00051860 - layer.2.k_cache 0.00119138 0.35131977 - layer.2.v_cache 0.00000112 0.00073930 - layer.3.k_cache 0.00131510 0.39147148 - layer.3.v_cache 0.00000211 0.00117857 - layer.4.k_cache 0.00352384 0.73666778 - layer.4.v_cache 0.00000335 0.00210012 - layer.4.output 0.00013081 0.05654985 - ------------------------------------------------------------------------------------- - TOTAL 0.00234480 0.67634417 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 1119440 -BPFP 1.7257 bits/point -EBPFP 3.4513 equivalent bits/point -MSE 0.676344 ----------------------- -------------------------------------------------------- -Time: 4.923s Load: 0.025s, Pack+Encode: 2.598s, Decode+Unpack: 2.300s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6763 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample33-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst (39/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 318, 128) -Output shape: (1, 318, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.output: torch.Size([1, 318, 4096]) -> torch.Size([1, 1, 318, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,196B, BPFP=0.4225 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,804B, BPFP=2.0589 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,984B, BPFP=1.4982 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 89,360B, BPFP=2.1954 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,664B, BPFP=1.8343 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 90,376B, BPFP=2.2203 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 75,156B, BPFP=1.8464 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,732B, BPFP=2.2045 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,644B, BPFP=1.5881 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,876B, BPFP=2.2326 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 265,432B, BPFP=1.6303 -⌛️ [2/4] FRONTEND: Frontend time: 2.328s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.910s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02477463 6.87928858 - layer.0.v_cache 0.00000027 0.00014032 - layer.1.k_cache 0.00289381 0.89942414 - layer.1.v_cache 0.00000083 0.00051824 - layer.2.k_cache 0.00119129 0.36182440 - layer.2.v_cache 0.00000117 0.00075193 - layer.3.k_cache 0.00131631 0.44791379 - layer.3.v_cache 0.00000257 0.00123300 - layer.4.k_cache 0.00356231 0.72936801 - layer.4.v_cache 0.00000346 0.00211839 - layer.4.output 0.00015646 0.05874466 - ------------------------------------------------------------------------------------- - TOTAL 0.00245518 0.68268282 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 1002224 -BPFP 1.7587 bits/point -EBPFP 3.5175 equivalent bits/point -MSE 0.682683 ----------------------- -------------------------------------------------------- -Time: 4.254s Load: 0.016s, Pack+Encode: 2.328s, Decode+Unpack: 1.910s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6827 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample34-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst (40/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,940B, BPFP=0.4098 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,660B, BPFP=2.2537 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,948B, BPFP=1.5750 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,564B, BPFP=2.3886 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,432B, BPFP=1.9744 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,856B, BPFP=2.4181 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,168B, BPFP=2.0141 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,852B, BPFP=2.4180 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,472B, BPFP=1.5870 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,052B, BPFP=2.4454 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 250,228B, BPFP=1.4290 -⌛️ [2/4] FRONTEND: Frontend time: 2.597s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.179s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02487079 6.57781733 - layer.0.v_cache 0.00000026 0.00013831 - layer.1.k_cache 0.00288341 0.80898334 - layer.1.v_cache 0.00000082 0.00051752 - layer.2.k_cache 0.00117177 0.34432854 - layer.2.v_cache 0.00000117 0.00074610 - layer.3.k_cache 0.00130255 0.40242500 - layer.3.v_cache 0.00000215 0.00118705 - layer.4.k_cache 0.00351771 0.71550822 - layer.4.v_cache 0.00000337 0.00205636 - layer.4.output 0.00013439 0.05135101 - ------------------------------------------------------------------------------------- - TOTAL 0.00244940 0.64707941 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 1103172 -BPFP 1.8000 bits/point -EBPFP 3.6001 equivalent bits/point -MSE 0.647079 ----------------------- -------------------------------------------------------- -Time: 4.793s Load: 0.017s, Pack+Encode: 2.597s, Decode+Unpack: 2.179s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6471 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample35-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst (41/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.020s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,392B, BPFP=0.4189 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,960B, BPFP=2.2540 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,792B, BPFP=1.5897 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,796B, BPFP=2.4097 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,184B, BPFP=1.9858 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,568B, BPFP=2.4273 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,472B, BPFP=1.9923 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,208B, BPFP=2.4191 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,596B, BPFP=1.5852 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,052B, BPFP=2.4611 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 251,112B, BPFP=1.4299 -⌛️ [2/4] FRONTEND: Frontend time: 2.648s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.138s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02787456 6.85308571 - layer.0.v_cache 0.00000027 0.00014080 - layer.1.k_cache 0.00288980 0.76633206 - layer.1.v_cache 0.00000078 0.00052103 - layer.2.k_cache 0.00118800 0.34666278 - layer.2.v_cache 0.00000116 0.00074679 - layer.3.k_cache 0.00131625 0.37887369 - layer.3.v_cache 0.00000215 0.00120600 - layer.4.k_cache 0.00354145 0.70910987 - layer.4.v_cache 0.00000355 0.00210977 - layer.4.output 0.00016614 0.06084832 - ------------------------------------------------------------------------------------- - TOTAL 0.00267732 0.66444156 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 1109132 -BPFP 1.8045 bits/point -EBPFP 3.6090 equivalent bits/point -MSE 0.664442 ----------------------- -------------------------------------------------------- -Time: 4.805s Load: 0.020s, Pack+Encode: 2.648s, Decode+Unpack: 2.138s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6644 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample36-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst (42/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.015s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,804B, BPFP=0.4177 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,056B, BPFP=2.3005 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,388B, BPFP=1.5575 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,264B, BPFP=2.4227 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,244B, BPFP=1.9764 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,712B, BPFP=2.4566 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,224B, BPFP=2.0229 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,492B, BPFP=2.4515 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,688B, BPFP=1.5646 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,088B, BPFP=2.4889 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 229,048B, BPFP=1.3434 -⌛️ [2/4] FRONTEND: Frontend time: 2.528s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.219s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02325115 7.01752900 - layer.0.v_cache 0.00000026 0.00014150 - layer.1.k_cache 0.00292822 0.81960968 - layer.1.v_cache 0.00000083 0.00052307 - layer.2.k_cache 0.00115284 0.35103261 - layer.2.v_cache 0.00000118 0.00074563 - layer.3.k_cache 0.00132958 0.40356606 - layer.3.v_cache 0.00000212 0.00119047 - layer.4.k_cache 0.00351398 0.68624447 - layer.4.v_cache 0.00000364 0.00210129 - layer.4.output 0.00014544 0.06194126 - ------------------------------------------------------------------------------------- - TOTAL 0.00234040 0.68074634 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 1067008 -BPFP 1.7881 bits/point -EBPFP 3.5761 equivalent bits/point -MSE 0.680746 ----------------------- -------------------------------------------------------- -Time: 4.762s Load: 0.015s, Pack+Encode: 2.528s, Decode+Unpack: 2.219s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6807 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample37-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst (43/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 356, 128) -Output shape: (1, 356, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.output: torch.Size([1, 356, 4096]) -> torch.Size([1, 1, 356, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,336B, BPFP=0.4024 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,600B, BPFP=2.1638 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,232B, BPFP=1.5851 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,488B, BPFP=2.2930 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,496B, BPFP=1.9421 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,392B, BPFP=2.3348 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,108B, BPFP=1.9555 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,556B, BPFP=2.3165 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,148B, BPFP=1.6272 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,564B, BPFP=2.3605 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 273,720B, BPFP=1.5017 -⌛️ [2/4] FRONTEND: Frontend time: 2.610s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.204s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02485108 6.91562815 - layer.0.v_cache 0.00000027 0.00013960 - layer.1.k_cache 0.00289550 0.80236448 - layer.1.v_cache 0.00000079 0.00051589 - layer.2.k_cache 0.00117641 0.35817798 - layer.2.v_cache 0.00000113 0.00075091 - layer.3.k_cache 0.00131994 0.42152872 - layer.3.v_cache 0.00000217 0.00121762 - layer.4.k_cache 0.00362269 0.73566840 - layer.4.v_cache 0.00000343 0.00215265 - layer.4.output 0.00016041 0.05536736 - ------------------------------------------------------------------------------------- - TOTAL 0.00246536 0.67568670 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 1138640 -BPFP 1.7848 bits/point -EBPFP 3.5697 equivalent bits/point -MSE 0.675687 ----------------------- -------------------------------------------------------- -Time: 4.831s Load: 0.017s, Pack+Encode: 2.610s, Decode+Unpack: 2.204s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6757 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample38-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst (44/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 331, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 331, 128) -Output shape: (1, 331, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.output: torch.Size([1, 331, 4096]) -> torch.Size([1, 1, 331, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,852B, BPFP=0.4214 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,816B, BPFP=2.2851 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,532B, BPFP=1.6175 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,884B, BPFP=2.4519 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,044B, BPFP=1.9837 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,356B, BPFP=2.4631 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,452B, BPFP=2.0405 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,100B, BPFP=2.4570 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,624B, BPFP=1.5961 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,616B, BPFP=2.4692 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 249,248B, BPFP=1.4707 -⌛️ [2/4] FRONTEND: Frontend time: 2.557s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.output: torch.Size([1, 331, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.263s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.output: torch.Size([1, 331, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02326727 6.70315229 - layer.0.v_cache 0.00000026 0.00013822 - layer.1.k_cache 0.00291335 0.87192971 - layer.1.v_cache 0.00000080 0.00051506 - layer.2.k_cache 0.00116981 0.36733827 - layer.2.v_cache 0.00000115 0.00074695 - layer.3.k_cache 0.00132745 0.41247651 - layer.3.v_cache 0.00000216 0.00117735 - layer.4.k_cache 0.00359183 0.72638202 - layer.4.v_cache 0.00000354 0.00207834 - layer.4.output 0.00014916 0.06099432 - ------------------------------------------------------------------------------------- - TOTAL 0.00234816 0.66642228 - (elements=4,745,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4745216 -Total Bytes 1087524 -BPFP 1.8335 bits/point -EBPFP 3.6669 equivalent bits/point -MSE 0.666422 ----------------------- -------------------------------------------------------- -Time: 4.835s Load: 0.016s, Pack+Encode: 2.557s, Decode+Unpack: 2.263s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 331, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6664 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample39-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst (45/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 345, 128) -Output shape: (1, 345, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.output: torch.Size([1, 345, 4096]) -> torch.Size([1, 1, 345, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,328B, BPFP=0.4150 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,816B, BPFP=2.2377 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,564B, BPFP=1.5526 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,588B, BPFP=2.3684 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,372B, BPFP=1.9785 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,108B, BPFP=2.4028 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,420B, BPFP=2.0023 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,012B, BPFP=2.4006 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 71,008B, BPFP=1.6080 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,536B, BPFP=2.4351 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 299,928B, BPFP=1.6980 -⌛️ [2/4] FRONTEND: Frontend time: 2.569s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.312s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02346668 7.20200054 - layer.0.v_cache 0.00000026 0.00013957 - layer.1.k_cache 0.00293293 0.84861813 - layer.1.v_cache 0.00000076 0.00048144 - layer.2.k_cache 0.00115776 0.34287169 - layer.2.v_cache 0.00000114 0.00071223 - layer.3.k_cache 0.00132774 0.40016037 - layer.3.v_cache 0.00000222 0.00116922 - layer.4.k_cache 0.00355356 0.74120231 - layer.4.v_cache 0.00000324 0.00201882 - layer.4.output 0.00016540 0.06239915 - ------------------------------------------------------------------------------------- - TOTAL 0.00236485 0.69921221 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 1156680 -BPFP 1.8709 bits/point -EBPFP 3.7418 equivalent bits/point -MSE 0.699212 ----------------------- -------------------------------------------------------- -Time: 4.898s Load: 0.016s, Pack+Encode: 2.569s, Decode+Unpack: 2.312s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6992 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample4-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst (46/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,208B, BPFP=0.4135 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,512B, BPFP=2.2373 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,232B, BPFP=1.5723 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,252B, BPFP=2.3904 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,936B, BPFP=1.9971 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,540B, BPFP=2.4196 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,404B, BPFP=2.0077 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,768B, BPFP=2.4021 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,028B, BPFP=1.5677 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,604B, BPFP=2.4438 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 258,088B, BPFP=1.4653 -⌛️ [2/4] FRONTEND: Frontend time: 2.572s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.293s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02380104 6.85800810 - layer.0.v_cache 0.00000026 0.00014070 - layer.1.k_cache 0.00287155 0.82222446 - layer.1.v_cache 0.00000078 0.00050547 - layer.2.k_cache 0.00117748 0.36217882 - layer.2.v_cache 0.00000114 0.00072273 - layer.3.k_cache 0.00132414 0.40737210 - layer.3.v_cache 0.00000215 0.00118100 - layer.4.k_cache 0.00352424 0.70468836 - layer.4.v_cache 0.00000349 0.00209055 - layer.4.output 0.00015726 0.05719887 - ------------------------------------------------------------------------------------- - TOTAL 0.00238109 0.67056484 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 1114572 -BPFP 1.8081 bits/point -EBPFP 3.6161 equivalent bits/point -MSE 0.670565 ----------------------- -------------------------------------------------------- -Time: 4.881s Load: 0.016s, Pack+Encode: 2.572s, Decode+Unpack: 2.293s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6706 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample40-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst (47/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,164B, BPFP=0.4186 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,428B, BPFP=2.2683 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,504B, BPFP=1.5557 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,480B, BPFP=2.4078 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,088B, BPFP=1.9840 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,104B, BPFP=2.4222 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,968B, BPFP=2.0042 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,760B, BPFP=2.4143 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 70,004B, BPFP=1.6133 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,308B, BPFP=2.4730 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 247,724B, BPFP=1.4272 -⌛️ [2/4] FRONTEND: Frontend time: 2.567s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02519707 6.83849486 - layer.0.v_cache 0.00000026 0.00014143 - layer.1.k_cache 0.00285639 0.78326713 - layer.1.v_cache 0.00000078 0.00051440 - layer.2.k_cache 0.00115969 0.34328625 - layer.2.v_cache 0.00000117 0.00073665 - layer.3.k_cache 0.00131371 0.39195391 - layer.3.v_cache 0.00000219 0.00118108 - layer.4.k_cache 0.00358480 0.68457315 - layer.4.v_cache 0.00000371 0.00208695 - layer.4.output 0.00015316 0.06072735 - ------------------------------------------------------------------------------------- - TOTAL 0.00248089 0.66351037 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 1096532 -BPFP 1.8050 bits/point -EBPFP 3.6101 equivalent bits/point -MSE 0.663510 ----------------------- -------------------------------------------------------- -Time: 4.865s Load: 0.018s, Pack+Encode: 2.567s, Decode+Unpack: 2.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6635 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample41-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst (48/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.015s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 316, 128) -Output shape: (1, 316, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.output: torch.Size([1, 316, 4096]) -> torch.Size([1, 1, 316, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,028B, BPFP=0.4210 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,812B, BPFP=2.0474 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,832B, BPFP=1.5287 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,336B, BPFP=2.1839 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,140B, BPFP=1.8330 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,944B, BPFP=2.2237 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,656B, BPFP=1.8457 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,176B, BPFP=2.2047 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,688B, BPFP=1.5746 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,724B, BPFP=2.2430 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 255,596B, BPFP=1.5798 -⌛️ [2/4] FRONTEND: Frontend time: 2.369s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.044s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02408213 7.06828289 - layer.0.v_cache 0.00000026 0.00014011 - layer.1.k_cache 0.00292725 0.86519121 - layer.1.v_cache 0.00000082 0.00050866 - layer.2.k_cache 0.00117417 0.36218561 - layer.2.v_cache 0.00000116 0.00074718 - layer.3.k_cache 0.00131215 0.45212618 - layer.3.v_cache 0.00000217 0.00121323 - layer.4.k_cache 0.00352248 0.73805565 - layer.4.v_cache 0.00000354 0.00211496 - layer.4.output 0.00015546 0.05890352 - ------------------------------------------------------------------------------------- - TOTAL 0.00240343 0.69472712 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 987932 -BPFP 1.7446 bits/point -EBPFP 3.4892 equivalent bits/point -MSE 0.694727 ----------------------- -------------------------------------------------------- -Time: 4.428s Load: 0.015s, Pack+Encode: 2.369s, Decode+Unpack: 2.044s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6947 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample42-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst (49/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,064B, BPFP=0.4114 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,540B, BPFP=2.2444 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,772B, BPFP=1.6120 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,404B, BPFP=2.4008 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,412B, BPFP=1.9682 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,392B, BPFP=2.4233 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,288B, BPFP=2.0109 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,864B, BPFP=2.4113 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 70,204B, BPFP=1.5990 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,852B, BPFP=2.4565 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 242,644B, BPFP=1.3817 -⌛️ [2/4] FRONTEND: Frontend time: 2.583s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.345s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02365684 6.77631452 - layer.0.v_cache 0.00000027 0.00013982 - layer.1.k_cache 0.00286131 0.82118052 - layer.1.v_cache 0.00000077 0.00051414 - layer.2.k_cache 0.00115688 0.36117950 - layer.2.v_cache 0.00000135 0.00075051 - layer.3.k_cache 0.00129955 0.41310823 - layer.3.v_cache 0.00000214 0.00118580 - layer.4.k_cache 0.00368945 0.70897659 - layer.4.v_cache 0.00000337 0.00209679 - layer.4.output 0.00015649 0.05994262 - ------------------------------------------------------------------------------------- - TOTAL 0.00237842 0.66608692 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 1100436 -BPFP 1.7903 bits/point -EBPFP 3.5807 equivalent bits/point -MSE 0.666087 ----------------------- -------------------------------------------------------- -Time: 4.945s Load: 0.017s, Pack+Encode: 2.583s, Decode+Unpack: 2.345s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6661 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample43-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst (50/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 326, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 326, 128) -Output shape: (1, 326, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.output: torch.Size([1, 326, 4096]) -> torch.Size([1, 1, 326, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,132B, BPFP=0.4106 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,428B, BPFP=2.3109 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 63,908B, BPFP=1.5315 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 101,972B, BPFP=2.4437 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 78,204B, BPFP=1.8741 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 102,740B, BPFP=2.4621 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,032B, BPFP=2.0378 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 102,348B, BPFP=2.4527 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,796B, BPFP=1.5768 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,336B, BPFP=2.5004 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 239,032B, BPFP=1.4321 -⌛️ [2/4] FRONTEND: Frontend time: 2.614s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.output: torch.Size([1, 326, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.206s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.output: torch.Size([1, 326, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438200 7.08043235 - layer.0.v_cache 0.00000027 0.00014176 - layer.1.k_cache 0.00290854 0.85503935 - layer.1.v_cache 0.00000077 0.00049581 - layer.2.k_cache 0.00114310 0.35248996 - layer.2.v_cache 0.00000112 0.00070061 - layer.3.k_cache 0.00131875 0.38926023 - layer.3.v_cache 0.00000210 0.00113513 - layer.4.k_cache 0.00360283 0.71140622 - layer.4.v_cache 0.00000318 0.00198278 - layer.4.output 0.00014710 0.06073221 - ------------------------------------------------------------------------------------- - TOTAL 0.00242508 0.68828665 - (elements=4,673,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4673536 -Total Bytes 1056928 -BPFP 1.8092 bits/point -EBPFP 3.6184 equivalent bits/point -MSE 0.688287 ----------------------- -------------------------------------------------------- -Time: 4.836s Load: 0.017s, Pack+Encode: 2.614s, Decode+Unpack: 2.206s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 326, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6883 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample44-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst (51/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,168B, BPFP=0.4237 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,988B, BPFP=2.2618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,824B, BPFP=1.5817 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,216B, BPFP=2.4071 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,852B, BPFP=1.9788 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,844B, BPFP=2.4451 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,864B, BPFP=2.0257 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,228B, BPFP=2.4307 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,956B, BPFP=1.5848 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,336B, BPFP=2.4799 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 272,536B, BPFP=1.5889 -⌛️ [2/4] FRONTEND: Frontend time: 2.624s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.100s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02422888 7.02526965 - layer.0.v_cache 0.00000026 0.00013793 - layer.1.k_cache 0.00291837 0.77245757 - layer.1.v_cache 0.00000081 0.00052990 - layer.2.k_cache 0.00116761 0.35129039 - layer.2.v_cache 0.00000118 0.00075752 - layer.3.k_cache 0.00133299 0.40562753 - layer.3.v_cache 0.00000217 0.00122917 - layer.4.k_cache 0.00352932 0.71971057 - layer.4.v_cache 0.00000340 0.00210250 - layer.4.output 0.00015509 0.05734708 - ------------------------------------------------------------------------------------- - TOTAL 0.00241467 0.67917865 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 1113812 -BPFP 1.8554 bits/point -EBPFP 3.7107 equivalent bits/point -MSE 0.679179 ----------------------- -------------------------------------------------------- -Time: 4.740s Load: 0.017s, Pack+Encode: 2.624s, Decode+Unpack: 2.100s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6792 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample45-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst (52/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 324, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 324, 128) -Output shape: (1, 324, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.output: torch.Size([1, 324, 4096]) -> torch.Size([1, 1, 324, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,088B, BPFP=0.4120 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 94,844B, BPFP=2.2869 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 63,288B, BPFP=1.5260 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 101,296B, BPFP=2.4425 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 80,788B, BPFP=1.9480 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 102,616B, BPFP=2.4743 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 82,896B, BPFP=1.9988 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 102,152B, BPFP=2.4632 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,060B, BPFP=1.5688 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,592B, BPFP=2.5220 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 250,188B, BPFP=1.5082 -⌛️ [2/4] FRONTEND: Frontend time: 2.586s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.output: torch.Size([1, 324, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.168s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.output: torch.Size([1, 324, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02313130 6.85702703 - layer.0.v_cache 0.00000026 0.00013834 - layer.1.k_cache 0.00295515 0.74579262 - layer.1.v_cache 0.00000079 0.00049928 - layer.2.k_cache 0.00117180 0.37366596 - layer.2.v_cache 0.00000115 0.00074453 - layer.3.k_cache 0.00132196 0.39344288 - layer.3.v_cache 0.00000215 0.00119141 - layer.4.k_cache 0.00355989 0.72896265 - layer.4.v_cache 0.00000355 0.00209868 - layer.4.output 0.00015612 0.06136071 - ------------------------------------------------------------------------------------- - TOTAL 0.00234089 0.66778616 - (elements=4,644,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4644864 -Total Bytes 1064808 -BPFP 1.8340 bits/point -EBPFP 3.6679 equivalent bits/point -MSE 0.667786 ----------------------- -------------------------------------------------------- -Time: 4.775s Load: 0.021s, Pack+Encode: 2.586s, Decode+Unpack: 2.168s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 324, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6678 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample46-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst (53/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 332, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 332, 128) -Output shape: (1, 332, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.output: torch.Size([1, 332, 4096]) -> torch.Size([1, 1, 332, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,184B, BPFP=0.4279 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,192B, BPFP=2.2871 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,024B, BPFP=1.6242 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,520B, BPFP=2.4125 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,232B, BPFP=1.9821 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,192B, BPFP=2.4753 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,048B, BPFP=2.0248 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,808B, BPFP=2.4663 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,496B, BPFP=1.5883 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,912B, BPFP=2.4923 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 262,900B, BPFP=1.5466 -⌛️ [2/4] FRONTEND: Frontend time: 2.574s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.output: torch.Size([1, 332, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.147s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.output: torch.Size([1, 332, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02360621 6.69204969 - layer.0.v_cache 0.00000026 0.00013896 - layer.1.k_cache 0.00288168 0.77376170 - layer.1.v_cache 0.00000079 0.00051987 - layer.2.k_cache 0.00118772 0.35895986 - layer.2.v_cache 0.00000117 0.00077964 - layer.3.k_cache 0.00132341 0.40104064 - layer.3.v_cache 0.00000218 0.00123572 - layer.4.k_cache 0.00354949 0.73576916 - layer.4.v_cache 0.00000378 0.00218050 - layer.4.output 0.00018345 0.05905116 - ------------------------------------------------------------------------------------- - TOTAL 0.00237789 0.65733146 - (elements=4,759,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4759552 -Total Bytes 1103508 -BPFP 1.8548 bits/point -EBPFP 3.7096 equivalent bits/point -MSE 0.657331 ----------------------- -------------------------------------------------------- -Time: 4.739s Load: 0.018s, Pack+Encode: 2.574s, Decode+Unpack: 2.147s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 332, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6573 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample47-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst (54/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 319, 128) -Output shape: (1, 319, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.output: torch.Size([1, 319, 4096]) -> torch.Size([1, 1, 319, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,204B, BPFP=0.4213 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,624B, BPFP=2.0480 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 62,128B, BPFP=1.5216 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,812B, BPFP=2.1751 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,252B, BPFP=1.8185 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,964B, BPFP=2.2033 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,312B, BPFP=1.8199 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,912B, BPFP=2.1775 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,236B, BPFP=1.5242 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,428B, BPFP=2.2146 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 222,740B, BPFP=1.3638 -⌛️ [2/4] FRONTEND: Frontend time: 2.369s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.993s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438342 7.16345177 - layer.0.v_cache 0.00000027 0.00013964 - layer.1.k_cache 0.00287626 0.86270572 - layer.1.v_cache 0.00000082 0.00051872 - layer.2.k_cache 0.00122126 0.36596790 - layer.2.v_cache 0.00000114 0.00072840 - layer.3.k_cache 0.00133159 0.43883230 - layer.3.v_cache 0.00000213 0.00115869 - layer.4.k_cache 0.00352266 0.72340613 - layer.4.v_cache 0.00000325 0.00201899 - layer.4.output 0.00013715 0.05410822 - ------------------------------------------------------------------------------------- - TOTAL 0.00242081 0.69824008 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 954612 -BPFP 1.6699 bits/point -EBPFP 3.3399 equivalent bits/point -MSE 0.698240 ----------------------- -------------------------------------------------------- -Time: 4.379s Load: 0.017s, Pack+Encode: 2.369s, Decode+Unpack: 1.993s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6982 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample48-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst (55/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,960B, BPFP=0.4188 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,180B, BPFP=2.2663 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,128B, BPFP=1.5655 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,264B, BPFP=2.4315 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,760B, BPFP=1.9767 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,728B, BPFP=2.4190 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,900B, BPFP=2.0033 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,312B, BPFP=2.4326 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,708B, BPFP=1.5790 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,336B, BPFP=2.4799 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 231,224B, BPFP=1.3481 -⌛️ [2/4] FRONTEND: Frontend time: 2.586s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.211s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02322828 6.87539937 - layer.0.v_cache 0.00000026 0.00013929 - layer.1.k_cache 0.00283722 0.82182982 - layer.1.v_cache 0.00000078 0.00051472 - layer.2.k_cache 0.00117890 0.34518378 - layer.2.v_cache 0.00000113 0.00072329 - layer.3.k_cache 0.00131429 0.39329338 - layer.3.v_cache 0.00000212 0.00117897 - layer.4.k_cache 0.00357066 0.72319454 - layer.4.v_cache 0.00000341 0.00208156 - layer.4.output 0.00013949 0.05754801 - ------------------------------------------------------------------------------------- - TOTAL 0.00233536 0.67098077 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 1070500 -BPFP 1.7832 bits/point -EBPFP 3.5664 equivalent bits/point -MSE 0.670981 ----------------------- -------------------------------------------------------- -Time: 4.815s Load: 0.018s, Pack+Encode: 2.586s, Decode+Unpack: 2.211s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6710 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample49-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst (56/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 354, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 354, 128) -Output shape: (1, 354, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.output: torch.Size([1, 354, 4096]) -> torch.Size([1, 1, 354, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,308B, BPFP=0.4261 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,780B, BPFP=2.2021 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,908B, BPFP=1.5649 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,336B, BPFP=2.3247 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,688B, BPFP=1.9352 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,612B, BPFP=2.3528 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,876B, BPFP=1.9614 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,832B, BPFP=2.3356 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 73,948B, BPFP=1.6320 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,824B, BPFP=2.3796 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 305,724B, BPFP=1.6868 -⌛️ [2/4] FRONTEND: Frontend time: 2.598s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.output: torch.Size([1, 354, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.300s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.output: torch.Size([1, 354, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02475441 6.91167075 - layer.0.v_cache 0.00000027 0.00014268 - layer.1.k_cache 0.00288960 0.89716869 - layer.1.v_cache 0.00000085 0.00052612 - layer.2.k_cache 0.00116959 0.35344000 - layer.2.v_cache 0.00000115 0.00074344 - layer.3.k_cache 0.00132243 0.41045181 - layer.3.v_cache 0.00000218 0.00121037 - layer.4.k_cache 0.00355194 0.70009781 - layer.4.v_cache 0.00000329 0.00212541 - layer.4.output 0.00015235 0.05604521 - ------------------------------------------------------------------------------------- - TOTAL 0.00245037 0.67869700 - (elements=5,074,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5074944 -Total Bytes 1171836 -BPFP 1.8472 bits/point -EBPFP 3.6945 equivalent bits/point -MSE 0.678697 ----------------------- -------------------------------------------------------- -Time: 4.916s Load: 0.019s, Pack+Encode: 2.598s, Decode+Unpack: 2.300s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 354, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6787 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample5-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst (57/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 327, 128) -Output shape: (1, 327, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.output: torch.Size([1, 327, 4096]) -> torch.Size([1, 1, 327, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,932B, BPFP=0.4284 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,880B, BPFP=2.3146 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 63,704B, BPFP=1.5220 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,640B, BPFP=2.4522 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 80,216B, BPFP=1.9165 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,824B, BPFP=2.4805 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 84,176B, BPFP=2.0111 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 102,900B, BPFP=2.4584 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,204B, BPFP=1.5339 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,032B, BPFP=2.4855 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 270,088B, BPFP=1.6132 -⌛️ [2/4] FRONTEND: Frontend time: 2.571s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.301s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02504845 6.61945868 - layer.0.v_cache 0.00000027 0.00013939 - layer.1.k_cache 0.00290162 0.79344658 - layer.1.v_cache 0.00000083 0.00052284 - layer.2.k_cache 0.00115956 0.34870640 - layer.2.v_cache 0.00000120 0.00074906 - layer.3.k_cache 0.00131802 0.40148478 - layer.3.v_cache 0.00000223 0.00120551 - layer.4.k_cache 0.00351161 0.72534702 - layer.4.v_cache 0.00000355 0.00211000 - layer.4.output 0.00016630 0.05836799 - ------------------------------------------------------------------------------------- - TOTAL 0.00247233 0.65190302 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 1090596 -BPFP 1.8611 bits/point -EBPFP 3.7223 equivalent bits/point -MSE 0.651903 ----------------------- -------------------------------------------------------- -Time: 4.894s Load: 0.022s, Pack+Encode: 2.571s, Decode+Unpack: 2.301s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6519 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample50-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst (58/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 313, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 313, 128) -Output shape: (1, 313, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.output: torch.Size([1, 313, 4096]) -> torch.Size([1, 1, 313, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,040B, BPFP=0.4253 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,276B, BPFP=2.0786 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,908B, BPFP=1.5203 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 87,956B, BPFP=2.1954 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 75,324B, BPFP=1.8801 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,688B, BPFP=2.2386 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,524B, BPFP=1.8601 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,876B, BPFP=2.2184 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,408B, BPFP=1.5577 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,632B, BPFP=2.2622 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 262,776B, BPFP=1.6397 -⌛️ [2/4] FRONTEND: Frontend time: 2.357s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.output: torch.Size([1, 313, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.970s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.output: torch.Size([1, 313, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02366765 7.05429204 - layer.0.v_cache 0.00000027 0.00014216 - layer.1.k_cache 0.00295685 0.79577954 - layer.1.v_cache 0.00000081 0.00048376 - layer.2.k_cache 0.00117427 0.36702541 - layer.2.v_cache 0.00000111 0.00072563 - layer.3.k_cache 0.00131811 0.43033388 - layer.3.v_cache 0.00000214 0.00118018 - layer.4.k_cache 0.00360037 0.72131855 - layer.4.v_cache 0.00000322 0.00206211 - layer.4.output 0.00015721 0.06112484 - ------------------------------------------------------------------------------------- - TOTAL 0.00238240 0.68698876 - (elements=4,487,168) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4487168 -Total Bytes 993408 -BPFP 1.7711 bits/point -EBPFP 3.5422 equivalent bits/point -MSE 0.686989 ----------------------- -------------------------------------------------------- -Time: 4.343s Load: 0.016s, Pack+Encode: 2.357s, Decode+Unpack: 1.970s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 313, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6870 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample51-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst (59/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,524B, BPFP=0.4161 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,504B, BPFP=2.3153 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 65,380B, BPFP=1.5525 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,752B, BPFP=2.4637 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,104B, BPFP=1.9734 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,144B, BPFP=2.4730 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,012B, BPFP=2.0187 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,172B, BPFP=2.4737 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,556B, BPFP=1.5567 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,644B, BPFP=2.5086 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 269,040B, BPFP=1.5972 -⌛️ [2/4] FRONTEND: Frontend time: 2.609s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.287s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435147 6.93655312 - layer.0.v_cache 0.00000027 0.00014130 - layer.1.k_cache 0.00282437 0.80512783 - layer.1.v_cache 0.00000081 0.00052132 - layer.2.k_cache 0.00117800 0.34997809 - layer.2.v_cache 0.00000117 0.00075675 - layer.3.k_cache 0.00131145 0.40333001 - layer.3.v_cache 0.00000228 0.00123058 - layer.4.k_cache 0.00354228 0.70106822 - layer.4.v_cache 0.00000348 0.00217444 - layer.4.output 0.00016143 0.05952393 - ------------------------------------------------------------------------------------- - TOTAL 0.00241866 0.67421267 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 1100832 -BPFP 1.8672 bits/point -EBPFP 3.7344 equivalent bits/point -MSE 0.674213 ----------------------- -------------------------------------------------------- -Time: 4.914s Load: 0.018s, Pack+Encode: 2.609s, Decode+Unpack: 2.287s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6742 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample52-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst (60/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.024s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,620B, BPFP=0.4171 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,876B, BPFP=2.3171 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 65,608B, BPFP=1.5532 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,820B, BPFP=2.4579 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 82,792B, BPFP=1.9600 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,276B, BPFP=2.4923 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,248B, BPFP=2.0182 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,072B, BPFP=2.4638 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,468B, BPFP=1.5736 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,580B, BPFP=2.4995 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 260,088B, BPFP=1.5393 -⌛️ [2/4] FRONTEND: Frontend time: 2.586s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.151s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02327405 6.75585864 - layer.0.v_cache 0.00000026 0.00014124 - layer.1.k_cache 0.00286502 0.80929334 - layer.1.v_cache 0.00000078 0.00051557 - layer.2.k_cache 0.00115867 0.35555233 - layer.2.v_cache 0.00000116 0.00074380 - layer.3.k_cache 0.00132713 0.39463792 - layer.3.v_cache 0.00000212 0.00118839 - layer.4.k_cache 0.00352212 0.70060499 - layer.4.v_cache 0.00000405 0.00216572 - layer.4.output 0.00014449 0.05966843 - ------------------------------------------------------------------------------------- - TOTAL 0.00233809 0.66138397 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 1094448 -BPFP 1.8507 bits/point -EBPFP 3.7015 equivalent bits/point -MSE 0.661384 ----------------------- -------------------------------------------------------- -Time: 4.761s Load: 0.024s, Pack+Encode: 2.586s, Decode+Unpack: 2.151s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6614 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample53-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst (61/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.015s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 320, 128) -Output shape: (1, 320, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.output: torch.Size([1, 320, 4096]) -> torch.Size([1, 1, 320, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,796B, BPFP=0.4101 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,204B, BPFP=2.0069 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,948B, BPFP=1.4880 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 87,260B, BPFP=2.1304 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 72,880B, BPFP=1.7793 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 88,784B, BPFP=2.1676 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 73,168B, BPFP=1.7863 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 87,852B, BPFP=2.1448 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,400B, BPFP=1.5234 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 89,568B, BPFP=2.1867 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 253,780B, BPFP=1.5490 -⌛️ [2/4] FRONTEND: Frontend time: 2.369s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.014s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461678 6.83154144 - layer.0.v_cache 0.00000026 0.00013221 - layer.1.k_cache 0.00291374 0.80770741 - layer.1.v_cache 0.00000082 0.00048403 - layer.2.k_cache 0.00121384 0.33962672 - layer.2.v_cache 0.00000117 0.00071167 - layer.3.k_cache 0.00132315 0.40912418 - layer.3.v_cache 0.00000219 0.00115891 - layer.4.k_cache 0.00353222 0.71546454 - layer.4.v_cache 0.00000326 0.00204754 - layer.4.output 0.00015660 0.05102996 - ------------------------------------------------------------------------------------- - TOTAL 0.00244527 0.66515132 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 975640 -BPFP 1.7014 bits/point -EBPFP 3.4028 equivalent bits/point -MSE 0.665151 ----------------------- -------------------------------------------------------- -Time: 4.398s Load: 0.015s, Pack+Encode: 2.369s, Decode+Unpack: 2.014s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6652 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample54-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst (62/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,084B, BPFP=0.4168 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,440B, BPFP=2.2686 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,892B, BPFP=1.5877 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,764B, BPFP=2.4144 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,520B, BPFP=1.9939 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,400B, BPFP=2.4290 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,460B, BPFP=2.0156 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,792B, BPFP=2.4150 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,692B, BPFP=1.6061 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,604B, BPFP=2.4568 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 246,032B, BPFP=1.4175 -⌛️ [2/4] FRONTEND: Frontend time: 2.573s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.218s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02391626 6.65290837 - layer.0.v_cache 0.00000026 0.00013934 - layer.1.k_cache 0.00286983 0.89282974 - layer.1.v_cache 0.00000081 0.00052146 - layer.2.k_cache 0.00115809 0.35162959 - layer.2.v_cache 0.00000117 0.00075151 - layer.3.k_cache 0.00131372 0.40183166 - layer.3.v_cache 0.00000240 0.00122346 - layer.4.k_cache 0.00359509 0.73761139 - layer.4.v_cache 0.00000368 0.00216053 - layer.4.output 0.00014345 0.05862570 - ------------------------------------------------------------------------------------- - TOTAL 0.00238822 0.66257927 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 1096680 -BPFP 1.8053 bits/point -EBPFP 3.6105 equivalent bits/point -MSE 0.662579 ----------------------- -------------------------------------------------------- -Time: 4.809s Load: 0.019s, Pack+Encode: 2.573s, Decode+Unpack: 2.218s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6626 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample55-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst (63/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 314, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.020s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 314, 128) -Output shape: (1, 314, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.output: torch.Size([1, 314, 4096]) -> torch.Size([1, 1, 314, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,944B, BPFP=0.4216 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,928B, BPFP=2.0633 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 62,456B, BPFP=1.5539 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,836B, BPFP=2.2103 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,720B, BPFP=1.8591 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 90,004B, BPFP=2.2394 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,764B, BPFP=1.8602 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,016B, BPFP=2.2148 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,728B, BPFP=1.5856 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 91,020B, BPFP=2.2646 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 247,864B, BPFP=1.5417 -⌛️ [2/4] FRONTEND: Frontend time: 2.365s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.output: torch.Size([1, 314, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.906s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.output: torch.Size([1, 314, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02380536 7.00350350 - layer.0.v_cache 0.00000027 0.00014645 - layer.1.k_cache 0.00292133 0.84581810 - layer.1.v_cache 0.00000080 0.00051463 - layer.2.k_cache 0.00120652 0.39931765 - layer.2.v_cache 0.00000113 0.00074274 - layer.3.k_cache 0.00133526 0.44274465 - layer.3.v_cache 0.00000215 0.00119204 - layer.4.k_cache 0.00353655 0.70567628 - layer.4.v_cache 0.00000336 0.00210958 - layer.4.output 0.00016938 0.06244926 - ------------------------------------------------------------------------------------- - TOTAL 0.00239216 0.68939733 - (elements=4,501,504) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4501504 -Total Bytes 982280 -BPFP 1.7457 bits/point -EBPFP 3.4914 equivalent bits/point -MSE 0.689397 ----------------------- -------------------------------------------------------- -Time: 4.291s Load: 0.020s, Pack+Encode: 2.365s, Decode+Unpack: 1.906s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 314, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6894 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample56-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst (64/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,184B, BPFP=0.4130 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,428B, BPFP=2.2354 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,136B, BPFP=1.5701 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,644B, BPFP=2.3993 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,340B, BPFP=2.0063 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,220B, BPFP=2.4350 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,628B, BPFP=2.0128 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,340B, BPFP=2.4151 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 71,144B, BPFP=1.6157 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,164B, BPFP=2.4338 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 244,432B, BPFP=1.3878 -⌛️ [2/4] FRONTEND: Frontend time: 2.580s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.241s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423059 7.02785616 - layer.0.v_cache 0.00000027 0.00014223 - layer.1.k_cache 0.00284490 0.81926443 - layer.1.v_cache 0.00000080 0.00052227 - layer.2.k_cache 0.00115895 0.36087666 - layer.2.v_cache 0.00000119 0.00076109 - layer.3.k_cache 0.00130300 0.39601956 - layer.3.v_cache 0.00000215 0.00120496 - layer.4.k_cache 0.00352154 0.69314903 - layer.4.v_cache 0.00000376 0.00214683 - layer.4.output 0.00013131 0.05564512 - ------------------------------------------------------------------------------------- - TOTAL 0.00239946 0.68032312 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 1104660 -BPFP 1.7920 bits/point -EBPFP 3.5840 equivalent bits/point -MSE 0.680323 ----------------------- -------------------------------------------------------- -Time: 4.838s Load: 0.017s, Pack+Encode: 2.580s, Decode+Unpack: 2.241s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6803 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample57-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst (65/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 308, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 308, 128) -Output shape: (1, 308, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.output: torch.Size([1, 308, 4096]) -> torch.Size([1, 1, 308, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,812B, BPFP=0.4264 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,436B, BPFP=2.1164 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 62,036B, BPFP=1.5736 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,732B, BPFP=2.2507 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,480B, BPFP=1.8892 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,928B, BPFP=2.2810 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 75,400B, BPFP=1.9125 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,244B, BPFP=2.2637 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,108B, BPFP=1.6261 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,948B, BPFP=2.3069 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 271,128B, BPFP=1.7193 -⌛️ [2/4] FRONTEND: Frontend time: 2.332s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.output: torch.Size([1, 308, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.008s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.output: torch.Size([1, 308, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02345257 6.77205668 - layer.0.v_cache 0.00000027 0.00014160 - layer.1.k_cache 0.00291045 0.86563249 - layer.1.v_cache 0.00000082 0.00052227 - layer.2.k_cache 0.00116968 0.35427094 - layer.2.v_cache 0.00000115 0.00073239 - layer.3.k_cache 0.00131855 0.43729614 - layer.3.v_cache 0.00000224 0.00122863 - layer.4.k_cache 0.00354502 0.72378659 - layer.4.v_cache 0.00000324 0.00210499 - layer.4.output 0.00015550 0.05989654 - ------------------------------------------------------------------------------------- - TOTAL 0.00235900 0.67123992 - (elements=4,415,488) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4415488 -Total Bytes 1006252 -BPFP 1.8231 bits/point -EBPFP 3.6463 equivalent bits/point -MSE 0.671240 ----------------------- -------------------------------------------------------- -Time: 4.359s Load: 0.018s, Pack+Encode: 2.332s, Decode+Unpack: 2.008s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 308, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6712 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample58-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst (66/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,272B, BPFP=0.4211 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,732B, BPFP=2.2523 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,232B, BPFP=1.5725 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,772B, BPFP=2.3915 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,884B, BPFP=1.9562 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,708B, BPFP=2.4361 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,576B, BPFP=2.0183 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,736B, BPFP=2.4137 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,508B, BPFP=1.5788 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,348B, BPFP=2.4739 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 271,004B, BPFP=1.5614 -⌛️ [2/4] FRONTEND: Frontend time: 2.621s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.152s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02515205 7.22200737 - layer.0.v_cache 0.00000026 0.00014145 - layer.1.k_cache 0.00294413 0.77193854 - layer.1.v_cache 0.00000079 0.00051029 - layer.2.k_cache 0.00117757 0.35783271 - layer.2.v_cache 0.00000116 0.00073679 - layer.3.k_cache 0.00130963 0.39319212 - layer.3.v_cache 0.00000218 0.00121409 - layer.4.k_cache 0.00367572 0.75271075 - layer.4.v_cache 0.00000325 0.00206194 - layer.4.output 0.00014293 0.05806879 - ------------------------------------------------------------------------------------- - TOTAL 0.00248846 0.69533009 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 1117772 -BPFP 1.8400 bits/point -EBPFP 3.6800 equivalent bits/point -MSE 0.695330 ----------------------- -------------------------------------------------------- -Time: 4.792s Load: 0.019s, Pack+Encode: 2.621s, Decode+Unpack: 2.152s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6953 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample59-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst (67/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 355, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 355, 128) -Output shape: (1, 355, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.output: torch.Size([1, 355, 4096]) -> torch.Size([1, 1, 355, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,284B, BPFP=0.4024 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,420B, BPFP=2.1879 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,712B, BPFP=1.6002 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,480B, BPFP=2.3213 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,672B, BPFP=1.9514 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,652B, BPFP=2.3471 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,232B, BPFP=1.9637 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,660B, BPFP=2.3253 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,900B, BPFP=1.6483 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,776B, BPFP=2.3718 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 243,020B, BPFP=1.3370 -⌛️ [2/4] FRONTEND: Frontend time: 2.620s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.output: torch.Size([1, 355, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.207s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.output: torch.Size([1, 355, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438693 6.61079583 - layer.0.v_cache 0.00000027 0.00013929 - layer.1.k_cache 0.00283684 0.80066528 - layer.1.v_cache 0.00000080 0.00052803 - layer.2.k_cache 0.00116829 0.35563806 - layer.2.v_cache 0.00000120 0.00076481 - layer.3.k_cache 0.00130356 0.41529438 - layer.3.v_cache 0.00000219 0.00119845 - layer.4.k_cache 0.00355311 0.71894836 - layer.4.v_cache 0.00000352 0.00210933 - layer.4.output 0.00013754 0.05279684 - ------------------------------------------------------------------------------------- - TOTAL 0.00241478 0.65123351 - (elements=5,089,280) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5089280 -Total Bytes 1111808 -BPFP 1.7477 bits/point -EBPFP 3.4954 equivalent bits/point -MSE 0.651234 ----------------------- -------------------------------------------------------- -Time: 4.845s Load: 0.018s, Pack+Encode: 2.620s, Decode+Unpack: 2.207s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 355, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6512 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample6-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst (68/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 334, 128) -Output shape: (1, 334, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.output: torch.Size([1, 334, 4096]) -> torch.Size([1, 1, 334, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,036B, BPFP=0.4219 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,252B, BPFP=2.2748 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 65,376B, BPFP=1.5292 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,216B, BPFP=2.4377 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,740B, BPFP=1.9821 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,360B, BPFP=2.4411 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,772B, BPFP=2.0297 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,620B, BPFP=2.4471 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,972B, BPFP=1.5665 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,596B, BPFP=2.4934 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 247,028B, BPFP=1.4445 -⌛️ [2/4] FRONTEND: Frontend time: 2.633s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.141s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02444427 6.93491094 - layer.0.v_cache 0.00000026 0.00014123 - layer.1.k_cache 0.00284576 0.84097564 - layer.1.v_cache 0.00000078 0.00052224 - layer.2.k_cache 0.00117801 0.35963472 - layer.2.v_cache 0.00000115 0.00073314 - layer.3.k_cache 0.00132187 0.39178248 - layer.3.v_cache 0.00000218 0.00122175 - layer.4.k_cache 0.00348374 0.71693946 - layer.4.v_cache 0.00000337 0.00215356 - layer.4.output 0.00014793 0.05431140 - ------------------------------------------------------------------------------------- - TOTAL 0.00241951 0.67616148 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 1085968 -BPFP 1.8144 bits/point -EBPFP 3.6288 equivalent bits/point -MSE 0.676161 ----------------------- -------------------------------------------------------- -Time: 4.792s Load: 0.018s, Pack+Encode: 2.633s, Decode+Unpack: 2.141s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6762 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample60-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst (69/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,688B, BPFP=0.4041 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,296B, BPFP=2.2454 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,640B, BPFP=1.5908 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,952B, BPFP=2.3975 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,108B, BPFP=2.0127 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,136B, BPFP=2.4245 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,068B, BPFP=2.0118 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,844B, BPFP=2.4179 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 71,112B, BPFP=1.6245 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,288B, BPFP=2.4508 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 224,360B, BPFP=1.2813 -⌛️ [2/4] FRONTEND: Frontend time: 2.583s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.233s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02429651 6.77597599 - layer.0.v_cache 0.00000027 0.00014017 - layer.1.k_cache 0.00293320 0.77166668 - layer.1.v_cache 0.00000079 0.00053144 - layer.2.k_cache 0.00117175 0.35468087 - layer.2.v_cache 0.00000114 0.00075110 - layer.3.k_cache 0.00133100 0.40569484 - layer.3.v_cache 0.00000223 0.00118787 - layer.4.k_cache 0.00354789 0.72753987 - layer.4.v_cache 0.00000344 0.00209291 - layer.4.output 0.00015145 0.05611440 - ------------------------------------------------------------------------------------- - TOTAL 0.00242100 0.66176567 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 1081492 -BPFP 1.7647 bits/point -EBPFP 3.5293 equivalent bits/point -MSE 0.661766 ----------------------- -------------------------------------------------------- -Time: 4.833s Load: 0.017s, Pack+Encode: 2.583s, Decode+Unpack: 2.233s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6618 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample61-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst (70/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 312, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 312, 128) -Output shape: (1, 312, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.output: torch.Size([1, 312, 4096]) -> torch.Size([1, 1, 312, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,704B, BPFP=0.4183 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,736B, BPFP=2.0717 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,460B, BPFP=1.5139 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,104B, BPFP=2.2061 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,412B, BPFP=1.8633 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,668B, BPFP=2.2453 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 75,292B, BPFP=1.8853 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,844B, BPFP=2.2247 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,776B, BPFP=1.5719 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,184B, BPFP=2.2582 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 257,708B, BPFP=1.6133 -⌛️ [2/4] FRONTEND: Frontend time: 2.397s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.output: torch.Size([1, 312, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.955s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.output: torch.Size([1, 312, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02396220 7.04134036 - layer.0.v_cache 0.00000026 0.00014138 - layer.1.k_cache 0.00299662 0.83917002 - layer.1.v_cache 0.00000081 0.00051087 - layer.2.k_cache 0.00115233 0.34397005 - layer.2.v_cache 0.00000114 0.00074910 - layer.3.k_cache 0.00133293 0.43444443 - layer.3.v_cache 0.00000220 0.00122975 - layer.4.k_cache 0.00362032 0.75740990 - layer.4.v_cache 0.00000325 0.00207599 - layer.4.output 0.00015781 0.06254315 - ------------------------------------------------------------------------------------- - TOTAL 0.00240738 0.69080103 - (elements=4,472,832) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4472832 -Total Bytes 986888 -BPFP 1.7651 bits/point -EBPFP 3.5302 equivalent bits/point -MSE 0.690801 ----------------------- -------------------------------------------------------- -Time: 4.369s Load: 0.017s, Pack+Encode: 2.397s, Decode+Unpack: 1.955s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 312, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6908 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample62-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst (71/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,284B, BPFP=0.4070 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,532B, BPFP=2.2154 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,272B, BPFP=1.5641 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,668B, BPFP=2.3519 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,208B, BPFP=1.9633 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,760B, BPFP=2.3762 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,348B, BPFP=1.9887 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,948B, BPFP=2.3582 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,904B, BPFP=1.6227 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,724B, BPFP=2.3977 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 257,584B, BPFP=1.4333 -⌛️ [2/4] FRONTEND: Frontend time: 2.574s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.248s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435836 6.63301004 - layer.0.v_cache 0.00000027 0.00014174 - layer.1.k_cache 0.00294291 0.79759351 - layer.1.v_cache 0.00000079 0.00052408 - layer.2.k_cache 0.00116231 0.38375015 - layer.2.v_cache 0.00000118 0.00073997 - layer.3.k_cache 0.00133045 0.41452870 - layer.3.v_cache 0.00000222 0.00120332 - layer.4.k_cache 0.00357138 0.72635375 - layer.4.v_cache 0.00000341 0.00209251 - layer.4.output 0.00014419 0.06015747 - ------------------------------------------------------------------------------------- - TOTAL 0.00242500 0.65718340 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 1122232 -BPFP 1.7842 bits/point -EBPFP 3.5684 equivalent bits/point -MSE 0.657183 ----------------------- -------------------------------------------------------- -Time: 4.841s Load: 0.018s, Pack+Encode: 2.574s, Decode+Unpack: 2.248s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6572 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample63-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst (72/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,896B, BPFP=0.4087 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 94,684B, BPFP=2.2902 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 64,064B, BPFP=1.5495 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,180B, BPFP=2.4715 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 77,772B, BPFP=1.8811 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,064B, BPFP=2.4928 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 82,220B, BPFP=1.9887 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 101,108B, BPFP=2.4455 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,572B, BPFP=1.5618 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 103,696B, BPFP=2.5081 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 219,648B, BPFP=1.3282 -⌛️ [2/4] FRONTEND: Frontend time: 2.605s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.261s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02343196 6.92341316 - layer.0.v_cache 0.00000026 0.00013968 - layer.1.k_cache 0.00287105 0.80481486 - layer.1.v_cache 0.00000078 0.00051776 - layer.2.k_cache 0.00117289 0.36698597 - layer.2.v_cache 0.00000112 0.00073805 - layer.3.k_cache 0.00130124 0.40062570 - layer.3.v_cache 0.00000213 0.00117088 - layer.4.k_cache 0.00357998 0.74483072 - layer.4.v_cache 0.00000342 0.00207280 - layer.4.output 0.00014651 0.05793483 - ------------------------------------------------------------------------------------- - TOTAL 0.00235363 0.67693206 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 1029904 -BPFP 1.7793 bits/point -EBPFP 3.5587 equivalent bits/point -MSE 0.676932 ----------------------- -------------------------------------------------------- -Time: 4.886s Load: 0.021s, Pack+Encode: 2.605s, Decode+Unpack: 2.261s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6769 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample64-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst (73/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 356, 128) -Output shape: (1, 356, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.output: torch.Size([1, 356, 4096]) -> torch.Size([1, 1, 356, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,860B, BPFP=0.4139 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,260B, BPFP=2.1783 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,276B, BPFP=1.5861 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,904B, BPFP=2.3021 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,752B, BPFP=1.9477 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,452B, BPFP=2.3361 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,304B, BPFP=1.9598 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,380B, BPFP=2.3126 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,848B, BPFP=1.6426 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,520B, BPFP=2.3596 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 272,576B, BPFP=1.4954 -⌛️ [2/4] FRONTEND: Frontend time: 2.602s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.320s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02477347 6.92850520 - layer.0.v_cache 0.00000027 0.00014211 - layer.1.k_cache 0.00294478 0.83587124 - layer.1.v_cache 0.00000080 0.00052018 - layer.2.k_cache 0.00115317 0.34196314 - layer.2.v_cache 0.00000113 0.00072953 - layer.3.k_cache 0.00130758 0.41377571 - layer.3.v_cache 0.00000212 0.00118891 - layer.4.k_cache 0.00357902 0.73504176 - layer.4.v_cache 0.00000333 0.00209542 - layer.4.output 0.00014592 0.05736571 - ------------------------------------------------------------------------------------- - TOTAL 0.00245352 0.67780686 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 1140132 -BPFP 1.7872 bits/point -EBPFP 3.5744 equivalent bits/point -MSE 0.677807 ----------------------- -------------------------------------------------------- -Time: 4.941s Load: 0.018s, Pack+Encode: 2.602s, Decode+Unpack: 2.320s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6778 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample65-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst (74/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,228B, BPFP=0.4164 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,088B, BPFP=2.2635 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,796B, BPFP=1.5487 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,168B, BPFP=2.4024 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,304B, BPFP=1.9943 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,220B, BPFP=2.4264 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,848B, BPFP=2.0068 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,604B, BPFP=2.4124 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,424B, BPFP=1.5859 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,852B, BPFP=2.4637 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 262,428B, BPFP=1.4987 -⌛️ [2/4] FRONTEND: Frontend time: 2.582s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.244s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02260964 6.86484461 - layer.0.v_cache 0.00000027 0.00014329 - layer.1.k_cache 0.00288942 0.84832237 - layer.1.v_cache 0.00000084 0.00052459 - layer.2.k_cache 0.00116702 0.35866846 - layer.2.v_cache 0.00000122 0.00074312 - layer.3.k_cache 0.00131477 0.39542777 - layer.3.v_cache 0.00000216 0.00119070 - layer.4.k_cache 0.00350203 0.69886173 - layer.4.v_cache 0.00000356 0.00210939 - layer.4.output 0.00013713 0.05963589 - ------------------------------------------------------------------------------------- - TOTAL 0.00228853 0.67209854 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 1116960 -BPFP 1.8225 bits/point -EBPFP 3.6451 equivalent bits/point -MSE 0.672099 ----------------------- -------------------------------------------------------- -Time: 4.845s Load: 0.019s, Pack+Encode: 2.582s, Decode+Unpack: 2.244s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6721 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample66-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst (75/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 360, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.023s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 360, 128) -Output shape: (1, 360, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.output: torch.Size([1, 360, 4096]) -> torch.Size([1, 1, 360, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,512B, BPFP=0.4234 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 100,040B, BPFP=2.1710 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 72,944B, BPFP=1.5830 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 105,872B, BPFP=2.2976 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 89,220B, BPFP=1.9362 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 107,112B, BPFP=2.3245 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,332B, BPFP=1.9386 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,492B, BPFP=2.3110 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 75,888B, BPFP=1.6469 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 108,716B, BPFP=2.3593 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 275,924B, BPFP=1.4970 -⌛️ [2/4] FRONTEND: Frontend time: 2.516s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.output: torch.Size([1, 360, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.102s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.output: torch.Size([1, 360, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02342221 6.70379028 - layer.0.v_cache 0.00000028 0.00014230 - layer.1.k_cache 0.00285409 0.85740272 - layer.1.v_cache 0.00000079 0.00051918 - layer.2.k_cache 0.00117169 0.35140485 - layer.2.v_cache 0.00000113 0.00073083 - layer.3.k_cache 0.00130395 0.41407619 - layer.3.v_cache 0.00000220 0.00123210 - layer.4.k_cache 0.00353751 0.70140767 - layer.4.v_cache 0.00000341 0.00219583 - layer.4.output 0.00014250 0.05920288 - ------------------------------------------------------------------------------------- - TOTAL 0.00234766 0.66212239 - (elements=5,160,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5160960 -Total Bytes 1151052 -BPFP 1.7842 bits/point -EBPFP 3.5685 equivalent bits/point -MSE 0.662122 ----------------------- -------------------------------------------------------- -Time: 4.640s Load: 0.023s, Pack+Encode: 2.516s, Decode+Unpack: 2.102s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 360, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6621 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample67-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst (76/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 294, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 294, 128) -Output shape: (1, 294, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.output: torch.Size([1, 294, 4096]) -> torch.Size([1, 1, 294, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 15,884B, BPFP=0.4221 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,864B, BPFP=2.2020 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 59,668B, BPFP=1.5856 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 87,800B, BPFP=2.3331 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 73,940B, BPFP=1.9648 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,152B, BPFP=2.3690 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 74,216B, BPFP=1.9722 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,280B, BPFP=2.3459 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 62,152B, BPFP=1.6516 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,028B, BPFP=2.3923 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 275,440B, BPFP=1.8298 -⌛️ [2/4] FRONTEND: Frontend time: 2.379s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.output: torch.Size([1, 294, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.000s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.output: torch.Size([1, 294, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02439844 7.11018216 - layer.0.v_cache 0.00000027 0.00014285 - layer.1.k_cache 0.00289989 0.81173966 - layer.1.v_cache 0.00000083 0.00051099 - layer.2.k_cache 0.00119190 0.35693743 - layer.2.v_cache 0.00000115 0.00074753 - layer.3.k_cache 0.00131407 0.39940550 - layer.3.v_cache 0.00000220 0.00121437 - layer.4.k_cache 0.00351532 0.71936798 - layer.4.v_cache 0.00000333 0.00209430 - layer.4.output 0.00015801 0.05823265 - ------------------------------------------------------------------------------------- - TOTAL 0.00242567 0.68823381 - (elements=4,214,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4214784 -Total Bytes 999424 -BPFP 1.8970 bits/point -EBPFP 3.7940 equivalent bits/point -MSE 0.688234 ----------------------- -------------------------------------------------------- -Time: 4.395s Load: 0.016s, Pack+Encode: 2.379s, Decode+Unpack: 2.000s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 294, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6882 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample68-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst (77/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 338, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 338, 128) -Output shape: (1, 338, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.output: torch.Size([1, 338, 4096]) -> torch.Size([1, 1, 338, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,992B, BPFP=0.4159 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,812B, BPFP=2.2839 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,540B, BPFP=1.5611 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,896B, BPFP=2.4246 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,136B, BPFP=1.9909 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,552B, BPFP=2.4397 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,604B, BPFP=2.0249 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,428B, BPFP=2.4137 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,012B, BPFP=1.5951 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,336B, BPFP=2.4810 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 253,788B, BPFP=1.4665 -⌛️ [2/4] FRONTEND: Frontend time: 2.579s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.output: torch.Size([1, 338, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.303s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.output: torch.Size([1, 338, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02370679 6.79207020 - layer.0.v_cache 0.00000026 0.00013991 - layer.1.k_cache 0.00288373 0.82264267 - layer.1.v_cache 0.00000083 0.00052061 - layer.2.k_cache 0.00117924 0.34656978 - layer.2.v_cache 0.00000113 0.00074403 - layer.3.k_cache 0.00131114 0.40633325 - layer.3.v_cache 0.00000214 0.00121294 - layer.4.k_cache 0.00355340 0.71003172 - layer.4.v_cache 0.00000355 0.00212123 - layer.4.output 0.00015048 0.06026838 - ------------------------------------------------------------------------------------- - TOTAL 0.00237458 0.66596142 - (elements=4,845,568) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4845568 -Total Bytes 1103096 -BPFP 1.8212 bits/point -EBPFP 3.6424 equivalent bits/point -MSE 0.665961 ----------------------- -------------------------------------------------------- -Time: 4.898s Load: 0.017s, Pack+Encode: 2.579s, Decode+Unpack: 2.303s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 338, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6660 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample69-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst (78/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 340, 128) -Output shape: (1, 340, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.output: torch.Size([1, 340, 4096]) -> torch.Size([1, 1, 340, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,512B, BPFP=0.4254 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,180B, BPFP=2.2560 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 68,100B, BPFP=1.5648 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,556B, BPFP=2.3795 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,092B, BPFP=2.0012 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,624B, BPFP=2.4270 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,864B, BPFP=2.0189 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,568B, BPFP=2.4028 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,492B, BPFP=1.5968 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,368B, BPFP=2.4671 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 292,348B, BPFP=1.6794 -⌛️ [2/4] FRONTEND: Frontend time: 2.585s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.161s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02367318 7.05153378 - layer.0.v_cache 0.00000026 0.00014080 - layer.1.k_cache 0.00294777 0.80340702 - layer.1.v_cache 0.00000081 0.00052653 - layer.2.k_cache 0.00121079 0.35547371 - layer.2.v_cache 0.00000125 0.00075794 - layer.3.k_cache 0.00131075 0.39137645 - layer.3.v_cache 0.00000218 0.00122414 - layer.4.k_cache 0.00357831 0.72395262 - layer.4.v_cache 0.00000337 0.00214800 - layer.4.output 0.00014951 0.05370301 - ------------------------------------------------------------------------------------- - TOTAL 0.00238048 0.68181093 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 1142704 -BPFP 1.8755 bits/point -EBPFP 3.7510 equivalent bits/point -MSE 0.681811 ----------------------- -------------------------------------------------------- -Time: 4.763s Load: 0.017s, Pack+Encode: 2.585s, Decode+Unpack: 2.161s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6818 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample7-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst (79/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 320, 128) -Output shape: (1, 320, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.output: torch.Size([1, 320, 4096]) -> torch.Size([1, 1, 320, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,860B, BPFP=0.4116 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,352B, BPFP=2.0105 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 60,264B, BPFP=1.4713 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 87,560B, BPFP=2.1377 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 72,828B, BPFP=1.7780 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 88,692B, BPFP=2.1653 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 72,944B, BPFP=1.7809 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 87,820B, BPFP=2.1440 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 61,088B, BPFP=1.4914 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 89,416B, BPFP=2.1830 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 226,340B, BPFP=1.3815 -⌛️ [2/4] FRONTEND: Frontend time: 2.413s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.037s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02419060 6.86125183 - layer.0.v_cache 0.00000027 0.00013205 - layer.1.k_cache 0.00288080 0.83858471 - layer.1.v_cache 0.00000084 0.00050225 - layer.2.k_cache 0.00117023 0.36457438 - layer.2.v_cache 0.00000114 0.00071289 - layer.3.k_cache 0.00131996 0.39565532 - layer.3.v_cache 0.00000215 0.00112731 - layer.4.k_cache 0.00348745 0.70254083 - layer.4.v_cache 0.00000346 0.00199045 - layer.4.output 0.00014756 0.05124633 - ------------------------------------------------------------------------------------- - TOTAL 0.00240337 0.66943267 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 946164 -BPFP 1.6500 bits/point -EBPFP 3.3000 equivalent bits/point -MSE 0.669433 ----------------------- -------------------------------------------------------- -Time: 4.467s Load: 0.017s, Pack+Encode: 2.413s, Decode+Unpack: 2.037s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6694 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample70-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst (80/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 328, 128) -Output shape: (1, 328, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.output: torch.Size([1, 328, 4096]) -> torch.Size([1, 1, 328, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,560B, BPFP=0.4183 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,688B, BPFP=2.3030 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 64,436B, BPFP=1.5348 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,020B, BPFP=2.4538 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 81,908B, BPFP=1.9509 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,896B, BPFP=2.4747 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,684B, BPFP=2.0409 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,600B, BPFP=2.4676 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,640B, BPFP=1.5873 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,536B, BPFP=2.5137 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 255,912B, BPFP=1.5239 -⌛️ [2/4] FRONTEND: Frontend time: 2.604s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.232s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02411815 7.52655848 - layer.0.v_cache 0.00000026 0.00014122 - layer.1.k_cache 0.00293126 0.86748319 - layer.1.v_cache 0.00000079 0.00052167 - layer.2.k_cache 0.00115822 0.36095328 - layer.2.v_cache 0.00000116 0.00072735 - layer.3.k_cache 0.00132993 0.39593827 - layer.3.v_cache 0.00000214 0.00118954 - layer.4.k_cache 0.00358323 0.74086910 - layer.4.v_cache 0.00000336 0.00206564 - layer.4.output 0.00014415 0.06085072 - ------------------------------------------------------------------------------------- - TOTAL 0.00240751 0.72427505 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 1084880 -BPFP 1.8457 bits/point -EBPFP 3.6915 equivalent bits/point -MSE 0.724275 ----------------------- -------------------------------------------------------- -Time: 4.854s Load: 0.018s, Pack+Encode: 2.604s, Decode+Unpack: 2.232s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7243 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample71-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst (81/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 328, 128) -Output shape: (1, 328, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.output: torch.Size([1, 328, 4096]) -> torch.Size([1, 1, 328, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,088B, BPFP=0.4070 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,692B, BPFP=2.3031 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 65,924B, BPFP=1.5702 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,644B, BPFP=2.4687 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 81,632B, BPFP=1.9444 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,748B, BPFP=2.4950 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,424B, BPFP=2.0347 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,008B, BPFP=2.4773 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,760B, BPFP=1.5425 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,876B, BPFP=2.4980 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 243,876B, BPFP=1.4522 -⌛️ [2/4] FRONTEND: Frontend time: 2.647s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.201s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02582944 6.69372707 - layer.0.v_cache 0.00000026 0.00014033 - layer.1.k_cache 0.00293683 0.86205273 - layer.1.v_cache 0.00000081 0.00052093 - layer.2.k_cache 0.00118071 0.36613743 - layer.2.v_cache 0.00000117 0.00075795 - layer.3.k_cache 0.00133336 0.40541356 - layer.3.v_cache 0.00000214 0.00117516 - layer.4.k_cache 0.00361319 0.74335140 - layer.4.v_cache 0.00000321 0.00203449 - layer.4.output 0.00015382 0.06208845 - ------------------------------------------------------------------------------------- - TOTAL 0.00253689 0.66597606 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 1072672 -BPFP 1.8250 bits/point -EBPFP 3.6499 equivalent bits/point -MSE 0.665976 ----------------------- -------------------------------------------------------- -Time: 4.865s Load: 0.017s, Pack+Encode: 2.647s, Decode+Unpack: 2.201s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6660 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample72-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst (82/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 352, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 352, 128) -Output shape: (1, 352, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.output: torch.Size([1, 352, 4096]) -> torch.Size([1, 1, 352, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,636B, BPFP=0.4136 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,028B, BPFP=2.1979 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,632B, BPFP=1.5676 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,576B, BPFP=2.3210 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,528B, BPFP=1.9648 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,652B, BPFP=2.3671 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,504B, BPFP=1.9865 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,536B, BPFP=2.3423 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 75,120B, BPFP=1.6673 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,768B, BPFP=2.3919 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 272,140B, BPFP=1.5100 -⌛️ [2/4] FRONTEND: Frontend time: 2.575s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.output: torch.Size([1, 352, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.200s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.output: torch.Size([1, 352, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02395736 7.04263306 - layer.0.v_cache 0.00000026 0.00014295 - layer.1.k_cache 0.00288394 0.82017682 - layer.1.v_cache 0.00000077 0.00052143 - layer.2.k_cache 0.00114480 0.34845660 - layer.2.v_cache 0.00000114 0.00075526 - layer.3.k_cache 0.00130173 0.40903486 - layer.3.v_cache 0.00000215 0.00122658 - layer.4.k_cache 0.00352615 0.73959247 - layer.4.v_cache 0.00000336 0.00219798 - layer.4.output 0.00014232 0.05690503 - ------------------------------------------------------------------------------------- - TOTAL 0.00238507 0.68516844 - (elements=5,046,272) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5046272 -Total Bytes 1138120 -BPFP 1.8043 bits/point -EBPFP 3.6086 equivalent bits/point -MSE 0.685168 ----------------------- -------------------------------------------------------- -Time: 4.794s Load: 0.019s, Pack+Encode: 2.575s, Decode+Unpack: 2.200s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 352, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6852 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample73-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst (83/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,908B, BPFP=0.4331 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 95,596B, BPFP=2.3122 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 63,244B, BPFP=1.5297 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 101,436B, BPFP=2.4535 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 79,416B, BPFP=1.9209 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 102,816B, BPFP=2.4868 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 83,228B, BPFP=2.0131 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 102,004B, BPFP=2.4672 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,628B, BPFP=1.5874 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 103,876B, BPFP=2.5125 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 278,340B, BPFP=1.6831 -⌛️ [2/4] FRONTEND: Frontend time: 2.652s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.237s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435790 6.76470286 - layer.0.v_cache 0.00000027 0.00014050 - layer.1.k_cache 0.00290123 0.82091879 - layer.1.v_cache 0.00000082 0.00051728 - layer.2.k_cache 0.00116542 0.34858472 - layer.2.v_cache 0.00000118 0.00076255 - layer.3.k_cache 0.00132756 0.39601178 - layer.3.v_cache 0.00000224 0.00124366 - layer.4.k_cache 0.00351721 0.70189811 - layer.4.v_cache 0.00000335 0.00217873 - layer.4.output 0.00015064 0.05688526 - ------------------------------------------------------------------------------------- - TOTAL 0.00241998 0.66175000 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 1093492 -BPFP 1.8892 bits/point -EBPFP 3.7784 equivalent bits/point -MSE 0.661750 ----------------------- -------------------------------------------------------- -Time: 4.911s Load: 0.022s, Pack+Encode: 2.652s, Decode+Unpack: 2.237s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6617 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample74-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst (84/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 327, 128) -Output shape: (1, 327, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.output: torch.Size([1, 327, 4096]) -> torch.Size([1, 1, 327, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,064B, BPFP=0.4316 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,188B, BPFP=2.2981 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 64,988B, BPFP=1.5527 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,264B, BPFP=2.4432 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 79,820B, BPFP=1.9070 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 102,972B, BPFP=2.4601 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 83,856B, BPFP=2.0034 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,088B, BPFP=2.4629 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,608B, BPFP=1.5436 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,188B, BPFP=2.4892 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 269,080B, BPFP=1.6072 -⌛️ [2/4] FRONTEND: Frontend time: 2.618s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.301s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02420299 6.86014699 - layer.0.v_cache 0.00000027 0.00013978 - layer.1.k_cache 0.00291079 0.83788055 - layer.1.v_cache 0.00000081 0.00052722 - layer.2.k_cache 0.00118771 0.35570577 - layer.2.v_cache 0.00000120 0.00076099 - layer.3.k_cache 0.00132021 0.39961411 - layer.3.v_cache 0.00000219 0.00122941 - layer.4.k_cache 0.00351552 0.69702624 - layer.4.v_cache 0.00000327 0.00208958 - layer.4.output 0.00014986 0.05551564 - ------------------------------------------------------------------------------------- - TOTAL 0.00241031 0.66979880 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 1089116 -BPFP 1.8586 bits/point -EBPFP 3.7172 equivalent bits/point -MSE 0.669799 ----------------------- -------------------------------------------------------- -Time: 4.936s Load: 0.016s, Pack+Encode: 2.618s, Decode+Unpack: 2.301s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6698 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample75-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst (85/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,576B, BPFP=0.4161 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,948B, BPFP=2.2952 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,168B, BPFP=1.5902 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,788B, BPFP=2.4334 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,796B, BPFP=1.9838 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,836B, BPFP=2.4582 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 84,172B, BPFP=1.9927 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,556B, BPFP=2.4516 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,488B, BPFP=1.5741 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,908B, BPFP=2.4836 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 270,296B, BPFP=1.5998 -⌛️ [2/4] FRONTEND: Frontend time: 2.686s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.298s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02537568 7.03235233 - layer.0.v_cache 0.00000027 0.00013877 - layer.1.k_cache 0.00294190 0.84578626 - layer.1.v_cache 0.00000076 0.00049504 - layer.2.k_cache 0.00116180 0.35741647 - layer.2.v_cache 0.00000113 0.00071442 - layer.3.k_cache 0.00131353 0.39243973 - layer.3.v_cache 0.00000217 0.00116053 - layer.4.k_cache 0.00362690 0.74854112 - layer.4.v_cache 0.00000322 0.00202215 - layer.4.output 0.00015895 0.06203181 - ------------------------------------------------------------------------------------- - TOTAL 0.00250451 0.68779958 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 1101532 -BPFP 1.8627 bits/point -EBPFP 3.7254 equivalent bits/point -MSE 0.687800 ----------------------- -------------------------------------------------------- -Time: 5.001s Load: 0.017s, Pack+Encode: 2.686s, Decode+Unpack: 2.298s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6878 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample76-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst (86/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,948B, BPFP=0.4025 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,164B, BPFP=2.3073 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 65,948B, BPFP=1.5660 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,616B, BPFP=2.4605 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,560B, BPFP=1.9842 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,264B, BPFP=2.4759 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,364B, BPFP=2.0271 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 103,560B, BPFP=2.4592 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,704B, BPFP=1.5602 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,352B, BPFP=2.5017 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 234,572B, BPFP=1.3925 -⌛️ [2/4] FRONTEND: Frontend time: 2.609s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.306s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02321163 7.26413047 - layer.0.v_cache 0.00000026 0.00013937 - layer.1.k_cache 0.00292509 0.81941265 - layer.1.v_cache 0.00000078 0.00051563 - layer.2.k_cache 0.00118155 0.37037376 - layer.2.v_cache 0.00000112 0.00072733 - layer.3.k_cache 0.00133849 0.40014481 - layer.3.v_cache 0.00000208 0.00114777 - layer.4.k_cache 0.00360187 0.70747705 - layer.4.v_cache 0.00000304 0.00199748 - layer.4.output 0.00015018 0.06212007 - ------------------------------------------------------------------------------------- - TOTAL 0.00234762 0.70103904 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 1066052 -BPFP 1.8082 bits/point -EBPFP 3.6164 equivalent bits/point -MSE 0.701039 ----------------------- -------------------------------------------------------- -Time: 4.932s Load: 0.017s, Pack+Encode: 2.609s, Decode+Unpack: 2.306s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.7010 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample77-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst (87/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,548B, BPFP=0.4117 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,916B, BPFP=2.2972 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,004B, BPFP=1.5720 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 102,780B, BPFP=2.4113 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 86,268B, BPFP=2.0239 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,332B, BPFP=2.4712 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 87,564B, BPFP=2.0543 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,968B, BPFP=2.4627 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,816B, BPFP=1.5676 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,320B, BPFP=2.4944 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 248,884B, BPFP=1.4598 -⌛️ [2/4] FRONTEND: Frontend time: 2.605s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.260s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02306361 6.74375572 - layer.0.v_cache 0.00000026 0.00014244 - layer.1.k_cache 0.00286224 0.82022507 - layer.1.v_cache 0.00000078 0.00051236 - layer.2.k_cache 0.00116467 0.36823069 - layer.2.v_cache 0.00000115 0.00073780 - layer.3.k_cache 0.00135416 0.42168415 - layer.3.v_cache 0.00000234 0.00123369 - layer.4.k_cache 0.00356469 0.73796343 - layer.4.v_cache 0.00000325 0.00209341 - layer.4.output 0.00015110 0.05598069 - ------------------------------------------------------------------------------------- - TOTAL 0.00233011 0.66575011 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 1091400 -BPFP 1.8289 bits/point -EBPFP 3.6579 equivalent bits/point -MSE 0.665750 ----------------------- -------------------------------------------------------- -Time: 4.881s Load: 0.017s, Pack+Encode: 2.605s, Decode+Unpack: 2.260s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6658 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample78-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst (88/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 304, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 304, 128) -Output shape: (1, 304, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.output: torch.Size([1, 304, 4096]) -> torch.Size([1, 1, 304, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,060B, BPFP=0.4127 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 82,952B, BPFP=2.1318 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,596B, BPFP=1.5830 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,228B, BPFP=2.2674 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,104B, BPFP=1.9044 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 89,648B, BPFP=2.3039 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 73,684B, BPFP=1.8936 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 88,704B, BPFP=2.2796 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,832B, BPFP=1.6404 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,488B, BPFP=2.3255 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 238,628B, BPFP=1.5331 -⌛️ [2/4] FRONTEND: Frontend time: 2.431s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.output: torch.Size([1, 304, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.010s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.output: torch.Size([1, 304, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02416075 7.05518863 - layer.0.v_cache 0.00000027 0.00014063 - layer.1.k_cache 0.00290015 0.73321563 - layer.1.v_cache 0.00000081 0.00051616 - layer.2.k_cache 0.00116609 0.36671528 - layer.2.v_cache 0.00000115 0.00074319 - layer.3.k_cache 0.00131777 0.40839885 - layer.3.v_cache 0.00000216 0.00121151 - layer.4.k_cache 0.00349415 0.74656130 - layer.4.v_cache 0.00000336 0.00213078 - layer.4.output 0.00015143 0.05611076 - ------------------------------------------------------------------------------------- - TOTAL 0.00240374 0.68137607 - (elements=4,358,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4358144 -Total Bytes 967924 -BPFP 1.7768 bits/point -EBPFP 3.5535 equivalent bits/point -MSE 0.681376 ----------------------- -------------------------------------------------------- -Time: 4.456s Load: 0.016s, Pack+Encode: 2.431s, Decode+Unpack: 2.010s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 304, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6814 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample79-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst (89/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,304B, BPFP=0.4074 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,148B, BPFP=2.2068 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 71,000B, BPFP=1.5803 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,664B, BPFP=2.3296 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,072B, BPFP=1.9603 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,676B, BPFP=2.3744 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,788B, BPFP=1.9762 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,100B, BPFP=2.3616 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 72,056B, BPFP=1.6038 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,872B, BPFP=2.4010 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 271,764B, BPFP=1.5122 -⌛️ [2/4] FRONTEND: Frontend time: 2.584s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.187s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02287757 6.93663612 - layer.0.v_cache 0.00000026 0.00013568 - layer.1.k_cache 0.00288319 0.83344723 - layer.1.v_cache 0.00000079 0.00049318 - layer.2.k_cache 0.00116994 0.35187470 - layer.2.v_cache 0.00000119 0.00074845 - layer.3.k_cache 0.00131288 0.39757788 - layer.3.v_cache 0.00000224 0.00122043 - layer.4.k_cache 0.00358885 0.77544966 - layer.4.v_cache 0.00000339 0.00210846 - layer.4.output 0.00014945 0.06015171 - ------------------------------------------------------------------------------------- - TOTAL 0.00231701 0.68144990 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 1134444 -BPFP 1.8036 bits/point -EBPFP 3.6072 equivalent bits/point -MSE 0.681450 ----------------------- -------------------------------------------------------- -Time: 4.789s Load: 0.018s, Pack+Encode: 2.584s, Decode+Unpack: 2.187s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6814 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample8-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst (90/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,096B, BPFP=0.4011 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,232B, BPFP=2.2812 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,736B, BPFP=1.5657 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 101,668B, BPFP=2.3852 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,620B, BPFP=1.9853 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,328B, BPFP=2.4711 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,580B, BPFP=2.0312 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,120B, BPFP=2.4428 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 67,056B, BPFP=1.5732 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,068B, BPFP=2.4885 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 226,368B, BPFP=1.3277 -⌛️ [2/4] FRONTEND: Frontend time: 2.614s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.257s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02328842 7.09296552 - layer.0.v_cache 0.00000026 0.00014105 - layer.1.k_cache 0.00290993 0.83825024 - layer.1.v_cache 0.00000078 0.00050319 - layer.2.k_cache 0.00115179 0.35509308 - layer.2.v_cache 0.00000114 0.00074393 - layer.3.k_cache 0.00134347 0.40717620 - layer.3.v_cache 0.00000212 0.00119816 - layer.4.k_cache 0.00362086 0.73832098 - layer.4.v_cache 0.00000365 0.00206017 - layer.4.output 0.00013924 0.06125849 - ------------------------------------------------------------------------------------- - TOTAL 0.00234853 0.69153475 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 1062872 -BPFP 1.7811 bits/point -EBPFP 3.5623 equivalent bits/point -MSE 0.691535 ----------------------- -------------------------------------------------------- -Time: 4.892s Load: 0.021s, Pack+Encode: 2.614s, Decode+Unpack: 2.257s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6915 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst (91/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 319, 128) -Output shape: (1, 319, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.output: torch.Size([1, 319, 4096]) -> torch.Size([1, 1, 319, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 16,872B, BPFP=0.4132 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 83,332B, BPFP=2.0409 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 61,328B, BPFP=1.5020 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 88,716B, BPFP=2.1727 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 74,232B, BPFP=1.8180 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 90,308B, BPFP=2.2117 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 73,820B, BPFP=1.8079 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 89,472B, BPFP=2.1912 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 63,512B, BPFP=1.5554 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 90,904B, BPFP=2.2263 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 257,040B, BPFP=1.5738 -⌛️ [2/4] FRONTEND: Frontend time: 2.451s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) -⌛️ [3/4] BACKEND: Backend time: 1.963s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02501136 6.50889774 - layer.0.v_cache 0.00000027 0.00013672 - layer.1.k_cache 0.00284013 0.82973803 - layer.1.v_cache 0.00000084 0.00050651 - layer.2.k_cache 0.00117936 0.35149788 - layer.2.v_cache 0.00000120 0.00075087 - layer.3.k_cache 0.00130692 0.42962747 - layer.3.v_cache 0.00000226 0.00119590 - layer.4.k_cache 0.00361670 0.72948339 - layer.4.v_cache 0.00000371 0.00213379 - layer.4.output 0.00016871 0.05782402 - ------------------------------------------------------------------------------------- - TOTAL 0.00247411 0.64894745 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 989536 -BPFP 1.7310 bits/point -EBPFP 3.4620 equivalent bits/point -MSE 0.648947 ----------------------- -------------------------------------------------------- -Time: 4.430s Load: 0.016s, Pack+Encode: 2.451s, Decode+Unpack: 1.963s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6489 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample81-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst (92/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 353, 128) -Output shape: (1, 353, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.output: torch.Size([1, 353, 4096]) -> torch.Size([1, 1, 353, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 19,668B, BPFP=0.4353 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,604B, BPFP=2.2044 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,984B, BPFP=1.5489 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,788B, BPFP=2.3191 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,432B, BPFP=1.9572 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,524B, BPFP=2.3576 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 89,464B, BPFP=1.9800 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,440B, BPFP=2.3336 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 74,148B, BPFP=1.6410 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,672B, BPFP=2.3830 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 331,836B, BPFP=1.8360 -⌛️ [2/4] FRONTEND: Frontend time: 2.607s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.217s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02381471 6.87225497 - layer.0.v_cache 0.00000027 0.00014202 - layer.1.k_cache 0.00292436 0.91333864 - layer.1.v_cache 0.00000085 0.00052576 - layer.2.k_cache 0.00118067 0.35683439 - layer.2.v_cache 0.00000120 0.00074707 - layer.3.k_cache 0.00130453 0.43315328 - layer.3.v_cache 0.00000223 0.00121425 - layer.4.k_cache 0.00350570 0.73291258 - layer.4.v_cache 0.00000328 0.00207091 - layer.4.output 0.00016368 0.05686635 - ------------------------------------------------------------------------------------- - TOTAL 0.00238518 0.68147566 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 1197560 -BPFP 1.8931 bits/point -EBPFP 3.7863 equivalent bits/point -MSE 0.681476 ----------------------- -------------------------------------------------------- -Time: 4.841s Load: 0.018s, Pack+Encode: 2.607s, Decode+Unpack: 2.217s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6815 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample82-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst (93/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,624B, BPFP=0.4230 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,000B, BPFP=2.2484 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 69,880B, BPFP=1.5870 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 106,116B, BPFP=2.4100 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,712B, BPFP=2.0147 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,548B, BPFP=2.4198 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,736B, BPFP=2.0153 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 106,452B, BPFP=2.4176 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 69,888B, BPFP=1.5872 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,784B, BPFP=2.4479 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 300,404B, BPFP=1.7056 -⌛️ [2/4] FRONTEND: Frontend time: 2.633s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.161s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02504009 6.88101267 - layer.0.v_cache 0.00000027 0.00014513 - layer.1.k_cache 0.00288154 0.85867815 - layer.1.v_cache 0.00000082 0.00053222 - layer.2.k_cache 0.00115445 0.33895308 - layer.2.v_cache 0.00000117 0.00075310 - layer.3.k_cache 0.00131257 0.40257015 - layer.3.v_cache 0.00000224 0.00125142 - layer.4.k_cache 0.00356118 0.71179363 - layer.4.v_cache 0.00000339 0.00211209 - layer.4.output 0.00014496 0.05522676 - ------------------------------------------------------------------------------------- - TOTAL 0.00246697 0.67276491 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 1162144 -BPFP 1.8852 bits/point -EBPFP 3.7705 equivalent bits/point -MSE 0.672765 ----------------------- -------------------------------------------------------- -Time: 4.811s Load: 0.017s, Pack+Encode: 2.633s, Decode+Unpack: 2.161s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6728 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample83-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst (94/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 321, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 321, 128) -Output shape: (1, 321, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.output: torch.Size([1, 321, 4096]) -> torch.Size([1, 1, 321, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,084B, BPFP=0.4158 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 95,120B, BPFP=2.3150 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 62,368B, BPFP=1.5179 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 100,300B, BPFP=2.4411 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 75,852B, BPFP=1.8461 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 102,484B, BPFP=2.4943 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 80,552B, BPFP=1.9605 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 102,068B, BPFP=2.4841 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,344B, BPFP=1.5660 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 103,328B, BPFP=2.5148 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 230,948B, BPFP=1.4052 -⌛️ [2/4] FRONTEND: Frontend time: 2.652s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.output: torch.Size([1, 321, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.288s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.output: torch.Size([1, 321, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02357271 6.86493852 - layer.0.v_cache 0.00000026 0.00014038 - layer.1.k_cache 0.00287206 0.78598346 - layer.1.v_cache 0.00000077 0.00050695 - layer.2.k_cache 0.00116438 0.35816330 - layer.2.v_cache 0.00000114 0.00073524 - layer.3.k_cache 0.00132116 0.39329198 - layer.3.v_cache 0.00000214 0.00119418 - layer.4.k_cache 0.00354096 0.69516115 - layer.4.v_cache 0.00000337 0.00211064 - layer.4.output 0.00013764 0.05840441 - ------------------------------------------------------------------------------------- - TOTAL 0.00235925 0.66684596 - (elements=4,601,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4601856 -Total Bytes 1034448 -BPFP 1.7983 bits/point -EBPFP 3.5966 equivalent bits/point -MSE 0.666846 ----------------------- -------------------------------------------------------- -Time: 4.957s Load: 0.016s, Pack+Encode: 2.652s, Decode+Unpack: 2.288s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 321, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6668 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample85-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst (95/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,492B, BPFP=0.4154 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 96,888B, BPFP=2.3007 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 64,896B, BPFP=1.5410 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,672B, BPFP=2.4618 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 83,876B, BPFP=1.9917 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,828B, BPFP=2.4893 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,936B, BPFP=2.0407 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,364B, BPFP=2.4782 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 65,376B, BPFP=1.5524 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 106,152B, BPFP=2.5207 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 236,284B, BPFP=1.4027 -⌛️ [2/4] FRONTEND: Frontend time: 2.609s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.223s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02395787 6.48219703 - layer.0.v_cache 0.00000026 0.00013817 - layer.1.k_cache 0.00285945 0.81566539 - layer.1.v_cache 0.00000078 0.00051380 - layer.2.k_cache 0.00114467 0.35601452 - layer.2.v_cache 0.00000113 0.00074694 - layer.3.k_cache 0.00130198 0.39494811 - layer.3.v_cache 0.00000219 0.00122881 - layer.4.k_cache 0.00350851 0.69742600 - layer.4.v_cache 0.00000338 0.00215801 - layer.4.output 0.00014624 0.05689131 - ------------------------------------------------------------------------------------- - TOTAL 0.00238323 0.64132871 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 1069764 -BPFP 1.8145 bits/point -EBPFP 3.6290 equivalent bits/point -MSE 0.641329 ----------------------- -------------------------------------------------------- -Time: 4.849s Load: 0.017s, Pack+Encode: 2.609s, Decode+Unpack: 2.223s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6413 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample86-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst (96/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 350, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.024s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 350, 128) -Output shape: (1, 350, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.output: torch.Size([1, 350, 4096]) -> torch.Size([1, 1, 350, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,896B, BPFP=0.3995 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 99,284B, BPFP=2.2162 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 70,028B, BPFP=1.5631 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,452B, BPFP=2.3315 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 88,612B, BPFP=1.9779 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 106,340B, BPFP=2.3737 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 88,508B, BPFP=1.9756 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 105,972B, BPFP=2.3654 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 73,396B, BPFP=1.6383 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 107,980B, BPFP=2.4103 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 266,528B, BPFP=1.4873 -⌛️ [2/4] FRONTEND: Frontend time: 2.624s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.output: torch.Size([1, 350, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.202s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.output: torch.Size([1, 350, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02252116 6.69065639 - layer.0.v_cache 0.00000028 0.00014190 - layer.1.k_cache 0.00283737 0.86655648 - layer.1.v_cache 0.00000078 0.00047318 - layer.2.k_cache 0.00114986 0.34463682 - layer.2.v_cache 0.00000111 0.00067812 - layer.3.k_cache 0.00128401 0.39077772 - layer.3.v_cache 0.00000216 0.00113282 - layer.4.k_cache 0.00352053 0.72493905 - layer.4.v_cache 0.00000313 0.00195614 - layer.4.output 0.00014459 0.05678670 - ------------------------------------------------------------------------------------- - TOTAL 0.00227848 0.66064967 - (elements=5,017,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5017600 -Total Bytes 1128996 -BPFP 1.8001 bits/point -EBPFP 3.6001 equivalent bits/point -MSE 0.660650 ----------------------- -------------------------------------------------------- -Time: 4.850s Load: 0.024s, Pack+Encode: 2.624s, Decode+Unpack: 2.202s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 350, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6606 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample87-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst (97/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,348B, BPFP=0.4196 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 94,736B, BPFP=2.2914 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 63,996B, BPFP=1.5479 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 101,896B, BPFP=2.4646 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 80,724B, BPFP=1.9525 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 103,216B, BPFP=2.4965 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 82,564B, BPFP=1.9970 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 101,388B, BPFP=2.4523 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 64,536B, BPFP=1.5610 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 104,136B, BPFP=2.5188 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 243,044B, BPFP=1.4696 -⌛️ [2/4] FRONTEND: Frontend time: 2.630s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.113s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02344667 7.23112559 - layer.0.v_cache 0.00000026 0.00013840 - layer.1.k_cache 0.00293942 0.74341586 - layer.1.v_cache 0.00000088 0.00051560 - layer.2.k_cache 0.00118806 0.34908793 - layer.2.v_cache 0.00000115 0.00074160 - layer.3.k_cache 0.00133183 0.40365119 - layer.3.v_cache 0.00000219 0.00122282 - layer.4.k_cache 0.00365354 0.75506469 - layer.4.v_cache 0.00000330 0.00208840 - layer.4.output 0.00014070 0.05864183 - ------------------------------------------------------------------------------------- - TOTAL 0.00236644 0.69440139 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 1057584 -BPFP 1.8272 bits/point -EBPFP 3.6543 equivalent bits/point -MSE 0.694401 ----------------------- -------------------------------------------------------- -Time: 4.760s Load: 0.017s, Pack+Encode: 2.630s, Decode+Unpack: 2.113s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6944 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst (98/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 419, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.026s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 419, 128) -Output shape: (1, 419, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.output: torch.Size([1, 419, 4096]) -> torch.Size([1, 1, 419, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 21,028B, BPFP=0.3921 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 114,112B, BPFP=2.1277 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 83,128B, BPFP=1.5500 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 119,964B, BPFP=2.2368 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 102,696B, BPFP=1.9148 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 122,132B, BPFP=2.2772 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 102,844B, BPFP=1.9176 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 121,792B, BPFP=2.2709 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 87,052B, BPFP=1.6231 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 124,576B, BPFP=2.3228 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 282,796B, BPFP=1.3182 -⌛️ [2/4] FRONTEND: Frontend time: 2.823s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.output: torch.Size([1, 419, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.394s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.output: torch.Size([1, 419, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02248870 6.72551893 - layer.0.v_cache 0.00000028 0.00014380 - layer.1.k_cache 0.00282601 0.80111403 - layer.1.v_cache 0.00000075 0.00046407 - layer.2.k_cache 0.00117249 0.34878129 - layer.2.v_cache 0.00000110 0.00065895 - layer.3.k_cache 0.00129145 0.40531328 - layer.3.v_cache 0.00000206 0.00112204 - layer.4.k_cache 0.00360397 0.69946508 - layer.4.v_cache 0.00000307 0.00192408 - layer.4.output 0.00013808 0.05294993 - ------------------------------------------------------------------------------------- - TOTAL 0.00228159 0.65687895 - (elements=6,006,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 6006784 -Total Bytes 1282120 -BPFP 1.7076 bits/point -EBPFP 3.4151 equivalent bits/point -MSE 0.656879 ----------------------- -------------------------------------------------------- -Time: 5.243s Load: 0.026s, Pack+Encode: 2.823s, Decode+Unpack: 2.394s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 419, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6569 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample9-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst (99/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 18,340B, BPFP=0.4264 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 97,840B, BPFP=2.2749 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 67,140B, BPFP=1.5611 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 103,432B, BPFP=2.4049 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 87,052B, BPFP=2.0241 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 105,076B, BPFP=2.4432 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 86,568B, BPFP=2.0128 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,344B, BPFP=2.4262 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 68,680B, BPFP=1.5969 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,588B, BPFP=2.4551 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 258,388B, BPFP=1.5020 -⌛️ [2/4] FRONTEND: Frontend time: 2.619s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.216s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02369028 7.10539682 - layer.0.v_cache 0.00000026 0.00014055 - layer.1.k_cache 0.00291391 0.84941001 - layer.1.v_cache 0.00000078 0.00051947 - layer.2.k_cache 0.00115383 0.35286797 - layer.2.v_cache 0.00000117 0.00075834 - layer.3.k_cache 0.00133386 0.39812224 - layer.3.v_cache 0.00000214 0.00121648 - layer.4.k_cache 0.00353874 0.72096153 - layer.4.v_cache 0.00000342 0.00214081 - layer.4.output 0.00014590 0.05918443 - ------------------------------------------------------------------------------------- - TOTAL 0.00237300 0.69059085 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 1102448 -BPFP 1.8310 bits/point -EBPFP 3.6619 equivalent bits/point -MSE 0.690591 ----------------------- -------------------------------------------------------- -Time: 4.850s Load: 0.016s, Pack+Encode: 2.619s, Decode+Unpack: 2.216s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6906 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample93-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst (100/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 17,200B, BPFP=0.4084 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 98,076B, BPFP=2.3289 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 66,248B, BPFP=1.5731 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 104,032B, BPFP=2.4704 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 84,736B, BPFP=2.0122 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 104,968B, BPFP=2.4926 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 85,600B, BPFP=2.0327 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 104,224B, BPFP=2.4749 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 66,856B, BPFP=1.5876 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 105,964B, BPFP=2.5162 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 227,604B, BPFP=1.3512 -⌛️ [2/4] FRONTEND: Frontend time: 2.623s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 2.207s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02541956 7.10208937 - layer.0.v_cache 0.00000027 0.00014010 - layer.1.k_cache 0.00292153 0.83156616 - layer.1.v_cache 0.00000078 0.00051665 - layer.2.k_cache 0.00117184 0.36085668 - layer.2.v_cache 0.00000114 0.00074243 - layer.3.k_cache 0.00131754 0.40214780 - layer.3.v_cache 0.00000217 0.00121948 - layer.4.k_cache 0.00353394 0.70682649 - layer.4.v_cache 0.00000342 0.00210255 - layer.4.output 0.00014204 0.06025474 - ------------------------------------------------------------------------------------- - TOTAL 0.00249574 0.68923047 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture elic-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 1065508 -BPFP 1.8073 bits/point -EBPFP 3.6145 equivalent bits/point -MSE 0.689230 ----------------------- -------------------------------------------------------- -Time: 4.851s Load: 0.021s, Pack+Encode: 2.623s, Decode+Unpack: 2.207s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 0.6892 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.01/elic-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 1.7975 bits/point -Avg EBPFP 3.5950 equivalent bits/point -Avg MSE 0.674355 -Avg Time 4.786s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:10fbd275dc203d5a4409a16733fa160901bcf97e50868058eb952b3875822c06 +size 1124079