diff --git a/.gitattributes b/.gitattributes index ce3df52ab3de0f2857bff14b4dc4e5e1dda46fee..9130a23fe8436c3c7fe710cf010662b5cb22f0ee 100644 --- a/.gitattributes +++ b/.gitattributes @@ -5114,3 +5114,5 @@ lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample57-layer4-item lambda0.007/elic-featurecoding-8bit-individual/fc_hellaswag/sample23-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.007/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample30-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.007/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample32-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +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 diff --git a/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log b/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log index 943f0ecbef857d73acea148988b6686fe7320772..2e28294ec74eb5c7595f6446722518d1c5cbedf9 100644 --- a/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log +++ b/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_hyperprior-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: hyperprior-featurecoding - handler: qwen - checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_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/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 599 -Loaded hyperprior-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 hyperprior-featurecoding -Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_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/lambda0.007/hyperprior-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: 0.011s - ------------------------------------------------------------- -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: 7,944B, BPFP=0.3587 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 37,736B, BPFP=1.7041 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 24,500B, BPFP=1.1064 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 40,408B, BPFP=1.8248 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 28,576B, BPFP=1.2905 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 41,176B, BPFP=1.8595 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 29,748B, BPFP=1.3434 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 40,436B, BPFP=1.8260 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 24,900B, BPFP=1.1245 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 41,476B, BPFP=1.8730 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 112,256B, BPFP=1.2673 -⌛️ [2/4] FRONTEND: Frontend time: 0.495s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.551s - -[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, 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 12.23617430 - layer.0.v_cache 0.00000027 0.00024740 - layer.1.k_cache 0.00314874 0.88755763 - layer.1.v_cache 0.00000072 0.00084044 - layer.2.k_cache 0.00114712 0.46391733 - layer.2.v_cache 0.00000110 0.00125916 - layer.3.k_cache 0.00139354 0.54422376 - layer.3.v_cache 0.00000204 0.00204967 - layer.4.k_cache 0.00353492 1.03861638 - layer.4.v_cache 0.00000301 0.00345948 - layer.4.output 0.00018661 0.07198587 - ------------------------------------------------------------------------------------- - TOTAL 0.00245579 1.10473493 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 429156 -BPFP 1.3843 bits/point -EBPFP 2.7686 equivalent bits/point -MSE 1.104735 ----------------------- -------------------------------------------------------- -Time: 1.056s Load: 0.011s, Pack+Encode: 0.495s, Decode+Unpack: 0.551s ----------------------- -------------------------------------------------------- -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 1.1047 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 5,016B, BPFP=0.3595 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,160B, BPFP=1.8033 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,420B, BPFP=1.1769 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,036B, BPFP=1.9378 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,040B, BPFP=1.3647 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,260B, BPFP=1.9538 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,844B, BPFP=1.4223 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,588B, BPFP=1.9057 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,672B, BPFP=1.1950 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,432B, BPFP=1.9662 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,284B, BPFP=1.3131 -⌛️ [2/4] FRONTEND: Frontend time: 0.270s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.308s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.38517327 - layer.0.v_cache 0.00000027 0.00025675 - layer.1.k_cache 0.00330346 1.01682526 - layer.1.v_cache 0.00000080 0.00086010 - layer.2.k_cache 0.00115767 0.47720988 - layer.2.v_cache 0.00000114 0.00122824 - layer.3.k_cache 0.00132842 0.53952824 - layer.3.v_cache 0.00000211 0.00199592 - layer.4.k_cache 0.00335301 1.08328156 - layer.4.v_cache 0.00000290 0.00326440 - layer.4.output 0.00024947 0.08458482 - ------------------------------------------------------------------------------------- - TOTAL 0.00248590 1.13199735 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 283752 -BPFP 1.4527 bits/point -EBPFP 2.9054 equivalent bits/point -MSE 1.131997 ----------------------- -------------------------------------------------------- -Time: 0.584s Load: 0.006s, Pack+Encode: 0.270s, Decode+Unpack: 0.308s ----------------------- -------------------------------------------------------- -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 1.1320 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,576B, BPFP=0.3686 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,728B, BPFP=1.8305 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,000B, BPFP=1.1276 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,360B, BPFP=1.9620 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,648B, BPFP=1.3409 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,752B, BPFP=1.9936 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,264B, BPFP=1.3905 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,112B, BPFP=1.9420 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,624B, BPFP=1.1778 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,988B, BPFP=2.0126 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,912B, BPFP=1.3272 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.294s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.67007289 - layer.0.v_cache 0.00000028 0.00025768 - layer.1.k_cache 0.00331373 0.92635794 - layer.1.v_cache 0.00000086 0.00093699 - layer.2.k_cache 0.00113316 0.50183200 - layer.2.v_cache 0.00000115 0.00132194 - layer.3.k_cache 0.00137718 0.55760366 - layer.3.v_cache 0.00000224 0.00217131 - layer.4.k_cache 0.00334605 1.04256738 - layer.4.v_cache 0.00000324 0.00381484 - layer.4.output 0.00031880 0.08092673 - ------------------------------------------------------------------------------------- - TOTAL 0.00258942 1.14504597 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 253964 -BPFP 1.4610 bits/point -EBPFP 2.9221 equivalent bits/point -MSE 1.145046 ----------------------- -------------------------------------------------------- -Time: 0.509s Load: 0.007s, Pack+Encode: 0.208s, Decode+Unpack: 0.294s ----------------------- -------------------------------------------------------- -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 1.1450 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 2,844B, BPFP=0.4444 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 12,188B, BPFP=1.9044 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 7,808B, BPFP=1.2200 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 13,504B, BPFP=2.1100 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 9,356B, BPFP=1.4619 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 13,944B, BPFP=2.1787 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 9,676B, BPFP=1.5119 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 13,348B, BPFP=2.0856 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 7,272B, BPFP=1.1362 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 13,656B, BPFP=2.1338 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 34,704B, BPFP=1.3556 -⌛️ [2/4] FRONTEND: Frontend time: 0.220s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.211s - -[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, 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 17.01605469 - layer.0.v_cache 0.00000029 0.00026465 - layer.1.k_cache 0.00403332 1.22934326 - layer.1.v_cache 0.00000087 0.00093150 - layer.2.k_cache 0.00117108 0.53017532 - layer.2.v_cache 0.00000108 0.00126579 - layer.3.k_cache 0.00141424 0.61221455 - layer.3.v_cache 0.00000205 0.00207411 - layer.4.k_cache 0.00319313 1.27476555 - layer.4.v_cache 0.00000283 0.00334711 - layer.4.output 0.00027592 0.13133595 - ------------------------------------------------------------------------------------- - TOTAL 0.00285987 1.51398431 - (elements=716,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 716800 -Total Bytes 138300 -BPFP 1.5435 bits/point -EBPFP 3.0871 equivalent bits/point -MSE 1.513984 ----------------------- -------------------------------------------------------- -Time: 0.435s Load: 0.004s, Pack+Encode: 0.220s, Decode+Unpack: 0.211s ----------------------- -------------------------------------------------------- -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 1.5140 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,008B, BPFP=0.4123 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 12,904B, BPFP=1.7686 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 8,836B, BPFP=1.2111 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 14,280B, BPFP=1.9572 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 9,968B, BPFP=1.3662 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 14,624B, BPFP=2.0044 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 10,564B, BPFP=1.4479 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 14,388B, BPFP=1.9720 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 8,544B, BPFP=1.1711 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 14,448B, BPFP=1.9803 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 39,444B, BPFP=1.3516 -⌛️ [2/4] FRONTEND: Frontend time: 0.157s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.201s - -[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, 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 16.23204174 - layer.0.v_cache 0.00000028 0.00025901 - layer.1.k_cache 0.00371865 1.21553026 - layer.1.v_cache 0.00000081 0.00088848 - layer.2.k_cache 0.00114644 0.52249815 - layer.2.v_cache 0.00000109 0.00123411 - layer.3.k_cache 0.00141512 0.58613292 - layer.3.v_cache 0.00000208 0.00202502 - layer.4.k_cache 0.00325023 1.16320493 - layer.4.v_cache 0.00000285 0.00330454 - layer.4.output 0.00023785 0.10431792 - ------------------------------------------------------------------------------------- - TOTAL 0.00282188 1.43888506 - (elements=817,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 817152 -Total Bytes 151008 -BPFP 1.4784 bits/point -EBPFP 2.9568 equivalent bits/point -MSE 1.438885 ----------------------- -------------------------------------------------------- -Time: 0.363s Load: 0.005s, Pack+Encode: 0.157s, Decode+Unpack: 0.201s ----------------------- -------------------------------------------------------- -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 1.4389 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 5,236B, BPFP=0.3620 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,952B, BPFP=1.7942 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,756B, BPFP=1.1585 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,596B, BPFP=1.9079 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,576B, BPFP=1.3534 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,960B, BPFP=1.9331 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,184B, BPFP=1.3955 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,396B, BPFP=1.8941 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,960B, BPFP=1.1726 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,176B, BPFP=1.9480 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 80,456B, BPFP=1.3906 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.83869975 - layer.0.v_cache 0.00000028 0.00025274 - layer.1.k_cache 0.00339028 0.88056284 - layer.1.v_cache 0.00000081 0.00083419 - layer.2.k_cache 0.00113693 0.46094999 - layer.2.v_cache 0.00000114 0.00120830 - layer.3.k_cache 0.00135151 0.52121373 - layer.3.v_cache 0.00000219 0.00201692 - layer.4.k_cache 0.00328814 0.93830149 - layer.4.v_cache 0.00000314 0.00344455 - layer.4.output 0.00017637 0.06726813 - ------------------------------------------------------------------------------------- - TOTAL 0.00286021 1.13689693 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 296248 -BPFP 1.4630 bits/point -EBPFP 2.9260 equivalent bits/point -MSE 1.136897 ----------------------- -------------------------------------------------------- -Time: 0.501s Load: 0.008s, Pack+Encode: 0.208s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.1369 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,656B, BPFP=0.3368 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 24,460B, BPFP=1.7694 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,156B, BPFP=1.1687 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,728B, BPFP=1.9334 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,744B, BPFP=1.3559 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,004B, BPFP=1.9534 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,604B, BPFP=1.4181 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,540B, BPFP=1.9198 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,364B, BPFP=1.1837 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,284B, BPFP=1.9737 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,908B, BPFP=1.3004 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.92476174 - layer.0.v_cache 0.00000028 0.00024946 - layer.1.k_cache 0.00351470 1.03130779 - layer.1.v_cache 0.00000073 0.00084076 - layer.2.k_cache 0.00116483 0.47962549 - layer.2.v_cache 0.00000104 0.00119453 - layer.3.k_cache 0.00137767 0.55724225 - layer.3.v_cache 0.00000202 0.00200079 - layer.4.k_cache 0.00329212 1.05805446 - layer.4.v_cache 0.00000308 0.00336952 - layer.4.output 0.00021651 0.08758612 - ------------------------------------------------------------------------------------- - TOTAL 0.00256588 1.17207080 - (elements=1,548,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1548288 -Total Bytes 279448 -BPFP 1.4439 bits/point -EBPFP 2.8878 equivalent bits/point -MSE 1.172071 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.008s, Pack+Encode: 0.203s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.1721 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 2,776B, BPFP=0.4614 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 10,972B, BPFP=1.8238 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 7,144B, BPFP=1.1875 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 11,968B, BPFP=1.9894 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 8,412B, BPFP=1.3983 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 12,412B, BPFP=2.0632 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 8,872B, BPFP=1.4747 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 12,260B, BPFP=2.0379 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 6,884B, BPFP=1.1443 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 12,324B, BPFP=2.0485 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 32,488B, BPFP=1.3501 -⌛️ [2/4] FRONTEND: Frontend time: 0.149s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.194s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 17.44106861 - layer.0.v_cache 0.00000031 0.00028353 - layer.1.k_cache 0.00397950 1.42124225 - layer.1.v_cache 0.00000076 0.00088694 - layer.2.k_cache 0.00125400 0.57218990 - layer.2.v_cache 0.00000109 0.00127142 - layer.3.k_cache 0.00146534 0.60984104 - layer.3.v_cache 0.00000206 0.00209490 - layer.4.k_cache 0.00325167 1.30942600 - layer.4.v_cache 0.00000289 0.00337286 - layer.4.output 0.00022000 0.12668963 - ------------------------------------------------------------------------------------- - TOTAL 0.00290784 1.56203114 - (elements=673,792) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 673792 -Total Bytes 126512 -BPFP 1.5021 bits/point -EBPFP 3.0042 equivalent bits/point -MSE 1.562031 ----------------------- -------------------------------------------------------- -Time: 0.348s Load: 0.004s, Pack+Encode: 0.149s, Decode+Unpack: 0.194s ----------------------- -------------------------------------------------------- -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 1.5620 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.009s - ------------------------------------------------------------- -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: 7,256B, BPFP=0.3681 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 34,052B, BPFP=1.7275 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,332B, BPFP=1.1329 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 36,972B, BPFP=1.8756 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,148B, BPFP=1.3265 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 38,164B, BPFP=1.9361 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,416B, BPFP=1.3908 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 37,908B, BPFP=1.9231 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,896B, BPFP=1.1615 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 38,724B, BPFP=1.9645 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 106,608B, BPFP=1.3521 -⌛️ [2/4] FRONTEND: Frontend time: 0.255s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.383s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.95910248 - layer.0.v_cache 0.00000027 0.00023857 - layer.1.k_cache 0.00328501 0.89339061 - layer.1.v_cache 0.00000091 0.00086146 - layer.2.k_cache 0.00115449 0.46730611 - layer.2.v_cache 0.00000127 0.00127121 - layer.3.k_cache 0.00140159 0.54219397 - layer.3.v_cache 0.00000237 0.00218384 - layer.4.k_cache 0.00336527 0.97177540 - layer.4.v_cache 0.00000311 0.00340303 - layer.4.output 0.00016660 0.07222412 - ------------------------------------------------------------------------------------- - TOTAL 0.00250283 1.15218737 - (elements=2,207,744) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2207744 -Total Bytes 398476 -BPFP 1.4439 bits/point -EBPFP 2.8878 equivalent bits/point -MSE 1.152187 ----------------------- -------------------------------------------------------- -Time: 0.648s Load: 0.009s, Pack+Encode: 0.255s, Decode+Unpack: 0.383s ----------------------- -------------------------------------------------------- -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 1.1522 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,812B, BPFP=0.3759 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 23,852B, BPFP=1.8634 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,088B, BPFP=1.1787 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 25,572B, BPFP=1.9978 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,668B, BPFP=1.3803 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,900B, BPFP=2.0234 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,392B, BPFP=1.4369 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 25,432B, BPFP=1.9869 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,548B, BPFP=1.2147 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 26,296B, BPFP=2.0544 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,752B, BPFP=1.4014 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.68817871 - layer.0.v_cache 0.00000027 0.00026014 - layer.1.k_cache 0.00330333 1.15150177 - layer.1.v_cache 0.00000088 0.00089654 - layer.2.k_cache 0.00114776 0.51566143 - layer.2.v_cache 0.00000114 0.00125261 - layer.3.k_cache 0.00139474 0.58514492 - layer.3.v_cache 0.00000229 0.00219600 - layer.4.k_cache 0.00328125 1.09343277 - layer.4.v_cache 0.00000316 0.00355347 - layer.4.output 0.00020624 0.10122705 - ------------------------------------------------------------------------------------- - TOTAL 0.00255682 1.24621332 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 270312 -BPFP 1.5084 bits/point -EBPFP 3.0169 equivalent bits/point -MSE 1.246213 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.2462 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,652B, BPFP=0.3826 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,380B, BPFP=1.8405 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,984B, BPFP=1.1500 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,968B, BPFP=1.9711 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,224B, BPFP=1.3342 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,380B, BPFP=2.0049 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,288B, BPFP=1.4217 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,004B, BPFP=1.9740 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,408B, BPFP=1.1849 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,932B, BPFP=2.0503 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,868B, BPFP=1.2925 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.29307155 - layer.0.v_cache 0.00000026 0.00024806 - layer.1.k_cache 0.00345234 1.09942916 - layer.1.v_cache 0.00000076 0.00088125 - layer.2.k_cache 0.00126989 0.50434835 - layer.2.v_cache 0.00000106 0.00121562 - layer.3.k_cache 0.00140666 0.55711156 - layer.3.v_cache 0.00000216 0.00210717 - layer.4.k_cache 0.00351815 1.08861100 - layer.4.v_cache 0.00000309 0.00362537 - layer.4.output 0.00017514 0.09964173 - ------------------------------------------------------------------------------------- - TOTAL 0.00254962 1.21065829 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 249088 -BPFP 1.4632 bits/point -EBPFP 2.9263 equivalent bits/point -MSE 1.210658 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.2107 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,472B, BPFP=0.3602 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,088B, BPFP=1.7790 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,028B, BPFP=1.1298 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,684B, BPFP=1.9075 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,360B, BPFP=1.3177 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,288B, BPFP=1.9562 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,080B, BPFP=1.3756 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,712B, BPFP=1.9098 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,280B, BPFP=1.1501 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,616B, BPFP=1.9826 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,228B, BPFP=1.2530 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.90134957 - layer.0.v_cache 0.00000027 0.00024839 - layer.1.k_cache 0.00344234 1.00640083 - layer.1.v_cache 0.00000085 0.00087386 - layer.2.k_cache 0.00116154 0.48957990 - layer.2.v_cache 0.00000108 0.00122351 - layer.3.k_cache 0.00150460 0.55903421 - layer.3.v_cache 0.00000231 0.00206965 - layer.4.k_cache 0.00334272 1.07653195 - layer.4.v_cache 0.00000310 0.00355328 - layer.4.output 0.00020164 0.08209801 - ------------------------------------------------------------------------------------- - TOTAL 0.00256925 1.09780408 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 246836 -BPFP 1.4200 bits/point -EBPFP 2.8401 equivalent bits/point -MSE 1.097804 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.0978 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,680B, BPFP=0.3731 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,752B, BPFP=1.8138 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,336B, BPFP=1.1429 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,712B, BPFP=1.9700 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,912B, BPFP=1.3482 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,388B, BPFP=2.0239 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,612B, BPFP=1.4040 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,740B, BPFP=1.9723 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,944B, BPFP=1.1913 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,440B, BPFP=2.0281 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 70,916B, BPFP=1.4133 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.21765884 - layer.0.v_cache 0.00000028 0.00024282 - layer.1.k_cache 0.00330389 1.03492729 - layer.1.v_cache 0.00000100 0.00087988 - layer.2.k_cache 0.00113731 0.47125614 - layer.2.v_cache 0.00000122 0.00130107 - layer.3.k_cache 0.00133429 0.53366614 - layer.3.v_cache 0.00000236 0.00214545 - layer.4.k_cache 0.00329518 1.04338486 - layer.4.v_cache 0.00000318 0.00356892 - layer.4.output 0.00020743 0.08642003 - ------------------------------------------------------------------------------------- - TOTAL 0.00260785 1.18962225 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 262432 -BPFP 1.4944 bits/point -EBPFP 2.9887 equivalent bits/point -MSE 1.189622 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.1896 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,764B, BPFP=0.3798 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 23,040B, BPFP=1.8367 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,400B, BPFP=1.1480 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,740B, BPFP=1.9723 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,840B, BPFP=1.3425 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,108B, BPFP=2.0016 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,424B, BPFP=1.3890 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,476B, BPFP=1.9512 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,624B, BPFP=1.1658 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,152B, BPFP=2.0051 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,036B, BPFP=1.3360 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.22034688 - layer.0.v_cache 0.00000028 0.00025766 - layer.1.k_cache 0.00343622 1.03123420 - layer.1.v_cache 0.00000088 0.00089208 - layer.2.k_cache 0.00122497 0.48950110 - layer.2.v_cache 0.00000114 0.00131204 - layer.3.k_cache 0.00135168 0.55596231 - layer.3.v_cache 0.00000224 0.00212045 - layer.4.k_cache 0.00342672 1.01415307 - layer.4.v_cache 0.00000300 0.00346961 - layer.4.output 0.00022158 0.08352728 - ------------------------------------------------------------------------------------- - TOTAL 0.00263494 1.18952561 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 257604 -BPFP 1.4669 bits/point -EBPFP 2.9337 equivalent bits/point -MSE 1.189526 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.1895 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,736B, BPFP=0.3700 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 23,580B, BPFP=1.8422 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,256B, BPFP=1.1919 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 25,648B, BPFP=2.0038 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,644B, BPFP=1.3784 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,980B, BPFP=2.0297 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,496B, BPFP=1.4450 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 25,504B, BPFP=1.9925 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,496B, BPFP=1.2106 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 26,156B, BPFP=2.0434 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 70,000B, BPFP=1.3672 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.98508423 - layer.0.v_cache 0.00000027 0.00024711 - layer.1.k_cache 0.00335783 1.08163765 - layer.1.v_cache 0.00000086 0.00087652 - layer.2.k_cache 0.00109031 0.49215988 - layer.2.v_cache 0.00000115 0.00125140 - layer.3.k_cache 0.00136642 0.55922653 - layer.3.v_cache 0.00000222 0.00210975 - layer.4.k_cache 0.00341075 1.04534927 - layer.4.v_cache 0.00000317 0.00345904 - layer.4.output 0.00031254 0.09304502 - ------------------------------------------------------------------------------------- - TOTAL 0.00265226 1.25311296 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 268496 -BPFP 1.4983 bits/point -EBPFP 2.9966 equivalent bits/point -MSE 1.253113 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.2531 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,560B, BPFP=0.3673 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,152B, BPFP=1.7841 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,104B, BPFP=1.1360 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,120B, BPFP=1.9427 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,668B, BPFP=1.3425 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,388B, BPFP=1.9642 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,008B, BPFP=1.3698 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,820B, BPFP=1.9185 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,520B, BPFP=1.1695 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,828B, BPFP=1.9997 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,600B, BPFP=1.3007 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.61766068 - layer.0.v_cache 0.00000027 0.00026487 - layer.1.k_cache 0.00347710 1.00105592 - layer.1.v_cache 0.00000089 0.00098182 - layer.2.k_cache 0.00114081 0.49794702 - layer.2.v_cache 0.00000111 0.00132640 - layer.3.k_cache 0.00140715 0.58224173 - layer.3.v_cache 0.00000211 0.00217248 - layer.4.k_cache 0.00324570 1.05407149 - layer.4.v_cache 0.00000329 0.00374030 - layer.4.output 0.00023085 0.07836600 - ------------------------------------------------------------------------------------- - TOTAL 0.00262500 1.14820905 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 250768 -BPFP 1.4427 bits/point -EBPFP 2.8853 equivalent bits/point -MSE 1.148209 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.008s, Pack+Encode: 0.202s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.1482 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,488B, BPFP=0.3940 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,912B, BPFP=1.8357 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,460B, BPFP=1.1815 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,976B, BPFP=2.0169 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,604B, BPFP=1.3697 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,600B, BPFP=2.0716 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,552B, BPFP=1.4529 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,092B, BPFP=2.0270 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,584B, BPFP=1.1924 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,864B, BPFP=2.0948 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,192B, BPFP=1.3209 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 16.39539946 - layer.0.v_cache 0.00000027 0.00024735 - layer.1.k_cache 0.00348241 1.08738966 - layer.1.v_cache 0.00000086 0.00089505 - layer.2.k_cache 0.00112258 0.50899634 - layer.2.v_cache 0.00000113 0.00124470 - layer.3.k_cache 0.00136851 0.57346207 - layer.3.v_cache 0.00000223 0.00207280 - layer.4.k_cache 0.00328884 1.11607704 - layer.4.v_cache 0.00000300 0.00335321 - layer.4.output 0.00019035 0.09451528 - ------------------------------------------------------------------------------------- - TOTAL 0.00264321 1.43337134 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 238324 -BPFP 1.4943 bits/point -EBPFP 2.9886 equivalent bits/point -MSE 1.433371 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.4334 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,944B, BPFP=0.3714 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 24,932B, BPFP=1.8729 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,000B, BPFP=1.2019 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,780B, BPFP=2.0117 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,616B, BPFP=1.3984 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,064B, BPFP=2.0331 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,356B, BPFP=1.4540 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,400B, BPFP=1.9832 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,184B, BPFP=1.2157 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,204B, BPFP=2.0436 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 74,436B, BPFP=1.3979 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.30089393 - layer.0.v_cache 0.00000027 0.00026013 - layer.1.k_cache 0.00323274 1.01783122 - layer.1.v_cache 0.00000093 0.00094280 - layer.2.k_cache 0.00116098 0.48070658 - layer.2.v_cache 0.00000116 0.00131822 - layer.3.k_cache 0.00137668 0.54488299 - layer.3.v_cache 0.00000213 0.00213062 - layer.4.k_cache 0.00339989 1.06257593 - layer.4.v_cache 0.00000305 0.00352026 - layer.4.output 0.00020049 0.08523784 - ------------------------------------------------------------------------------------- - TOTAL 0.00250768 1.26828672 - (elements=1,490,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1490944 -Total Bytes 281916 -BPFP 1.5127 bits/point -EBPFP 3.0254 equivalent bits/point -MSE 1.268287 ----------------------- -------------------------------------------------------- -Time: 0.502s Load: 0.008s, Pack+Encode: 0.209s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.2683 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,072B, BPFP=0.3880 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,656B, BPFP=1.8727 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,672B, BPFP=1.2073 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,464B, BPFP=2.0450 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,576B, BPFP=1.3887 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,760B, BPFP=2.0732 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,660B, BPFP=1.4920 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,424B, BPFP=2.0412 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,604B, BPFP=1.2008 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,344B, BPFP=2.1288 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 56,408B, BPFP=1.3436 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 16.04294493 - layer.0.v_cache 0.00000027 0.00024496 - layer.1.k_cache 0.00355746 1.15157821 - layer.1.v_cache 0.00000072 0.00081005 - layer.2.k_cache 0.00112546 0.50456461 - layer.2.v_cache 0.00000104 0.00114245 - layer.3.k_cache 0.00143209 0.58159907 - layer.3.v_cache 0.00000205 0.00193762 - layer.4.k_cache 0.00323843 1.16904282 - layer.4.v_cache 0.00000284 0.00318410 - layer.4.output 0.00022045 0.10536664 - ------------------------------------------------------------------------------------- - TOTAL 0.00273527 1.41989396 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 222640 -BPFP 1.5151 bits/point -EBPFP 3.0303 equivalent bits/point -MSE 1.419894 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.4199 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,948B, BPFP=0.3546 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,080B, BPFP=1.7976 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,440B, BPFP=1.1783 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,092B, BPFP=1.9418 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,224B, BPFP=1.3779 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,440B, BPFP=1.9667 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,844B, BPFP=1.4223 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,796B, BPFP=1.9206 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,728B, BPFP=1.1990 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,604B, BPFP=1.9785 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 75,672B, BPFP=1.3559 -⌛️ [2/4] FRONTEND: Frontend time: 0.207s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.288s - -[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, 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 12.96803662 - layer.0.v_cache 0.00000028 0.00024949 - layer.1.k_cache 0.00329736 1.00357504 - layer.1.v_cache 0.00000085 0.00086740 - layer.2.k_cache 0.00114487 0.48470037 - layer.2.v_cache 0.00000123 0.00127182 - layer.3.k_cache 0.00134342 0.53531486 - layer.3.v_cache 0.00000206 0.00203020 - layer.4.k_cache 0.00348010 1.06140291 - layer.4.v_cache 0.00000308 0.00344861 - layer.4.output 0.00017639 0.08515500 - ------------------------------------------------------------------------------------- - TOTAL 0.00245771 1.17153695 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 286868 -BPFP 1.4686 bits/point -EBPFP 2.9373 equivalent bits/point -MSE 1.171537 ----------------------- -------------------------------------------------------- -Time: 0.502s Load: 0.007s, Pack+Encode: 0.207s, Decode+Unpack: 0.288s ----------------------- -------------------------------------------------------- -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 1.1715 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,420B, BPFP=0.3635 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,000B, BPFP=1.8092 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,224B, BPFP=1.1697 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,876B, BPFP=1.9635 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,408B, BPFP=1.3493 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,608B, BPFP=2.0237 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,060B, BPFP=1.4030 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,960B, BPFP=1.9704 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,388B, BPFP=1.1832 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,792B, BPFP=2.0388 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,260B, BPFP=1.3828 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.286s - -[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, 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 13.52332571 - layer.0.v_cache 0.00000028 0.00026291 - layer.1.k_cache 0.00367247 1.03780791 - layer.1.v_cache 0.00000082 0.00092938 - layer.2.k_cache 0.00112740 0.48923107 - layer.2.v_cache 0.00000120 0.00136622 - layer.3.k_cache 0.00137809 0.55545160 - layer.3.v_cache 0.00000227 0.00219495 - layer.4.k_cache 0.00319271 1.06842844 - layer.4.v_cache 0.00000332 0.00365517 - layer.4.output 0.00022674 0.09083351 - ------------------------------------------------------------------------------------- - TOTAL 0.00268704 1.21757053 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 252996 -BPFP 1.4861 bits/point -EBPFP 2.9722 equivalent bits/point -MSE 1.217571 ----------------------- -------------------------------------------------------- -Time: 0.497s Load: 0.007s, Pack+Encode: 0.205s, Decode+Unpack: 0.286s ----------------------- -------------------------------------------------------- -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 1.2176 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,888B, BPFP=0.3708 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 24,288B, BPFP=1.8422 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,608B, BPFP=1.1839 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,560B, BPFP=2.0146 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,180B, BPFP=1.3789 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 26,816B, BPFP=2.0340 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,832B, BPFP=1.4284 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,028B, BPFP=1.9742 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,716B, BPFP=1.1921 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,080B, BPFP=2.0540 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 70,976B, BPFP=1.3459 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.286s - -[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, 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 13.62851989 - layer.0.v_cache 0.00000028 0.00025200 - layer.1.k_cache 0.00329593 1.05759897 - layer.1.v_cache 0.00000087 0.00089426 - layer.2.k_cache 0.00114743 0.47674557 - layer.2.v_cache 0.00000126 0.00123299 - layer.3.k_cache 0.00133553 0.53101279 - layer.3.v_cache 0.00000217 0.00209278 - layer.4.k_cache 0.00338515 0.99372649 - layer.4.v_cache 0.00000318 0.00349121 - layer.4.output 0.00017408 0.08163691 - ------------------------------------------------------------------------------------- - TOTAL 0.00284966 1.21586533 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 274972 -BPFP 1.4897 bits/point -EBPFP 2.9795 equivalent bits/point -MSE 1.215865 ----------------------- -------------------------------------------------------- -Time: 0.497s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.286s ----------------------- -------------------------------------------------------- -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 1.2159 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,968B, BPFP=0.3768 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 24,696B, BPFP=1.8732 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,728B, BPFP=1.1930 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,432B, BPFP=2.0049 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,588B, BPFP=1.4099 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 26,876B, BPFP=2.0385 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,200B, BPFP=1.4563 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,268B, BPFP=1.9924 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,088B, BPFP=1.2203 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,256B, BPFP=2.0674 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,300B, BPFP=1.3899 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.304s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.90395674 - layer.0.v_cache 0.00000028 0.00025172 - layer.1.k_cache 0.00340973 1.03384533 - layer.1.v_cache 0.00000088 0.00088173 - layer.2.k_cache 0.00113950 0.48229962 - layer.2.v_cache 0.00000120 0.00128052 - layer.3.k_cache 0.00133761 0.55323540 - layer.3.v_cache 0.00000227 0.00211949 - layer.4.k_cache 0.00339299 1.05040645 - layer.4.v_cache 0.00000316 0.00351718 - layer.4.output 0.00023648 0.08299268 - ------------------------------------------------------------------------------------- - TOTAL 0.00261420 1.31169749 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 279400 -BPFP 1.5137 bits/point -EBPFP 3.0275 equivalent bits/point -MSE 1.311697 ----------------------- -------------------------------------------------------- -Time: 0.520s Load: 0.007s, Pack+Encode: 0.209s, Decode+Unpack: 0.304s ----------------------- -------------------------------------------------------- -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 1.3117 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,380B, BPFP=0.3719 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,936B, BPFP=1.8628 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,100B, BPFP=1.1974 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,992B, BPFP=2.0374 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,468B, BPFP=1.3984 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,280B, BPFP=2.0618 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,368B, BPFP=1.4749 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,004B, BPFP=2.0384 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,232B, BPFP=1.2086 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,440B, BPFP=2.0754 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 66,084B, BPFP=1.4029 -⌛️ [2/4] FRONTEND: Frontend time: 0.210s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.296s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.34777832 - layer.0.v_cache 0.00000028 0.00024971 - layer.1.k_cache 0.00347829 1.08664554 - layer.1.v_cache 0.00000086 0.00088666 - layer.2.k_cache 0.00113727 0.48434602 - layer.2.v_cache 0.00000116 0.00122707 - layer.3.k_cache 0.00132770 0.54903387 - layer.3.v_cache 0.00000243 0.00205605 - layer.4.k_cache 0.00339865 1.05269482 - layer.4.v_cache 0.00000297 0.00327175 - layer.4.output 0.00018871 0.08675475 - ------------------------------------------------------------------------------------- - TOTAL 0.00247791 1.34822920 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 251284 -BPFP 1.5242 bits/point -EBPFP 3.0484 equivalent bits/point -MSE 1.348229 ----------------------- -------------------------------------------------------- -Time: 0.514s Load: 0.008s, Pack+Encode: 0.210s, Decode+Unpack: 0.296s ----------------------- -------------------------------------------------------- -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 1.3482 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,460B, BPFP=0.3630 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,124B, BPFP=1.8005 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,180B, BPFP=1.1540 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,748B, BPFP=1.9326 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,240B, BPFP=1.3216 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,208B, BPFP=1.9701 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,272B, BPFP=1.4056 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,104B, BPFP=1.9616 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,416B, BPFP=1.1732 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,720B, BPFP=2.0117 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 68,180B, BPFP=1.3871 -⌛️ [2/4] FRONTEND: Frontend time: 0.210s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.293s - -[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, 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 12.91423798 - layer.0.v_cache 0.00000027 0.00024849 - layer.1.k_cache 0.00351230 0.97879823 - layer.1.v_cache 0.00000075 0.00083548 - layer.2.k_cache 0.00114531 0.46662446 - layer.2.v_cache 0.00000110 0.00119642 - layer.3.k_cache 0.00136477 0.52964036 - layer.3.v_cache 0.00000217 0.00208503 - layer.4.k_cache 0.00356451 1.03530963 - layer.4.v_cache 0.00000291 0.00330092 - layer.4.output 0.00018809 0.09310284 - ------------------------------------------------------------------------------------- - TOTAL 0.00262048 1.16462060 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 253652 -BPFP 1.4744 bits/point -EBPFP 2.9489 equivalent bits/point -MSE 1.164621 ----------------------- -------------------------------------------------------- -Time: 0.509s Load: 0.007s, Pack+Encode: 0.210s, Decode+Unpack: 0.293s ----------------------- -------------------------------------------------------- -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 1.1646 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,528B, BPFP=0.3804 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,340B, BPFP=1.8767 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,300B, BPFP=1.2013 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,888B, BPFP=2.0067 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,604B, BPFP=1.3948 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,380B, BPFP=2.0481 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,268B, BPFP=1.4506 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,812B, BPFP=2.0003 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,516B, BPFP=1.2194 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,644B, BPFP=2.0702 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 66,424B, BPFP=1.3950 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.295s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.65328881 - layer.0.v_cache 0.00000028 0.00025135 - layer.1.k_cache 0.00358459 1.14176416 - layer.1.v_cache 0.00000078 0.00087789 - layer.2.k_cache 0.00113483 0.48362543 - layer.2.v_cache 0.00000114 0.00126554 - layer.3.k_cache 0.00136506 0.56404097 - layer.3.v_cache 0.00000214 0.00206013 - layer.4.k_cache 0.00334622 1.10133821 - layer.4.v_cache 0.00000303 0.00345494 - layer.4.output 0.00022383 0.10340593 - ------------------------------------------------------------------------------------- - TOTAL 0.00260854 1.31182794 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 252704 -BPFP 1.5163 bits/point -EBPFP 3.0326 equivalent bits/point -MSE 1.311828 ----------------------- -------------------------------------------------------- -Time: 0.511s Load: 0.007s, Pack+Encode: 0.208s, Decode+Unpack: 0.295s ----------------------- -------------------------------------------------------- -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 1.3118 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,564B, BPFP=0.3714 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,440B, BPFP=1.8262 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,360B, BPFP=1.1686 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,404B, BPFP=1.9860 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,816B, BPFP=1.3685 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,728B, BPFP=2.0124 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,188B, BPFP=1.3988 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,088B, BPFP=1.9603 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,752B, BPFP=1.2005 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,960B, BPFP=2.0312 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,872B, BPFP=1.3809 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.294s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.33801015 - layer.0.v_cache 0.00000028 0.00024570 - layer.1.k_cache 0.00334712 0.94870559 - layer.1.v_cache 0.00000079 0.00091081 - layer.2.k_cache 0.00118939 0.47461200 - layer.2.v_cache 0.00000116 0.00133238 - layer.3.k_cache 0.00138481 0.53672012 - layer.3.v_cache 0.00000224 0.00213282 - layer.4.k_cache 0.00336247 0.99583658 - layer.4.v_cache 0.00000319 0.00355050 - layer.4.output 0.00019453 0.07905197 - ------------------------------------------------------------------------------------- - TOTAL 0.00260570 1.11559032 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 256172 -BPFP 1.4891 bits/point -EBPFP 2.9782 equivalent bits/point -MSE 1.115590 ----------------------- -------------------------------------------------------- -Time: 0.508s Load: 0.006s, Pack+Encode: 0.209s, Decode+Unpack: 0.294s ----------------------- -------------------------------------------------------- -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 1.1156 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,412B, BPFP=0.3747 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,608B, BPFP=1.8349 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,940B, BPFP=1.1838 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,672B, BPFP=2.0102 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,460B, BPFP=1.3978 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,136B, BPFP=2.0496 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,072B, BPFP=1.4497 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,652B, BPFP=2.0085 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,128B, BPFP=1.1997 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,416B, BPFP=2.0734 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 63,448B, BPFP=1.3470 -⌛️ [2/4] FRONTEND: Frontend time: 0.210s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.296s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.51838087 - layer.0.v_cache 0.00000028 0.00024718 - layer.1.k_cache 0.00330708 1.09101627 - layer.1.v_cache 0.00000075 0.00089829 - layer.2.k_cache 0.00113337 0.49412325 - layer.2.v_cache 0.00000121 0.00134155 - layer.3.k_cache 0.00137180 0.56000494 - layer.3.v_cache 0.00000212 0.00209191 - layer.4.k_cache 0.00335567 1.06886325 - layer.4.v_cache 0.00000315 0.00358078 - layer.4.output 0.00020664 0.08681630 - ------------------------------------------------------------------------------------- - TOTAL 0.00253069 1.36341525 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 246944 -BPFP 1.4979 bits/point -EBPFP 2.9957 equivalent bits/point -MSE 1.363415 ----------------------- -------------------------------------------------------- -Time: 0.513s Load: 0.007s, Pack+Encode: 0.210s, Decode+Unpack: 0.296s ----------------------- -------------------------------------------------------- -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 1.3634 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.011s - ------------------------------------------------------------- -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: 8,376B, BPFP=0.3676 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 38,424B, BPFP=1.6864 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 25,128B, BPFP=1.1029 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 41,088B, BPFP=1.8034 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 29,124B, BPFP=1.2783 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 41,692B, BPFP=1.8299 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 30,308B, BPFP=1.3302 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 41,148B, BPFP=1.8060 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 25,196B, BPFP=1.1059 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 42,048B, BPFP=1.8455 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 112,780B, BPFP=1.2375 -⌛️ [2/4] FRONTEND: Frontend time: 0.263s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.399s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.37050123 - layer.0.v_cache 0.00000026 0.00023786 - layer.1.k_cache 0.00309449 0.81083559 - layer.1.v_cache 0.00000082 0.00083417 - layer.2.k_cache 0.00118451 0.44889270 - layer.2.v_cache 0.00000118 0.00119714 - layer.3.k_cache 0.00136933 0.51766496 - layer.3.v_cache 0.00000226 0.00207204 - layer.4.k_cache 0.00340187 0.95339889 - layer.4.v_cache 0.00000300 0.00323518 - layer.4.output 0.00017401 0.06449198 - ------------------------------------------------------------------------------------- - TOTAL 0.00241758 1.02620269 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 435312 -BPFP 1.3647 bits/point -EBPFP 2.7294 equivalent bits/point -MSE 1.026203 ----------------------- -------------------------------------------------------- -Time: 0.673s Load: 0.011s, Pack+Encode: 0.263s, Decode+Unpack: 0.399s ----------------------- -------------------------------------------------------- -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 1.0262 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,376B, BPFP=0.3599 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,968B, BPFP=1.8066 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,324B, BPFP=1.1780 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,524B, BPFP=1.9345 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,556B, BPFP=1.3615 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,132B, BPFP=1.9845 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,220B, BPFP=1.4161 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,488B, BPFP=1.9316 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,476B, BPFP=1.1905 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,416B, BPFP=2.0079 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,388B, BPFP=1.3443 -⌛️ [2/4] FRONTEND: Frontend time: 0.212s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.295s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.40626413 - layer.0.v_cache 0.00000028 0.00024499 - layer.1.k_cache 0.00332260 1.04390066 - layer.1.v_cache 0.00000077 0.00087257 - layer.2.k_cache 0.00114542 0.50299683 - layer.2.v_cache 0.00000120 0.00128773 - layer.3.k_cache 0.00143781 0.56807408 - layer.3.v_cache 0.00000216 0.00201885 - layer.4.k_cache 0.00324414 1.08517255 - layer.4.v_cache 0.00000298 0.00335782 - layer.4.output 0.00027467 0.09612493 - ------------------------------------------------------------------------------------- - TOTAL 0.00256159 1.21419214 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 249868 -BPFP 1.4677 bits/point -EBPFP 2.9355 equivalent bits/point -MSE 1.214192 ----------------------- -------------------------------------------------------- -Time: 0.514s Load: 0.007s, Pack+Encode: 0.212s, Decode+Unpack: 0.295s ----------------------- -------------------------------------------------------- -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 1.2142 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,580B, BPFP=0.3807 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,676B, BPFP=1.8015 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,212B, BPFP=1.1812 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,556B, BPFP=1.9578 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,508B, BPFP=1.3720 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,240B, BPFP=2.0146 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,256B, BPFP=1.4342 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,768B, BPFP=1.9754 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,580B, BPFP=1.2118 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,500B, BPFP=2.0362 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,948B, BPFP=1.3495 -⌛️ [2/4] FRONTEND: Frontend time: 0.210s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.298s - -[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, 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 13.57097350 - layer.0.v_cache 0.00000029 0.00024220 - layer.1.k_cache 0.00347189 1.08238423 - layer.1.v_cache 0.00000073 0.00080836 - layer.2.k_cache 0.00115370 0.48461111 - layer.2.v_cache 0.00000109 0.00121345 - layer.3.k_cache 0.00136415 0.53779334 - layer.3.v_cache 0.00000218 0.00197341 - layer.4.k_cache 0.00350239 1.04119914 - layer.4.v_cache 0.00000301 0.00340726 - layer.4.output 0.00017689 0.09718250 - ------------------------------------------------------------------------------------- - TOTAL 0.00254021 1.22238114 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 249824 -BPFP 1.4831 bits/point -EBPFP 2.9662 equivalent bits/point -MSE 1.222381 ----------------------- -------------------------------------------------------- -Time: 0.516s Load: 0.008s, Pack+Encode: 0.210s, Decode+Unpack: 0.298s ----------------------- -------------------------------------------------------- -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 1.2224 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,664B, BPFP=0.3681 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 23,204B, BPFP=1.8311 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,780B, BPFP=1.1664 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 25,268B, BPFP=1.9940 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,284B, BPFP=1.3640 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,664B, BPFP=2.0253 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,132B, BPFP=1.4309 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,920B, BPFP=1.9665 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,076B, BPFP=1.1897 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,756B, BPFP=2.0325 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,620B, BPFP=1.3340 -⌛️ [2/4] FRONTEND: Frontend time: 0.232s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 13.62776693 - layer.0.v_cache 0.00000028 0.00024644 - layer.1.k_cache 0.00326434 1.02712782 - layer.1.v_cache 0.00000081 0.00083672 - layer.2.k_cache 0.00115982 0.49383217 - layer.2.v_cache 0.00000131 0.00125560 - layer.3.k_cache 0.00139121 0.55692519 - layer.3.v_cache 0.00000205 0.00199595 - layer.4.k_cache 0.00335893 1.07618713 - layer.4.v_cache 0.00000299 0.00344639 - layer.4.output 0.00020438 0.09285371 - ------------------------------------------------------------------------------------- - TOTAL 0.00245722 1.22578823 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 262368 -BPFP 1.4789 bits/point -EBPFP 2.9578 equivalent bits/point -MSE 1.225788 ----------------------- -------------------------------------------------------- -Time: 0.525s Load: 0.006s, Pack+Encode: 0.232s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.2258 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,812B, BPFP=0.3650 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 24,068B, BPFP=1.8255 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 15,732B, BPFP=1.1933 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,088B, BPFP=1.9788 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 18,112B, BPFP=1.3738 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 26,348B, BPFP=1.9985 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,064B, BPFP=1.4460 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 25,940B, BPFP=1.9675 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,048B, BPFP=1.2172 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 26,636B, BPFP=2.0203 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 71,864B, BPFP=1.3627 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 14.75521465 - layer.0.v_cache 0.00000026 0.00023992 - layer.1.k_cache 0.00325535 1.05610909 - layer.1.v_cache 0.00000093 0.00088429 - layer.2.k_cache 0.00112553 0.48792867 - layer.2.v_cache 0.00000114 0.00122621 - layer.3.k_cache 0.00141218 0.55958150 - layer.3.v_cache 0.00000204 0.00202535 - layer.4.k_cache 0.00324713 1.09697020 - layer.4.v_cache 0.00000302 0.00339936 - layer.4.output 0.00020093 0.09749151 - ------------------------------------------------------------------------------------- - TOTAL 0.00256365 1.31096752 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 274712 -BPFP 1.4883 bits/point -EBPFP 2.9767 equivalent bits/point -MSE 1.310968 ----------------------- -------------------------------------------------------- -Time: 0.498s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.3110 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,744B, BPFP=0.3861 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,512B, BPFP=1.8320 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,432B, BPFP=1.1745 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,284B, BPFP=1.9762 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,820B, BPFP=1.3688 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,540B, BPFP=1.9971 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,420B, BPFP=1.4176 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,212B, BPFP=1.9704 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,832B, BPFP=1.2070 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,028B, BPFP=2.0368 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,220B, BPFP=1.3676 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.81838226 - layer.0.v_cache 0.00000028 0.00024084 - layer.1.k_cache 0.00321835 0.96878091 - layer.1.v_cache 0.00000081 0.00092075 - layer.2.k_cache 0.00115341 0.47995766 - layer.2.v_cache 0.00000115 0.00124502 - layer.3.k_cache 0.00138578 0.54283762 - layer.3.v_cache 0.00000239 0.00218852 - layer.4.k_cache 0.00338578 1.01710693 - layer.4.v_cache 0.00000322 0.00357577 - layer.4.output 0.00018605 0.09651041 - ------------------------------------------------------------------------------------- - TOTAL 0.00255769 1.08723414 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 256044 -BPFP 1.4884 bits/point -EBPFP 2.9767 equivalent bits/point -MSE 1.087234 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.0872 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,504B, BPFP=0.4140 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,788B, BPFP=1.9107 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,276B, BPFP=1.2202 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,848B, BPFP=2.1000 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,448B, BPFP=1.4199 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,348B, BPFP=2.1460 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,136B, BPFP=1.4831 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,744B, BPFP=2.0904 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,336B, BPFP=1.2257 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,564B, BPFP=2.1658 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 63,648B, BPFP=1.4625 -⌛️ [2/4] FRONTEND: Frontend time: 0.216s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.302s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.23650333 - layer.0.v_cache 0.00000027 0.00025973 - layer.1.k_cache 0.00328410 1.12596256 - layer.1.v_cache 0.00000085 0.00093709 - layer.2.k_cache 0.00114159 0.52142195 - layer.2.v_cache 0.00000125 0.00141390 - layer.3.k_cache 0.00133976 0.58871819 - layer.3.v_cache 0.00000266 0.00234009 - layer.4.k_cache 0.00332377 1.15684500 - layer.4.v_cache 0.00000336 0.00386589 - layer.4.output 0.00023926 0.10130849 - ------------------------------------------------------------------------------------- - TOTAL 0.00266509 1.36025012 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 239640 -BPFP 1.5733 bits/point -EBPFP 3.1465 equivalent bits/point -MSE 1.360250 ----------------------- -------------------------------------------------------- -Time: 0.526s Load: 0.007s, Pack+Encode: 0.216s, Decode+Unpack: 0.302s ----------------------- -------------------------------------------------------- -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 1.3603 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,452B, BPFP=0.3586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,052B, BPFP=1.7761 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,220B, BPFP=1.1453 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,040B, BPFP=1.9362 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,508B, BPFP=1.3296 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,520B, BPFP=1.9749 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,096B, BPFP=1.3769 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,764B, BPFP=1.9140 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,304B, BPFP=1.1521 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,852B, BPFP=2.0016 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,068B, BPFP=1.3102 -⌛️ [2/4] FRONTEND: Frontend time: 0.211s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.87218987 - layer.0.v_cache 0.00000029 0.00026439 - layer.1.k_cache 0.00322990 0.98046497 - layer.1.v_cache 0.00000089 0.00092440 - layer.2.k_cache 0.00117012 0.47426684 - layer.2.v_cache 0.00000108 0.00127872 - layer.3.k_cache 0.00135684 0.53708141 - layer.3.v_cache 0.00000209 0.00205206 - layer.4.k_cache 0.00331928 1.00876224 - layer.4.v_cache 0.00000299 0.00345917 - layer.4.output 0.00021425 0.07935840 - ------------------------------------------------------------------------------------- - TOTAL 0.00253374 1.08558412 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 250876 -BPFP 1.4433 bits/point -EBPFP 2.8866 equivalent bits/point -MSE 1.085584 ----------------------- -------------------------------------------------------- -Time: 0.501s Load: 0.006s, Pack+Encode: 0.211s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.0856 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,532B, BPFP=0.3650 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 22,604B, BPFP=1.8206 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,396B, BPFP=1.1595 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,408B, BPFP=1.9659 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,704B, BPFP=1.3454 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,932B, BPFP=2.0081 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,480B, BPFP=1.4079 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,460B, BPFP=1.9700 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,676B, BPFP=1.1820 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,204B, BPFP=2.0300 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 70,584B, BPFP=1.4212 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 12.63056364 - layer.0.v_cache 0.00000028 0.00026913 - layer.1.k_cache 0.00330002 0.96665782 - layer.1.v_cache 0.00000090 0.00088729 - layer.2.k_cache 0.00113900 0.46444309 - layer.2.v_cache 0.00000130 0.00127179 - layer.3.k_cache 0.00136322 0.52253574 - layer.3.v_cache 0.00000251 0.00222470 - layer.4.k_cache 0.00342626 0.96921822 - layer.4.v_cache 0.00000319 0.00344746 - layer.4.output 0.00021399 0.08323184 - ------------------------------------------------------------------------------------- - TOTAL 0.00244156 1.13531759 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 259980 -BPFP 1.4957 bits/point -EBPFP 2.9913 equivalent bits/point -MSE 1.135318 ----------------------- -------------------------------------------------------- -Time: 0.497s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.1353 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 4,588B, BPFP=0.3621 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 23,188B, BPFP=1.8299 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,872B, BPFP=1.1736 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 24,880B, BPFP=1.9634 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 17,196B, BPFP=1.3570 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 25,304B, BPFP=1.9968 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 18,016B, BPFP=1.4217 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 24,628B, BPFP=1.9435 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 15,012B, BPFP=1.1847 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 25,500B, BPFP=2.0123 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 67,796B, BPFP=1.3375 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.35899646 - layer.0.v_cache 0.00000029 0.00025245 - layer.1.k_cache 0.00335574 1.06414040 - layer.1.v_cache 0.00000081 0.00082114 - layer.2.k_cache 0.00112037 0.46179045 - layer.2.v_cache 0.00000110 0.00115944 - layer.3.k_cache 0.00138480 0.55920973 - layer.3.v_cache 0.00000194 0.00187060 - layer.4.k_cache 0.00346640 1.04827049 - layer.4.v_cache 0.00000296 0.00319342 - layer.4.output 0.00019387 0.08125792 - ------------------------------------------------------------------------------------- - TOTAL 0.00250096 1.20176687 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 260980 -BPFP 1.4711 bits/point -EBPFP 2.9421 equivalent bits/point -MSE 1.201767 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.008s, Pack+Encode: 0.203s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.2018 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,656B, BPFP=0.4230 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,068B, BPFP=1.9139 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,528B, BPFP=1.2289 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,052B, BPFP=2.0941 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,752B, BPFP=1.4310 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,756B, BPFP=2.1581 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,560B, BPFP=1.5044 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,184B, BPFP=2.1061 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,512B, BPFP=1.2275 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,664B, BPFP=2.1497 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,816B, BPFP=1.4266 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.04213271 - layer.0.v_cache 0.00000027 0.00024565 - layer.1.k_cache 0.00348465 1.08692169 - layer.1.v_cache 0.00000080 0.00090175 - layer.2.k_cache 0.00118276 0.50395797 - layer.2.v_cache 0.00000133 0.00145528 - layer.3.k_cache 0.00137135 0.57945504 - layer.3.v_cache 0.00000258 0.00228602 - layer.4.k_cache 0.00338397 1.09483195 - layer.4.v_cache 0.00000307 0.00362297 - layer.4.output 0.00019889 0.09920468 - ------------------------------------------------------------------------------------- - TOTAL 0.00259439 1.26518784 - (elements=1,232,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1232896 -Total Bytes 241548 -BPFP 1.5674 bits/point -EBPFP 3.1347 equivalent bits/point -MSE 1.265188 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.2652 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,788B, BPFP=0.4203 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,964B, BPFP=1.8402 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,596B, BPFP=1.1935 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,956B, BPFP=2.0151 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,848B, BPFP=1.3912 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,724B, BPFP=2.0825 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,060B, BPFP=1.4975 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,260B, BPFP=2.0418 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,864B, BPFP=1.2170 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,016B, BPFP=2.1081 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 65,680B, BPFP=1.4414 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.42075168 - layer.0.v_cache 0.00000027 0.00025483 - layer.1.k_cache 0.00344681 1.07228757 - layer.1.v_cache 0.00000079 0.00092651 - layer.2.k_cache 0.00114729 0.51403029 - layer.2.v_cache 0.00000121 0.00133197 - layer.3.k_cache 0.00138721 0.60999654 - layer.3.v_cache 0.00000216 0.00223326 - layer.4.k_cache 0.00341441 1.11139293 - layer.4.v_cache 0.00000316 0.00377021 - layer.4.output 0.00022172 0.10403665 - ------------------------------------------------------------------------------------- - TOTAL 0.00265058 1.29665160 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 245756 -BPFP 1.5409 bits/point -EBPFP 3.0818 equivalent bits/point -MSE 1.296652 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.2967 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,556B, BPFP=0.3999 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,476B, BPFP=1.8852 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,884B, BPFP=1.2188 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,660B, BPFP=2.0769 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,176B, BPFP=1.4199 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,172B, BPFP=2.1218 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,740B, BPFP=1.4695 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,552B, BPFP=2.0674 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,988B, BPFP=1.2279 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,176B, BPFP=2.1222 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,180B, BPFP=1.4084 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.75021671 - layer.0.v_cache 0.00000029 0.00025600 - layer.1.k_cache 0.00343373 1.06226417 - layer.1.v_cache 0.00000078 0.00090669 - layer.2.k_cache 0.00114421 0.48666785 - layer.2.v_cache 0.00000115 0.00128482 - layer.3.k_cache 0.00135741 0.55252015 - layer.3.v_cache 0.00000236 0.00211827 - layer.4.k_cache 0.00332665 1.03403387 - layer.4.v_cache 0.00000319 0.00350864 - layer.4.output 0.00018046 0.08069562 - ------------------------------------------------------------------------------------- - TOTAL 0.00244566 1.22975426 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 246560 -BPFP 1.5459 bits/point -EBPFP 3.0919 equivalent bits/point -MSE 1.229754 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.2298 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,352B, BPFP=0.3696 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,688B, BPFP=1.8417 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,108B, BPFP=1.1980 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,056B, BPFP=1.9579 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,416B, BPFP=1.3940 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,812B, BPFP=2.0221 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,132B, BPFP=1.4548 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,204B, BPFP=1.9704 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,056B, BPFP=1.1936 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,324B, BPFP=2.0656 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 61,144B, BPFP=1.2981 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.67483122 - layer.0.v_cache 0.00000027 0.00024513 - layer.1.k_cache 0.00347722 1.07455403 - layer.1.v_cache 0.00000076 0.00082050 - layer.2.k_cache 0.00117662 0.48676632 - layer.2.v_cache 0.00000108 0.00121346 - layer.3.k_cache 0.00141157 0.54547865 - layer.3.v_cache 0.00000195 0.00192206 - layer.4.k_cache 0.00338827 1.06032446 - layer.4.v_cache 0.00000294 0.00342724 - layer.4.output 0.00021021 0.08223228 - ------------------------------------------------------------------------------------- - TOTAL 0.00255851 1.29846516 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 243292 -BPFP 1.4757 bits/point -EBPFP 2.9514 equivalent bits/point -MSE 1.298465 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.2985 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,548B, BPFP=0.4084 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,968B, BPFP=1.8829 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,260B, BPFP=1.1907 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,936B, BPFP=2.0596 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,416B, BPFP=1.3843 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,380B, BPFP=2.0995 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,428B, BPFP=1.4752 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,076B, BPFP=2.0722 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,404B, BPFP=1.2037 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,736B, BPFP=2.1315 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 61,412B, BPFP=1.3787 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 13.90804177 - layer.0.v_cache 0.00000026 0.00024765 - layer.1.k_cache 0.00374749 1.09473919 - layer.1.v_cache 0.00000077 0.00087476 - layer.2.k_cache 0.00116905 0.48845041 - layer.2.v_cache 0.00000114 0.00127956 - layer.3.k_cache 0.00135864 0.55604974 - layer.3.v_cache 0.00000233 0.00216615 - layer.4.k_cache 0.00339113 1.10349791 - layer.4.v_cache 0.00000304 0.00345323 - layer.4.output 0.00030438 0.08924318 - ------------------------------------------------------------------------------------- - TOTAL 0.00262918 1.25112665 - (elements=1,247,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1247232 -Total Bytes 238564 -BPFP 1.5302 bits/point -EBPFP 3.0604 equivalent bits/point -MSE 1.251127 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.2511 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,704B, BPFP=0.3910 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,832B, BPFP=1.8145 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,128B, BPFP=1.1742 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,492B, BPFP=1.9525 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,304B, BPFP=1.3551 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,972B, BPFP=1.9924 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,916B, BPFP=1.4059 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,520B, BPFP=1.9548 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,224B, BPFP=1.1822 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,308B, BPFP=2.0203 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,848B, BPFP=1.3474 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.44488396 - layer.0.v_cache 0.00000028 0.00024874 - layer.1.k_cache 0.00344140 1.06339832 - layer.1.v_cache 0.00000074 0.00086510 - layer.2.k_cache 0.00110852 0.46707661 - layer.2.v_cache 0.00000113 0.00123517 - layer.3.k_cache 0.00135732 0.54817829 - layer.3.v_cache 0.00000223 0.00209866 - layer.4.k_cache 0.00337875 1.01241773 - layer.4.v_cache 0.00000288 0.00326344 - layer.4.output 0.00018208 0.08041968 - ------------------------------------------------------------------------------------- - TOTAL 0.00244665 1.27609605 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 248248 -BPFP 1.4737 bits/point -EBPFP 2.9475 equivalent bits/point -MSE 1.276096 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.2761 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.009s - ------------------------------------------------------------- -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: 7,348B, BPFP=0.3853 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 33,948B, BPFP=1.7800 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,664B, BPFP=1.1359 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 36,964B, BPFP=1.9381 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 25,428B, BPFP=1.3333 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 37,392B, BPFP=1.9606 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 26,368B, BPFP=1.3826 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 36,660B, BPFP=1.9222 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 22,272B, BPFP=1.1678 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 38,160B, BPFP=2.0008 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 101,620B, BPFP=1.3321 -⌛️ [2/4] FRONTEND: Frontend time: 0.253s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.386s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.86356144 - layer.0.v_cache 0.00000028 0.00025417 - layer.1.k_cache 0.00314236 0.91399092 - layer.1.v_cache 0.00000089 0.00089951 - layer.2.k_cache 0.00115430 0.46321352 - layer.2.v_cache 0.00000109 0.00124247 - layer.3.k_cache 0.00134690 0.51866739 - layer.3.v_cache 0.00000207 0.00203286 - layer.4.k_cache 0.00351049 0.98893748 - layer.4.v_cache 0.00000318 0.00358047 - layer.4.output 0.00015300 0.06698414 - ------------------------------------------------------------------------------------- - TOTAL 0.00246521 1.07316549 - (elements=2,136,064) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2136064 -Total Bytes 387824 -BPFP 1.4525 bits/point -EBPFP 2.9050 equivalent bits/point -MSE 1.073165 ----------------------- -------------------------------------------------------- -Time: 0.648s Load: 0.009s, Pack+Encode: 0.253s, Decode+Unpack: 0.386s ----------------------- -------------------------------------------------------- -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 1.0732 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,500B, BPFP=0.3780 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,788B, BPFP=1.8303 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,476B, BPFP=1.2161 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,744B, BPFP=1.9946 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,700B, BPFP=1.4029 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,092B, BPFP=2.0239 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,192B, BPFP=1.4442 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,756B, BPFP=1.9956 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,348B, BPFP=1.2053 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,748B, BPFP=2.0790 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 64,692B, BPFP=1.3586 -⌛️ [2/4] FRONTEND: Frontend time: 0.207s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.297s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.62630208 - layer.0.v_cache 0.00000029 0.00024839 - layer.1.k_cache 0.00337012 1.03140382 - layer.1.v_cache 0.00000087 0.00085040 - layer.2.k_cache 0.00111931 0.47522293 - layer.2.v_cache 0.00000117 0.00119357 - layer.3.k_cache 0.00133665 0.53287539 - layer.3.v_cache 0.00000243 0.00201008 - layer.4.k_cache 0.00331438 1.02767633 - layer.4.v_cache 0.00000316 0.00337363 - layer.4.output 0.00021472 0.09087301 - ------------------------------------------------------------------------------------- - TOTAL 0.00244186 1.21890348 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 250036 -BPFP 1.5003 bits/point -EBPFP 3.0006 equivalent bits/point -MSE 1.218903 ----------------------- -------------------------------------------------------- -Time: 0.511s Load: 0.007s, Pack+Encode: 0.207s, Decode+Unpack: 0.297s ----------------------- -------------------------------------------------------- -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 1.2189 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,104B, BPFP=0.3958 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,896B, BPFP=1.9190 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,856B, BPFP=1.2400 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,968B, BPFP=2.1188 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,980B, BPFP=1.4448 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,780B, BPFP=2.1971 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,584B, BPFP=1.5031 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,040B, BPFP=2.1258 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,044B, BPFP=1.2581 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,792B, BPFP=2.1983 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 59,944B, BPFP=1.4454 -⌛️ [2/4] FRONTEND: Frontend time: 0.213s (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, 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: 0.303s - -[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, 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 14.64456290 - layer.0.v_cache 0.00000026 0.00024892 - layer.1.k_cache 0.00346748 1.17966942 - layer.1.v_cache 0.00000092 0.00090941 - layer.2.k_cache 0.00113221 0.51280128 - layer.2.v_cache 0.00000143 0.00129732 - layer.3.k_cache 0.00133539 0.59295899 - layer.3.v_cache 0.00000229 0.00210371 - layer.4.k_cache 0.00342405 1.12099071 - layer.4.v_cache 0.00000310 0.00354352 - layer.4.output 0.00020696 0.08800032 - ------------------------------------------------------------------------------------- - TOTAL 0.00258429 1.31507767 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 229988 -BPFP 1.5845 bits/point -EBPFP 3.1689 equivalent bits/point -MSE 1.315078 ----------------------- -------------------------------------------------------- -Time: 0.522s Load: 0.006s, Pack+Encode: 0.213s, Decode+Unpack: 0.303s ----------------------- -------------------------------------------------------- -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 1.3151 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 4,472B, BPFP=0.3757 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,692B, BPFP=1.8222 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 14,292B, BPFP=1.2006 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,756B, BPFP=1.9956 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,196B, BPFP=1.3606 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 24,196B, BPFP=2.0326 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,072B, BPFP=1.4341 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,508B, BPFP=1.9748 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,280B, BPFP=1.1996 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,456B, BPFP=2.0544 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,864B, BPFP=1.3202 -⌛️ [2/4] FRONTEND: Frontend time: 0.214s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.290s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.91312006 - layer.0.v_cache 0.00000028 0.00025915 - layer.1.k_cache 0.00336978 1.07558925 - layer.1.v_cache 0.00000078 0.00091169 - layer.2.k_cache 0.00117610 0.48411421 - layer.2.v_cache 0.00000111 0.00129778 - layer.3.k_cache 0.00133121 0.54779816 - layer.3.v_cache 0.00000213 0.00209864 - layer.4.k_cache 0.00346230 1.09997788 - layer.4.v_cache 0.00000315 0.00355331 - layer.4.output 0.00018648 0.08042985 - ------------------------------------------------------------------------------------- - TOTAL 0.00247111 1.31788854 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 246784 -BPFP 1.4808 bits/point -EBPFP 2.9616 equivalent bits/point -MSE 1.317889 ----------------------- -------------------------------------------------------- -Time: 0.511s Load: 0.006s, Pack+Encode: 0.214s, Decode+Unpack: 0.290s ----------------------- -------------------------------------------------------- -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 1.3179 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,400B, BPFP=0.3819 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,752B, BPFP=1.8882 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,640B, BPFP=1.1840 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,092B, BPFP=2.0045 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 16,112B, BPFP=1.3986 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,804B, BPFP=2.0663 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,008B, BPFP=1.4764 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,412B, BPFP=2.0323 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,480B, BPFP=1.1701 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,968B, BPFP=2.0806 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,076B, BPFP=1.3037 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.286s - -[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, 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 16.08494331 - layer.0.v_cache 0.00000028 0.00024530 - layer.1.k_cache 0.00333766 1.07414042 - layer.1.v_cache 0.00000075 0.00082555 - layer.2.k_cache 0.00114903 0.48203447 - layer.2.v_cache 0.00000112 0.00117555 - layer.3.k_cache 0.00131913 0.56347258 - layer.3.v_cache 0.00000216 0.00198241 - layer.4.k_cache 0.00329656 1.06162389 - layer.4.v_cache 0.00000300 0.00328609 - layer.4.output 0.00024019 0.08854710 - ------------------------------------------------------------------------------------- - TOTAL 0.00279529 1.40199414 - (elements=1,290,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1290240 -Total Bytes 240744 -BPFP 1.4927 bits/point -EBPFP 2.9854 equivalent bits/point -MSE 1.401994 ----------------------- -------------------------------------------------------- -Time: 0.497s Load: 0.006s, Pack+Encode: 0.206s, Decode+Unpack: 0.286s ----------------------- -------------------------------------------------------- -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 1.4020 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 5,912B, BPFP=0.3881 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,344B, BPFP=1.7295 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,224B, BPFP=1.1308 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,276B, BPFP=1.8564 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,220B, BPFP=1.3275 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,712B, BPFP=1.8850 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,672B, BPFP=1.3571 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,992B, BPFP=1.8377 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,392B, BPFP=1.1418 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,668B, BPFP=1.8821 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 83,344B, BPFP=1.3679 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.69977576 - layer.0.v_cache 0.00000028 0.00025293 - layer.1.k_cache 0.00331476 0.92155508 - layer.1.v_cache 0.00000080 0.00091235 - layer.2.k_cache 0.00114230 0.47909245 - layer.2.v_cache 0.00000115 0.00133406 - layer.3.k_cache 0.00138709 0.53514926 - layer.3.v_cache 0.00000225 0.00220117 - layer.4.k_cache 0.00333808 0.97908494 - layer.4.v_cache 0.00000314 0.00357534 - layer.4.output 0.00020645 0.08107782 - ------------------------------------------------------------------------------------- - TOTAL 0.00247129 1.13908890 - (elements=1,705,984) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1705984 -Total Bytes 304756 -BPFP 1.4291 bits/point -EBPFP 2.8582 equivalent bits/point -MSE 1.139089 ----------------------- -------------------------------------------------------- -Time: 0.497s Load: 0.008s, Pack+Encode: 0.204s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.1391 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,124B, BPFP=0.3929 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,188B, BPFP=1.9234 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,988B, BPFP=1.2374 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,232B, BPFP=2.1181 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,960B, BPFP=1.4253 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,764B, BPFP=2.1688 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,500B, BPFP=1.4768 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,240B, BPFP=2.1189 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,652B, BPFP=1.2054 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,716B, BPFP=2.1643 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 57,900B, BPFP=1.3791 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.28053842 - layer.0.v_cache 0.00000027 0.00024958 - layer.1.k_cache 0.00344214 1.20252535 - layer.1.v_cache 0.00000078 0.00086249 - layer.2.k_cache 0.00114699 0.49319849 - layer.2.v_cache 0.00000119 0.00125557 - layer.3.k_cache 0.00140111 0.55650474 - layer.3.v_cache 0.00000200 0.00199003 - layer.4.k_cache 0.00332288 1.09543824 - layer.4.v_cache 0.00000299 0.00327930 - layer.4.output 0.00018752 0.08920751 - ------------------------------------------------------------------------------------- - TOTAL 0.00258402 1.35661945 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 228264 -BPFP 1.5534 bits/point -EBPFP 3.1068 equivalent bits/point -MSE 1.356619 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.3566 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.009s - ------------------------------------------------------------- -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: 6,220B, BPFP=0.3919 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,492B, BPFP=1.6691 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,848B, BPFP=1.1245 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,284B, BPFP=1.7820 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,764B, BPFP=1.3082 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,444B, BPFP=1.7921 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,460B, BPFP=1.3521 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,948B, BPFP=1.7608 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,340B, BPFP=1.1555 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,612B, BPFP=1.8027 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 81,412B, BPFP=1.2823 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 12.49716679 - layer.0.v_cache 0.00000027 0.00024844 - layer.1.k_cache 0.00324291 0.87260609 - layer.1.v_cache 0.00000080 0.00084574 - layer.2.k_cache 0.00117169 0.45266791 - layer.2.v_cache 0.00000108 0.00120378 - layer.3.k_cache 0.00137525 0.54599737 - layer.3.v_cache 0.00000203 0.00190600 - layer.4.k_cache 0.00340243 0.96029503 - layer.4.v_cache 0.00000299 0.00334391 - layer.4.output 0.00019780 0.07785213 - ------------------------------------------------------------------------------------- - TOTAL 0.00255239 1.11769211 - (elements=1,777,664) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1777664 -Total Bytes 305824 -BPFP 1.3763 bits/point -EBPFP 2.7526 equivalent bits/point -MSE 1.117692 ----------------------- -------------------------------------------------------- -Time: 0.501s Load: 0.009s, Pack+Encode: 0.205s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.1177 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,164B, BPFP=0.3827 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,108B, BPFP=1.8482 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,964B, BPFP=1.1915 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,888B, BPFP=2.0118 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,116B, BPFP=1.3893 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,500B, BPFP=2.0680 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,016B, BPFP=1.4721 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,156B, BPFP=2.0364 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,068B, BPFP=1.2011 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,120B, BPFP=2.1250 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 58,000B, BPFP=1.3327 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.96664752 - layer.0.v_cache 0.00000028 0.00026557 - layer.1.k_cache 0.00337091 1.13849415 - layer.1.v_cache 0.00000080 0.00083670 - layer.2.k_cache 0.00114287 0.51136017 - layer.2.v_cache 0.00000105 0.00117402 - layer.3.k_cache 0.00139664 0.57330475 - layer.3.v_cache 0.00000196 0.00192182 - layer.4.k_cache 0.00332559 1.16752212 - layer.4.v_cache 0.00000294 0.00337590 - layer.4.output 0.00025832 0.10541409 - ------------------------------------------------------------------------------------- - TOTAL 0.00265546 1.41332565 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 229100 -BPFP 1.5041 bits/point -EBPFP 3.0081 equivalent bits/point -MSE 1.413326 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.4133 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 3,732B, BPFP=0.3940 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,304B, BPFP=2.0380 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,396B, BPFP=1.3087 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,416B, BPFP=2.2610 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,400B, BPFP=1.5203 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,224B, BPFP=2.3463 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,512B, BPFP=1.6377 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,428B, BPFP=2.2622 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,540B, BPFP=1.3239 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,192B, BPFP=2.3429 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 55,364B, BPFP=1.4613 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 16.47197662 - layer.0.v_cache 0.00000027 0.00026453 - layer.1.k_cache 0.00342263 1.24955636 - layer.1.v_cache 0.00000079 0.00097888 - layer.2.k_cache 0.00118243 0.52128385 - layer.2.v_cache 0.00000111 0.00129619 - layer.3.k_cache 0.00141981 0.60637345 - layer.3.v_cache 0.00000233 0.00215341 - layer.4.k_cache 0.00326747 1.21055789 - layer.4.v_cache 0.00000299 0.00347041 - layer.4.output 0.00023783 0.10583587 - ------------------------------------------------------------------------------------- - TOTAL 0.00280131 1.46366108 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 220508 -BPFP 1.6629 bits/point -EBPFP 3.3257 equivalent bits/point -MSE 1.463661 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.007s, Pack+Encode: 0.201s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -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 1.4637 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 7,028B, BPFP=0.4358 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,964B, BPFP=1.6099 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 18,056B, BPFP=1.1195 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,776B, BPFP=1.7222 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,572B, BPFP=1.2755 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,960B, BPFP=1.7336 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,192B, BPFP=1.3140 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,480B, BPFP=1.7039 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,524B, BPFP=1.1486 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,180B, BPFP=1.7473 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 79,076B, BPFP=1.2258 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.289s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.26818073 - layer.0.v_cache 0.00000026 0.00023860 - layer.1.k_cache 0.00313964 0.81256225 - layer.1.v_cache 0.00000077 0.00085272 - layer.2.k_cache 0.00115655 0.46417981 - layer.2.v_cache 0.00000107 0.00120056 - layer.3.k_cache 0.00144242 0.53400373 - layer.3.v_cache 0.00000205 0.00196623 - layer.4.k_cache 0.00340494 0.98988875 - layer.4.v_cache 0.00000295 0.00337276 - layer.4.output 0.00023638 0.08218713 - ------------------------------------------------------------------------------------- - TOTAL 0.00263232 0.95751390 - (elements=1,806,336) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1806336 -Total Bytes 301808 -BPFP 1.3367 bits/point -EBPFP 2.6733 equivalent bits/point -MSE 0.957514 ----------------------- -------------------------------------------------------- -Time: 0.501s Load: 0.008s, Pack+Encode: 0.205s, Decode+Unpack: 0.289s ----------------------- -------------------------------------------------------- -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 0.9575 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,748B, BPFP=0.3853 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,956B, BPFP=1.9486 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,552B, BPFP=1.2903 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 20,972B, BPFP=2.1558 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,316B, BPFP=1.4716 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,460B, BPFP=2.2060 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,340B, BPFP=1.5769 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,088B, BPFP=2.1678 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,692B, BPFP=1.3047 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,596B, BPFP=2.2200 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 56,100B, BPFP=1.4417 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.97657374 - layer.0.v_cache 0.00000027 0.00025150 - layer.1.k_cache 0.00360181 1.21341043 - layer.1.v_cache 0.00000073 0.00083001 - layer.2.k_cache 0.00113352 0.53067860 - layer.2.v_cache 0.00000108 0.00124327 - layer.3.k_cache 0.00139474 0.59887886 - layer.3.v_cache 0.00000224 0.00207391 - layer.4.k_cache 0.00318908 1.23018787 - layer.4.v_cache 0.00000296 0.00345628 - layer.4.output 0.00022408 0.11564054 - ------------------------------------------------------------------------------------- - TOTAL 0.00266343 1.43001047 - (elements=1,089,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1089536 -Total Bytes 218820 -BPFP 1.6067 bits/point -EBPFP 3.2134 equivalent bits/point -MSE 1.430010 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.4300 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,392B, BPFP=0.4037 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,312B, BPFP=1.8669 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,084B, BPFP=1.2026 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,072B, BPFP=2.0287 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,288B, BPFP=1.4051 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,680B, BPFP=2.0846 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,084B, BPFP=1.4783 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,160B, BPFP=2.0368 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,988B, BPFP=1.1938 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,972B, BPFP=2.1114 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 59,308B, BPFP=1.3628 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.92465389 - layer.0.v_cache 0.00000028 0.00023768 - layer.1.k_cache 0.00340719 1.07873715 - layer.1.v_cache 0.00000073 0.00077272 - layer.2.k_cache 0.00115796 0.49523603 - layer.2.v_cache 0.00000109 0.00113925 - layer.3.k_cache 0.00135314 0.55090435 - layer.3.v_cache 0.00000196 0.00181799 - layer.4.k_cache 0.00328730 1.08786056 - layer.4.v_cache 0.00000286 0.00318326 - layer.4.output 0.00022769 0.09548336 - ------------------------------------------------------------------------------------- - TOTAL 0.00248105 1.25189117 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 231340 -BPFP 1.5188 bits/point -EBPFP 3.0376 equivalent bits/point -MSE 1.251891 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.007s, Pack+Encode: 0.199s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -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 1.2519 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 6,528B, BPFP=0.3806 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,048B, BPFP=1.8102 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,732B, BPFP=1.2087 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,720B, BPFP=1.9660 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,316B, BPFP=1.4177 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,160B, BPFP=1.9916 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 25,384B, BPFP=1.4799 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 33,920B, BPFP=1.9776 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,316B, BPFP=1.2428 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 34,920B, BPFP=2.0359 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 86,980B, BPFP=1.2678 -⌛️ [2/4] FRONTEND: Frontend time: 0.247s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.385s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.29041985 - layer.0.v_cache 0.00000029 0.00025425 - layer.1.k_cache 0.00318200 0.94481169 - layer.1.v_cache 0.00000078 0.00082654 - layer.2.k_cache 0.00115661 0.47366191 - layer.2.v_cache 0.00000102 0.00111259 - layer.3.k_cache 0.00142778 0.55432806 - layer.3.v_cache 0.00000209 0.00198537 - layer.4.k_cache 0.00344207 1.04761733 - layer.4.v_cache 0.00000292 0.00324563 - layer.4.output 0.00019652 0.08521235 - ------------------------------------------------------------------------------------- - TOTAL 0.00258848 1.26136519 - (elements=1,921,024) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1921024 -Total Bytes 353024 -BPFP 1.4701 bits/point -EBPFP 2.9403 equivalent bits/point -MSE 1.261365 ----------------------- -------------------------------------------------------- -Time: 0.640s Load: 0.008s, Pack+Encode: 0.247s, Decode+Unpack: 0.385s ----------------------- -------------------------------------------------------- -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 1.2614 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.009s - ------------------------------------------------------------- -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: 6,444B, BPFP=0.3814 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,084B, BPFP=1.8397 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,148B, BPFP=1.1925 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,976B, BPFP=2.0109 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,184B, BPFP=1.4313 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,836B, BPFP=2.0618 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 25,304B, BPFP=1.4976 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,328B, BPFP=2.0317 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,292B, BPFP=1.2602 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,244B, BPFP=2.0859 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 95,248B, BPFP=1.4093 -⌛️ [2/4] FRONTEND: Frontend time: 0.252s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.384s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.28859641 - layer.0.v_cache 0.00000027 0.00024790 - layer.1.k_cache 0.00323066 0.99261521 - layer.1.v_cache 0.00000090 0.00089481 - layer.2.k_cache 0.00115984 0.47793972 - layer.2.v_cache 0.00000123 0.00128244 - layer.3.k_cache 0.00138390 0.56394126 - layer.3.v_cache 0.00000215 0.00214032 - layer.4.k_cache 0.00364779 1.02419258 - layer.4.v_cache 0.00000297 0.00330681 - layer.4.output 0.00020533 0.08870221 - ------------------------------------------------------------------------------------- - TOTAL 0.00254814 1.19356902 - (elements=1,892,352) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1892352 -Total Bytes 362088 -BPFP 1.5307 bits/point -EBPFP 3.0615 equivalent bits/point -MSE 1.193569 ----------------------- -------------------------------------------------------- -Time: 0.645s Load: 0.009s, Pack+Encode: 0.252s, Decode+Unpack: 0.384s ----------------------- -------------------------------------------------------- -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 1.1936 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,016B, BPFP=0.4075 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,160B, BPFP=2.0455 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,420B, BPFP=1.2601 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,888B, BPFP=2.2208 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,784B, BPFP=1.5000 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,612B, BPFP=2.2942 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,552B, BPFP=1.5779 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,148B, BPFP=2.2472 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,856B, BPFP=1.3044 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,816B, BPFP=2.3149 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 59,596B, BPFP=1.5117 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 15.61783748 - layer.0.v_cache 0.00000028 0.00026574 - layer.1.k_cache 0.00355533 1.20482259 - layer.1.v_cache 0.00000093 0.00093000 - layer.2.k_cache 0.00115642 0.52177603 - layer.2.v_cache 0.00000119 0.00133389 - layer.3.k_cache 0.00136722 0.58104958 - layer.3.v_cache 0.00000230 0.00219257 - layer.4.k_cache 0.00337541 1.10262814 - layer.4.v_cache 0.00000304 0.00366113 - layer.4.output 0.00023573 0.10184857 - ------------------------------------------------------------------------------------- - TOTAL 0.00261829 1.38884939 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 228848 -BPFP 1.6585 bits/point -EBPFP 3.3170 equivalent bits/point -MSE 1.388849 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.007s, Pack+Encode: 0.201s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.3888 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 4,628B, BPFP=0.3846 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 21,760B, BPFP=1.8085 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,948B, BPFP=1.1592 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 23,084B, BPFP=1.9186 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,924B, BPFP=1.3235 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 23,744B, BPFP=1.9734 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 17,000B, BPFP=1.4129 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 23,552B, BPFP=1.9574 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 14,296B, BPFP=1.1882 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 24,320B, BPFP=2.0213 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 66,144B, BPFP=1.3743 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.64579870 - layer.0.v_cache 0.00000029 0.00025884 - layer.1.k_cache 0.00339276 1.03956710 - layer.1.v_cache 0.00000074 0.00076936 - layer.2.k_cache 0.00116230 0.46220333 - layer.2.v_cache 0.00000115 0.00117755 - layer.3.k_cache 0.00135573 0.53181283 - layer.3.v_cache 0.00000225 0.00197876 - layer.4.k_cache 0.00337629 1.05693233 - layer.4.v_cache 0.00000290 0.00302885 - layer.4.output 0.00017695 0.08609026 - ------------------------------------------------------------------------------------- - TOTAL 0.00232786 1.22056348 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 248400 -BPFP 1.4746 bits/point -EBPFP 2.9493 equivalent bits/point -MSE 1.220563 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.007s, Pack+Encode: 0.199s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.2206 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,876B, BPFP=0.3933 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,428B, BPFP=1.9712 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,492B, BPFP=1.2675 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,428B, BPFP=2.1741 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,772B, BPFP=1.4988 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,228B, BPFP=2.2553 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,360B, BPFP=1.5584 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,648B, BPFP=2.1964 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,700B, BPFP=1.2886 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,396B, BPFP=2.2723 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 57,084B, BPFP=1.4480 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 16.27410968 - layer.0.v_cache 0.00000027 0.00026819 - layer.1.k_cache 0.00339772 1.25371264 - layer.1.v_cache 0.00000076 0.00093280 - layer.2.k_cache 0.00114556 0.54294398 - layer.2.v_cache 0.00000112 0.00134675 - layer.3.k_cache 0.00146787 0.61489046 - layer.3.v_cache 0.00000204 0.00215678 - layer.4.k_cache 0.00318992 1.22423702 - layer.4.v_cache 0.00000300 0.00363360 - layer.4.output 0.00024419 0.11676769 - ------------------------------------------------------------------------------------- - TOTAL 0.00266143 1.45609305 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 223412 -BPFP 1.6191 bits/point -EBPFP 3.2382 equivalent bits/point -MSE 1.456093 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.006s, Pack+Encode: 0.199s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.4561 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 5,336B, BPFP=0.3722 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,316B, BPFP=1.7659 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,260B, BPFP=1.1342 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 26,976B, BPFP=1.8817 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,056B, BPFP=1.3292 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,364B, BPFP=1.9088 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,860B, BPFP=1.3853 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,760B, BPFP=1.8666 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,652B, BPFP=1.1616 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,520B, BPFP=1.9196 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 73,228B, BPFP=1.2770 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.65539333 - layer.0.v_cache 0.00000026 0.00024913 - layer.1.k_cache 0.00327242 0.94267150 - layer.1.v_cache 0.00000072 0.00083555 - layer.2.k_cache 0.00117792 0.49384485 - layer.2.v_cache 0.00000108 0.00120536 - layer.3.k_cache 0.00137547 0.55666542 - layer.3.v_cache 0.00000205 0.00200910 - layer.4.k_cache 0.00331726 1.07306984 - layer.4.v_cache 0.00000297 0.00335912 - layer.4.output 0.00024922 0.09535772 - ------------------------------------------------------------------------------------- - TOTAL 0.00263713 1.07933815 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 284328 -BPFP 1.4167 bits/point -EBPFP 2.8333 equivalent bits/point -MSE 1.079338 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -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 1.0793 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,972B, BPFP=0.3978 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,512B, BPFP=1.9543 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,568B, BPFP=1.2588 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,364B, BPFP=2.1398 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,816B, BPFP=1.4840 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,452B, BPFP=2.2488 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,436B, BPFP=1.5461 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,880B, BPFP=2.1915 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,844B, BPFP=1.2865 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,632B, BPFP=2.2668 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 59,132B, BPFP=1.4807 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.05928861 - layer.0.v_cache 0.00000030 0.00025418 - layer.1.k_cache 0.00350441 1.19667308 - layer.1.v_cache 0.00000082 0.00091506 - layer.2.k_cache 0.00114279 0.52174681 - layer.2.v_cache 0.00000132 0.00138869 - layer.3.k_cache 0.00135780 0.57692342 - layer.3.v_cache 0.00000231 0.00225489 - layer.4.k_cache 0.00328533 1.13537294 - layer.4.v_cache 0.00000329 0.00381525 - layer.4.output 0.00022272 0.10439224 - ------------------------------------------------------------------------------------- - TOTAL 0.00254428 1.27972870 - (elements=1,118,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1118208 -Total Bytes 226608 -BPFP 1.6212 bits/point -EBPFP 3.2424 equivalent bits/point -MSE 1.279729 ----------------------- -------------------------------------------------------- -Time: 0.485s Load: 0.006s, Pack+Encode: 0.199s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -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 1.2797 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,936B, BPFP=0.3892 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,496B, BPFP=1.9280 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,800B, BPFP=1.2658 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,428B, BPFP=2.1191 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,436B, BPFP=1.4276 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,904B, BPFP=2.1661 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,244B, BPFP=1.5075 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,664B, BPFP=2.1424 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,608B, BPFP=1.2468 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,388B, BPFP=2.2140 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 55,276B, BPFP=1.3666 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.51745605 - layer.0.v_cache 0.00000029 0.00025550 - layer.1.k_cache 0.00331158 1.20601529 - layer.1.v_cache 0.00000077 0.00088042 - layer.2.k_cache 0.00112976 0.50541088 - layer.2.v_cache 0.00000111 0.00122398 - layer.3.k_cache 0.00133964 0.56980486 - layer.3.v_cache 0.00000223 0.00206053 - layer.4.k_cache 0.00332594 1.13602100 - layer.4.v_cache 0.00000299 0.00322377 - layer.4.output 0.00020400 0.10060672 - ------------------------------------------------------------------------------------- - TOTAL 0.00252135 1.31034137 - (elements=1,132,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1132544 -Total Bytes 221180 -BPFP 1.5624 bits/point -EBPFP 3.1247 equivalent bits/point -MSE 1.310341 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.006s, Pack+Encode: 0.199s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.3103 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 4,204B, BPFP=0.4055 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,340B, BPFP=1.9618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,044B, BPFP=1.2581 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,296B, BPFP=2.1505 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,204B, BPFP=1.4664 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,796B, BPFP=2.1987 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,940B, BPFP=1.5374 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,816B, BPFP=2.2006 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,312B, BPFP=1.2840 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,168B, BPFP=2.2346 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 62,956B, BPFP=1.5180 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.66884735 - layer.0.v_cache 0.00000029 0.00026563 - layer.1.k_cache 0.00348817 1.18706371 - layer.1.v_cache 0.00000087 0.00091949 - layer.2.k_cache 0.00114702 0.51782533 - layer.2.v_cache 0.00000128 0.00142080 - layer.3.k_cache 0.00135614 0.56971331 - layer.3.v_cache 0.00000252 0.00219869 - layer.4.k_cache 0.00328397 1.10929042 - layer.4.v_cache 0.00000321 0.00362188 - layer.4.output 0.00022006 0.10571792 - ------------------------------------------------------------------------------------- - TOTAL 0.00262492 1.32028845 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 236076 -BPFP 1.6264 bits/point -EBPFP 3.2528 equivalent bits/point -MSE 1.320288 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.007s, Pack+Encode: 0.200s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.3203 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 5,796B, BPFP=0.3870 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,888B, BPFP=1.7286 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,936B, BPFP=1.1309 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,868B, BPFP=1.8608 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,744B, BPFP=1.3184 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,160B, BPFP=1.8803 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,412B, BPFP=1.3630 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,588B, BPFP=1.8421 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,180B, BPFP=1.1472 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,328B, BPFP=1.8916 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 78,744B, BPFP=1.3145 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 12.76514715 - layer.0.v_cache 0.00000026 0.00023705 - layer.1.k_cache 0.00326448 0.95160538 - layer.1.v_cache 0.00000079 0.00087640 - layer.2.k_cache 0.00117087 0.48222853 - layer.2.v_cache 0.00000122 0.00129044 - layer.3.k_cache 0.00141366 0.54434250 - layer.3.v_cache 0.00000241 0.00211418 - layer.4.k_cache 0.00339367 1.05597289 - layer.4.v_cache 0.00000311 0.00343631 - layer.4.output 0.00025967 0.09894658 - ------------------------------------------------------------------------------------- - TOTAL 0.00263308 1.15735980 - (elements=1,677,312) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1677312 -Total Bytes 296644 -BPFP 1.4149 bits/point -EBPFP 2.8297 equivalent bits/point -MSE 1.157360 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.1574 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 4,128B, BPFP=0.3839 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 20,260B, BPFP=1.8843 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 13,164B, BPFP=1.2243 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 22,460B, BPFP=2.0889 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 15,200B, BPFP=1.4137 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 22,852B, BPFP=2.1254 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 16,040B, BPFP=1.4918 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 22,376B, BPFP=2.0811 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 13,228B, BPFP=1.2303 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 23,316B, BPFP=2.1685 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 60,632B, BPFP=1.4098 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 15.61908250 - layer.0.v_cache 0.00000028 0.00024890 - layer.1.k_cache 0.00342466 1.10922595 - layer.1.v_cache 0.00000080 0.00086476 - layer.2.k_cache 0.00116354 0.48586518 - layer.2.v_cache 0.00000112 0.00123762 - layer.3.k_cache 0.00133515 0.54713199 - layer.3.v_cache 0.00000214 0.00200783 - layer.4.k_cache 0.00358854 1.10606557 - layer.4.v_cache 0.00000313 0.00336581 - layer.4.output 0.00016584 0.08991419 - ------------------------------------------------------------------------------------- - TOTAL 0.00257403 1.37391092 - (elements=1,204,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1204224 -Total Bytes 233656 -BPFP 1.5522 bits/point -EBPFP 3.1045 equivalent bits/point -MSE 1.373911 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.3739 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 6,364B, BPFP=0.3738 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 31,200B, BPFP=1.8327 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,348B, BPFP=1.1953 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 33,976B, BPFP=1.9958 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,104B, BPFP=1.4159 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 34,700B, BPFP=2.0383 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 25,024B, BPFP=1.4699 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 34,056B, BPFP=2.0005 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 20,792B, BPFP=1.2213 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 35,132B, BPFP=2.0637 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 92,644B, BPFP=1.3605 -⌛️ [2/4] FRONTEND: Frontend time: 0.246s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.381s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 13.79287146 - layer.0.v_cache 0.00000028 0.00025778 - layer.1.k_cache 0.00324298 0.97873131 - layer.1.v_cache 0.00000087 0.00084610 - layer.2.k_cache 0.00114197 0.46777011 - layer.2.v_cache 0.00000110 0.00126124 - layer.3.k_cache 0.00134409 0.54349070 - layer.3.v_cache 0.00000211 0.00204908 - layer.4.k_cache 0.00375619 1.04712224 - layer.4.v_cache 0.00000306 0.00346633 - layer.4.output 0.00014515 0.07301389 - ------------------------------------------------------------------------------------- - TOTAL 0.00232996 1.22356585 - (elements=1,906,688) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1906688 -Total Bytes 358340 -BPFP 1.5035 bits/point -EBPFP 3.0070 equivalent bits/point -MSE 1.223566 ----------------------- -------------------------------------------------------- -Time: 0.635s Load: 0.008s, Pack+Encode: 0.246s, Decode+Unpack: 0.381s ----------------------- -------------------------------------------------------- -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 1.2236 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 6,424B, BPFP=0.4015 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,856B, BPFP=1.6160 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,876B, BPFP=1.1173 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,828B, BPFP=1.7392 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,572B, BPFP=1.2857 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,184B, BPFP=1.7615 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,260B, BPFP=1.3288 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,804B, BPFP=1.7377 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,192B, BPFP=1.1370 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,336B, BPFP=1.7710 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 80,624B, BPFP=1.2597 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.286s - -[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, 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 12.09155371 - layer.0.v_cache 0.00000027 0.00024368 - layer.1.k_cache 0.00318185 0.81388483 - layer.1.v_cache 0.00000072 0.00081389 - layer.2.k_cache 0.00113729 0.44547852 - layer.2.v_cache 0.00000107 0.00118096 - layer.3.k_cache 0.00136705 0.51610046 - layer.3.v_cache 0.00000203 0.00196933 - layer.4.k_cache 0.00364986 0.92769836 - layer.4.v_cache 0.00000297 0.00317668 - layer.4.output 0.00019454 0.07483486 - ------------------------------------------------------------------------------------- - TOTAL 0.00247738 1.07867428 - (elements=1,792,000) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1792000 -Total Bytes 302956 -BPFP 1.3525 bits/point -EBPFP 2.7050 equivalent bits/point -MSE 1.078674 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.286s ----------------------- -------------------------------------------------------- -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 1.0787 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,644B, BPFP=0.4010 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,924B, BPFP=1.9723 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,084B, BPFP=1.3297 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 20,244B, BPFP=2.2276 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,788B, BPFP=1.5172 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,064B, BPFP=2.3178 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,756B, BPFP=1.6237 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,720B, BPFP=2.2799 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,304B, BPFP=1.3539 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,032B, BPFP=2.3143 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 55,588B, BPFP=1.5292 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 16.83804192 - layer.0.v_cache 0.00000028 0.00026840 - layer.1.k_cache 0.00376352 1.23551855 - layer.1.v_cache 0.00000080 0.00095532 - layer.2.k_cache 0.00117123 0.51594646 - layer.2.v_cache 0.00000118 0.00129946 - layer.3.k_cache 0.00138233 0.59983707 - layer.3.v_cache 0.00000223 0.00221221 - layer.4.k_cache 0.00337686 1.17514221 - layer.4.v_cache 0.00000300 0.00359700 - layer.4.output 0.00019204 0.09821696 - ------------------------------------------------------------------------------------- - TOTAL 0.00278304 1.48326332 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 213148 -BPFP 1.6753 bits/point -EBPFP 3.3505 equivalent bits/point -MSE 1.483263 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.006s, Pack+Encode: 0.199s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.4833 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.004s - ------------------------------------------------------------- -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: 3,656B, BPFP=0.4200 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,776B, BPFP=1.9274 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 11,604B, BPFP=1.3332 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,028B, BPFP=2.1861 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,900B, BPFP=1.5970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 19,972B, BPFP=2.2946 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,648B, BPFP=1.6829 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 19,896B, BPFP=2.2858 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,444B, BPFP=1.3148 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 20,160B, BPFP=2.3162 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 53,460B, BPFP=1.5355 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 16.67268102 - layer.0.v_cache 0.00000027 0.00027065 - layer.1.k_cache 0.00364027 1.33006758 - layer.1.v_cache 0.00000080 0.00101146 - layer.2.k_cache 0.00115812 0.55406615 - layer.2.v_cache 0.00000117 0.00141147 - layer.3.k_cache 0.00137295 0.60441404 - layer.3.v_cache 0.00000227 0.00239040 - layer.4.k_cache 0.00327663 1.18702642 - layer.4.v_cache 0.00000341 0.00402742 - layer.4.output 0.00018782 0.10159283 - ------------------------------------------------------------------------------------- - TOTAL 0.00272859 1.48312414 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 204544 -BPFP 1.6786 bits/point -EBPFP 3.3571 equivalent bits/point -MSE 1.483124 ----------------------- -------------------------------------------------------- -Time: 0.483s Load: 0.004s, Pack+Encode: 0.199s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.4831 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,704B, BPFP=0.4019 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,080B, BPFP=1.9618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,220B, BPFP=1.3260 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,992B, BPFP=2.1693 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,300B, BPFP=1.5516 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,860B, BPFP=2.2635 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,160B, BPFP=1.6450 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,628B, BPFP=2.2383 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,316B, BPFP=1.3364 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 20,812B, BPFP=2.2582 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,700B, BPFP=1.4296 -⌛️ [2/4] FRONTEND: Frontend time: 0.199s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.279s - -[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, 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 16.79572211 - layer.0.v_cache 0.00000028 0.00027464 - layer.1.k_cache 0.00362044 1.26617813 - layer.1.v_cache 0.00000088 0.00094041 - layer.2.k_cache 0.00116887 0.51998372 - layer.2.v_cache 0.00000122 0.00137391 - layer.3.k_cache 0.00138034 0.60859013 - layer.3.v_cache 0.00000238 0.00227341 - layer.4.k_cache 0.00335768 1.15008651 - layer.4.v_cache 0.00000294 0.00349847 - layer.4.output 0.00021991 0.09688889 - ------------------------------------------------------------------------------------- - TOTAL 0.00266936 1.48117693 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 210772 -BPFP 1.6336 bits/point -EBPFP 3.2672 equivalent bits/point -MSE 1.481177 ----------------------- -------------------------------------------------------- -Time: 0.484s Load: 0.006s, Pack+Encode: 0.199s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -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 1.4812 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,692B, BPFP=0.3951 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,140B, BPFP=1.9414 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 11,880B, BPFP=1.2714 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 20,272B, BPFP=2.1695 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,816B, BPFP=1.4786 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,976B, BPFP=2.2449 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,432B, BPFP=1.5445 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,368B, BPFP=2.1798 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,940B, BPFP=1.2778 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,000B, BPFP=2.2474 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 51,792B, BPFP=1.3857 -⌛️ [2/4] FRONTEND: Frontend time: 0.198s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 16.19381421 - layer.0.v_cache 0.00000028 0.00026390 - layer.1.k_cache 0.00357364 1.17209866 - layer.1.v_cache 0.00000075 0.00090147 - layer.2.k_cache 0.00114681 0.51715098 - layer.2.v_cache 0.00000108 0.00125909 - layer.3.k_cache 0.00136272 0.58566023 - layer.3.v_cache 0.00000212 0.00202954 - layer.4.k_cache 0.00341161 1.17503733 - layer.4.v_cache 0.00000293 0.00346256 - layer.4.output 0.00018354 0.09572816 - ------------------------------------------------------------------------------------- - TOTAL 0.00259573 1.43104219 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 208308 -BPFP 1.5924 bits/point -EBPFP 3.1847 equivalent bits/point -MSE 1.431042 ----------------------- -------------------------------------------------------- -Time: 0.485s Load: 0.005s, Pack+Encode: 0.198s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.4310 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,608B, BPFP=0.3915 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,920B, BPFP=1.9444 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,212B, BPFP=1.3251 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,476B, BPFP=2.1133 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,032B, BPFP=1.5226 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,656B, BPFP=2.2413 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,836B, BPFP=1.6098 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,608B, BPFP=2.2361 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,304B, BPFP=1.3351 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 20,976B, BPFP=2.2760 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,320B, BPFP=1.4193 -⌛️ [2/4] FRONTEND: Frontend time: 0.207s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.293s - -[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, 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 17.04265171 - layer.0.v_cache 0.00000029 0.00026883 - layer.1.k_cache 0.00335621 1.16433080 - layer.1.v_cache 0.00000072 0.00084129 - layer.2.k_cache 0.00111533 0.49373086 - layer.2.v_cache 0.00000112 0.00120967 - layer.3.k_cache 0.00138591 0.58048333 - layer.3.v_cache 0.00000206 0.00198838 - layer.4.k_cache 0.00323834 1.12102445 - layer.4.v_cache 0.00000293 0.00337521 - layer.4.output 0.00022510 0.09880121 - ------------------------------------------------------------------------------------- - TOTAL 0.00266260 1.48607924 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 208948 -BPFP 1.6195 bits/point -EBPFP 3.2389 equivalent bits/point -MSE 1.486079 ----------------------- -------------------------------------------------------- -Time: 0.506s Load: 0.006s, Pack+Encode: 0.207s, Decode+Unpack: 0.293s ----------------------- -------------------------------------------------------- -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 1.4861 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 5,736B, BPFP=0.3897 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,180B, BPFP=1.7785 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,868B, BPFP=1.1459 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,980B, BPFP=1.9008 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,772B, BPFP=1.3432 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,208B, BPFP=1.9163 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,408B, BPFP=1.3864 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,656B, BPFP=1.8788 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,216B, BPFP=1.1696 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,420B, BPFP=1.9307 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 81,156B, BPFP=1.3783 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.288s - -[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, 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 12.04249533 - layer.0.v_cache 0.00000027 0.00025520 - layer.1.k_cache 0.00345256 0.93037176 - layer.1.v_cache 0.00000081 0.00091894 - layer.2.k_cache 0.00114556 0.46706570 - layer.2.v_cache 0.00000111 0.00127427 - layer.3.k_cache 0.00139466 0.54852295 - layer.3.v_cache 0.00000215 0.00212160 - layer.4.k_cache 0.00339940 1.02309584 - layer.4.v_cache 0.00000318 0.00373823 - layer.4.output 0.00018696 0.07668838 - ------------------------------------------------------------------------------------- - TOTAL 0.00250116 1.09475809 - (elements=1,648,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1648640 -Total Bytes 299600 -BPFP 1.4538 bits/point -EBPFP 2.9076 equivalent bits/point -MSE 1.094758 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.008s, Pack+Encode: 0.201s, Decode+Unpack: 0.288s ----------------------- -------------------------------------------------------- -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 1.0948 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,736B, BPFP=0.4111 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 17,628B, BPFP=1.9397 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,000B, BPFP=1.3204 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,616B, BPFP=2.1585 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,916B, BPFP=1.5312 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,440B, BPFP=2.2491 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,624B, BPFP=1.6092 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,424B, BPFP=2.2474 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,104B, BPFP=1.3319 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 20,484B, BPFP=2.2540 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,552B, BPFP=1.4456 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 16.12062610 - layer.0.v_cache 0.00000027 0.00026057 - layer.1.k_cache 0.00355294 1.24668100 - layer.1.v_cache 0.00000089 0.00096981 - layer.2.k_cache 0.00117088 0.51789791 - layer.2.v_cache 0.00000116 0.00134287 - layer.3.k_cache 0.00139091 0.58191101 - layer.3.v_cache 0.00000218 0.00219024 - layer.4.k_cache 0.00325425 1.20565205 - layer.4.v_cache 0.00000306 0.00354895 - layer.4.output 0.00022008 0.10172186 - ------------------------------------------------------------------------------------- - TOTAL 0.00263517 1.43485485 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 207524 -BPFP 1.6311 bits/point -EBPFP 3.2621 equivalent bits/point -MSE 1.434855 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.006s, Pack+Encode: 0.200s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.4349 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,672B, BPFP=0.3930 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,548B, BPFP=1.9850 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,004B, BPFP=1.2847 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 20,596B, BPFP=2.2042 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,972B, BPFP=1.4953 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,332B, BPFP=2.2830 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,712B, BPFP=1.5745 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,868B, BPFP=2.2333 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,080B, BPFP=1.2928 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,748B, BPFP=2.3275 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 54,148B, BPFP=1.4487 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.279s - -[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, 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 16.59505431 - layer.0.v_cache 0.00000028 0.00027328 - layer.1.k_cache 0.00347943 1.24088110 - layer.1.v_cache 0.00000079 0.00095781 - layer.2.k_cache 0.00114165 0.52487935 - layer.2.v_cache 0.00000118 0.00136461 - layer.3.k_cache 0.00136377 0.61907478 - layer.3.v_cache 0.00000212 0.00221245 - layer.4.k_cache 0.00321585 1.15599457 - layer.4.v_cache 0.00000299 0.00361665 - layer.4.output 0.00023923 0.10709144 - ------------------------------------------------------------------------------------- - TOTAL 0.00247661 1.46947676 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 213680 -BPFP 1.6334 bits/point -EBPFP 3.2669 equivalent bits/point -MSE 1.469477 ----------------------- -------------------------------------------------------- -Time: 0.485s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -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 1.4695 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 5,876B, BPFP=0.4099 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,088B, BPFP=1.8198 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,512B, BPFP=1.1518 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,844B, BPFP=1.9422 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,196B, BPFP=1.3390 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,052B, BPFP=1.9568 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,924B, BPFP=1.3898 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,408B, BPFP=1.9118 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,716B, BPFP=1.1660 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,056B, BPFP=1.9570 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 82,120B, BPFP=1.4321 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 11.65647997 - layer.0.v_cache 0.00000027 0.00026265 - layer.1.k_cache 0.00329348 0.94913462 - layer.1.v_cache 0.00000088 0.00101169 - layer.2.k_cache 0.00114370 0.47562844 - layer.2.v_cache 0.00000135 0.00133282 - layer.3.k_cache 0.00133332 0.54026093 - layer.3.v_cache 0.00000257 0.00233327 - layer.4.k_cache 0.00339413 0.97893735 - layer.4.v_cache 0.00000321 0.00358396 - layer.4.output 0.00019581 0.08341160 - ------------------------------------------------------------------------------------- - TOTAL 0.00274709 1.06732944 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 297792 -BPFP 1.4837 bits/point -EBPFP 2.9675 equivalent bits/point -MSE 1.067329 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.0673 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,680B, BPFP=0.4228 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,308B, BPFP=1.8736 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 11,196B, BPFP=1.2863 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,264B, BPFP=2.2132 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,300B, BPFP=1.5280 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,028B, BPFP=2.3010 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 13,936B, BPFP=1.6011 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 19,680B, BPFP=2.2610 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,216B, BPFP=1.2886 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 19,596B, BPFP=2.2514 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 51,728B, BPFP=1.4858 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.96449908 - layer.0.v_cache 0.00000029 0.00026048 - layer.1.k_cache 0.00369932 1.34318049 - layer.1.v_cache 0.00000081 0.00091994 - layer.2.k_cache 0.00117793 0.52493477 - layer.2.v_cache 0.00000112 0.00129863 - layer.3.k_cache 0.00140028 0.61859507 - layer.3.v_cache 0.00000224 0.00216167 - layer.4.k_cache 0.00315349 1.15678720 - layer.4.v_cache 0.00000301 0.00353088 - layer.4.output 0.00021496 0.10303315 - ------------------------------------------------------------------------------------- - TOTAL 0.00267943 1.43059291 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 199932 -BPFP 1.6407 bits/point -EBPFP 3.2814 equivalent bits/point -MSE 1.430593 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.200s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -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 1.4306 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 6,628B, BPFP=0.3807 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,144B, BPFP=1.8465 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 21,104B, BPFP=1.2123 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 34,752B, BPFP=1.9963 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,600B, BPFP=1.4131 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 35,736B, BPFP=2.0528 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 25,708B, BPFP=1.4768 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,192B, BPFP=2.0216 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,608B, BPFP=1.2413 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 36,040B, BPFP=2.0703 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 101,168B, BPFP=1.4529 -⌛️ [2/4] FRONTEND: Frontend time: 0.247s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.382s - -[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, 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 13.12691902 - layer.0.v_cache 0.00000027 0.00025299 - layer.1.k_cache 0.00321745 1.00510979 - layer.1.v_cache 0.00000084 0.00087074 - layer.2.k_cache 0.00114365 0.46438845 - layer.2.v_cache 0.00000128 0.00123593 - layer.3.k_cache 0.00137918 0.53130038 - layer.3.v_cache 0.00000219 0.00204511 - layer.4.k_cache 0.00340721 1.02665531 - layer.4.v_cache 0.00000338 0.00348946 - layer.4.output 0.00019585 0.08812297 - ------------------------------------------------------------------------------------- - TOTAL 0.00265666 1.17962565 - (elements=1,949,696) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1949696 -Total Bytes 374680 -BPFP 1.5374 bits/point -EBPFP 3.0748 equivalent bits/point -MSE 1.179626 ----------------------- -------------------------------------------------------- -Time: 0.637s Load: 0.008s, Pack+Encode: 0.247s, Decode+Unpack: 0.382s ----------------------- -------------------------------------------------------- -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 1.1796 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,632B, BPFP=0.4235 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 16,468B, BPFP=1.9202 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 11,536B, BPFP=1.3451 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 18,276B, BPFP=2.1311 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,672B, BPFP=1.5942 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 19,812B, BPFP=2.3102 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,260B, BPFP=1.6628 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 19,624B, BPFP=2.2882 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,332B, BPFP=1.3214 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 19,336B, BPFP=2.2547 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,760B, BPFP=1.5380 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.279s - -[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, 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 17.69990562 - layer.0.v_cache 0.00000028 0.00026651 - layer.1.k_cache 0.00368886 1.26521734 - layer.1.v_cache 0.00000112 0.00097680 - layer.2.k_cache 0.00116499 0.54201656 - layer.2.v_cache 0.00000120 0.00138241 - layer.3.k_cache 0.00139456 0.63246901 - layer.3.v_cache 0.00000232 0.00232950 - layer.4.k_cache 0.00321227 1.13715283 - layer.4.v_cache 0.00000307 0.00372231 - layer.4.output 0.00021890 0.10519653 - ------------------------------------------------------------------------------------- - TOTAL 0.00271674 1.55044464 - (elements=960,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 960512 -Total Bytes 200708 -BPFP 1.6717 bits/point -EBPFP 3.3434 equivalent bits/point -MSE 1.550445 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.004s, Pack+Encode: 0.202s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -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 1.5504 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 6,600B, BPFP=0.4192 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,396B, BPFP=1.6766 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,676B, BPFP=1.1227 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,656B, BPFP=1.7566 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,300B, BPFP=1.2894 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,364B, BPFP=1.8016 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 21,252B, BPFP=1.3498 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 28,108B, BPFP=1.7853 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 18,144B, BPFP=1.1524 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,548B, BPFP=1.8133 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 82,588B, BPFP=1.3114 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.287s - -[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, 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 13.46366572 - layer.0.v_cache 0.00000027 0.00024289 - layer.1.k_cache 0.00336899 0.94576605 - layer.1.v_cache 0.00000079 0.00084018 - layer.2.k_cache 0.00113470 0.46226374 - layer.2.v_cache 0.00000113 0.00120765 - layer.3.k_cache 0.00144113 0.54558414 - layer.3.v_cache 0.00000264 0.00212389 - layer.4.k_cache 0.00356668 1.07525473 - layer.4.v_cache 0.00000295 0.00333586 - layer.4.output 0.00019508 0.08160973 - ------------------------------------------------------------------------------------- - TOTAL 0.00259466 1.20190884 - (elements=1,763,328) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1763328 -Total Bytes 305632 -BPFP 1.3866 bits/point -EBPFP 2.7732 equivalent bits/point -MSE 1.201909 ----------------------- -------------------------------------------------------- -Time: 0.498s Load: 0.008s, Pack+Encode: 0.203s, Decode+Unpack: 0.287s ----------------------- -------------------------------------------------------- -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 1.2019 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.011s - ------------------------------------------------------------- -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: 7,456B, BPFP=0.3710 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 34,668B, BPFP=1.7251 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 22,556B, BPFP=1.1224 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 37,656B, BPFP=1.8738 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 26,396B, BPFP=1.3135 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 38,364B, BPFP=1.9090 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 27,448B, BPFP=1.3658 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 37,696B, BPFP=1.8758 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 23,152B, BPFP=1.1521 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 38,688B, BPFP=1.9252 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 105,252B, BPFP=1.3094 -⌛️ [2/4] FRONTEND: Frontend time: 0.266s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.397s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.91710493 - layer.0.v_cache 0.00000027 0.00025205 - layer.1.k_cache 0.00309445 0.89000114 - layer.1.v_cache 0.00000088 0.00087203 - layer.2.k_cache 0.00117908 0.46142180 - layer.2.v_cache 0.00000110 0.00122062 - layer.3.k_cache 0.00138553 0.52383010 - layer.3.v_cache 0.00000215 0.00210877 - layer.4.k_cache 0.00349636 0.98346317 - layer.4.v_cache 0.00000310 0.00339369 - layer.4.output 0.00020206 0.07117409 - ------------------------------------------------------------------------------------- - TOTAL 0.00244317 1.07631176 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 399332 -BPFP 1.4194 bits/point -EBPFP 2.8387 equivalent bits/point -MSE 1.076312 ----------------------- -------------------------------------------------------- -Time: 0.673s Load: 0.011s, Pack+Encode: 0.266s, Decode+Unpack: 0.397s ----------------------- -------------------------------------------------------- -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 1.0763 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 3,832B, BPFP=0.3992 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 19,384B, BPFP=2.0192 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,652B, BPFP=1.3179 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 21,336B, BPFP=2.2225 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,284B, BPFP=1.4879 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,672B, BPFP=2.2575 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,084B, BPFP=1.5713 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,024B, BPFP=2.1900 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,472B, BPFP=1.2992 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,796B, BPFP=2.2704 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 57,776B, BPFP=1.5046 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.295s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.88327474 - layer.0.v_cache 0.00000028 0.00026633 - layer.1.k_cache 0.00341811 1.15208740 - layer.1.v_cache 0.00000080 0.00090928 - layer.2.k_cache 0.00110890 0.50646657 - layer.2.v_cache 0.00000112 0.00123114 - layer.3.k_cache 0.00134674 0.56674932 - layer.3.v_cache 0.00000209 0.00202667 - layer.4.k_cache 0.00330252 1.14954010 - layer.4.v_cache 0.00000296 0.00344638 - layer.4.output 0.00020017 0.09816450 - ------------------------------------------------------------------------------------- - TOTAL 0.00249143 1.40418971 - (elements=1,075,200) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1075200 -Total Bytes 221312 -BPFP 1.6467 bits/point -EBPFP 3.2933 equivalent bits/point -MSE 1.404190 ----------------------- -------------------------------------------------------- -Time: 0.510s Load: 0.005s, Pack+Encode: 0.209s, Decode+Unpack: 0.295s ----------------------- -------------------------------------------------------- -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 1.4042 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 6,116B, BPFP=0.3916 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,408B, BPFP=1.6911 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,548B, BPFP=1.1237 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,320B, BPFP=1.8135 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,408B, BPFP=1.3069 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,600B, BPFP=1.8315 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,916B, BPFP=1.3394 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 28,116B, BPFP=1.8005 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,716B, BPFP=1.1345 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,728B, BPFP=1.8397 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 83,740B, BPFP=1.3406 -⌛️ [2/4] FRONTEND: Frontend time: 0.210s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.313s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.78521929 - layer.0.v_cache 0.00000028 0.00025176 - layer.1.k_cache 0.00316206 0.88178728 - layer.1.v_cache 0.00000097 0.00089805 - layer.2.k_cache 0.00113900 0.44976663 - layer.2.v_cache 0.00000117 0.00124511 - layer.3.k_cache 0.00136642 0.52263804 - layer.3.v_cache 0.00000221 0.00203757 - layer.4.k_cache 0.00339647 0.95754461 - layer.4.v_cache 0.00000316 0.00332600 - layer.4.output 0.00022994 0.07709056 - ------------------------------------------------------------------------------------- - TOTAL 0.00251773 1.13664833 - (elements=1,748,992) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1748992 -Total Bytes 306616 -BPFP 1.4025 bits/point -EBPFP 2.8050 equivalent bits/point -MSE 1.136648 ----------------------- -------------------------------------------------------- -Time: 0.532s Load: 0.008s, Pack+Encode: 0.210s, Decode+Unpack: 0.313s ----------------------- -------------------------------------------------------- -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 1.1366 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 6,708B, BPFP=0.3691 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 32,916B, BPFP=1.8110 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 20,908B, BPFP=1.1503 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 35,728B, BPFP=1.9657 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 24,356B, BPFP=1.3400 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 36,572B, BPFP=2.0121 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 25,748B, BPFP=1.4166 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 35,760B, BPFP=1.9674 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 21,712B, BPFP=1.1945 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 37,140B, BPFP=2.0434 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 104,704B, BPFP=1.4401 -⌛️ [2/4] FRONTEND: Frontend time: 0.255s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.401s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.55271976 - layer.0.v_cache 0.00000029 0.00025214 - layer.1.k_cache 0.00315333 0.95953616 - layer.1.v_cache 0.00000086 0.00084655 - layer.2.k_cache 0.00112591 0.44927624 - layer.2.v_cache 0.00000130 0.00119913 - layer.3.k_cache 0.00130425 0.51902647 - layer.3.v_cache 0.00000237 0.00216857 - layer.4.k_cache 0.00348660 1.00569432 - layer.4.v_cache 0.00000314 0.00332538 - layer.4.output 0.00018111 0.08275463 - ------------------------------------------------------------------------------------- - TOTAL 0.00234429 1.05893309 - (elements=2,035,712) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2035712 -Total Bytes 382252 -BPFP 1.5022 bits/point -EBPFP 3.0044 equivalent bits/point -MSE 1.058933 ----------------------- -------------------------------------------------------- -Time: 0.665s Load: 0.009s, Pack+Encode: 0.255s, Decode+Unpack: 0.401s ----------------------- -------------------------------------------------------- -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 1.0589 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.004s - ------------------------------------------------------------- -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: 3,636B, BPFP=0.3839 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,740B, BPFP=1.9785 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,216B, BPFP=1.2897 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 20,900B, BPFP=2.2065 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 14,352B, BPFP=1.5152 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 21,732B, BPFP=2.2943 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 15,008B, BPFP=1.5845 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 21,392B, BPFP=2.2584 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 12,212B, BPFP=1.2893 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 22,364B, BPFP=2.3611 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 52,212B, BPFP=1.3781 -⌛️ [2/4] FRONTEND: Frontend time: 0.207s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.301s - -[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, 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 17.05666042 - layer.0.v_cache 0.00000028 0.00026303 - layer.1.k_cache 0.00343224 1.23070990 - layer.1.v_cache 0.00000079 0.00089821 - layer.2.k_cache 0.00111603 0.50685063 - layer.2.v_cache 0.00000116 0.00126877 - layer.3.k_cache 0.00133047 0.57315435 - layer.3.v_cache 0.00000218 0.00205307 - layer.4.k_cache 0.00334566 1.12497041 - layer.4.v_cache 0.00000301 0.00333708 - layer.4.output 0.00024685 0.08633784 - ------------------------------------------------------------------------------------- - TOTAL 0.00255562 1.48896552 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 214764 -BPFP 1.6195 bits/point -EBPFP 3.2391 equivalent bits/point -MSE 1.488966 ----------------------- -------------------------------------------------------- -Time: 0.512s Load: 0.004s, Pack+Encode: 0.207s, Decode+Unpack: 0.301s ----------------------- -------------------------------------------------------- -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 1.4890 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,716B, BPFP=0.4759 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 13,700B, BPFP=1.7546 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 9,848B, BPFP=1.2613 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 14,592B, BPFP=1.8689 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 10,892B, BPFP=1.3950 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 14,648B, BPFP=1.8760 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 11,396B, BPFP=1.4595 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 14,384B, BPFP=1.8422 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 9,944B, BPFP=1.2736 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 14,728B, BPFP=1.8863 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 45,064B, BPFP=1.4429 -⌛️ [2/4] FRONTEND: Frontend time: 0.157s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.197s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 14.59533992 - layer.0.v_cache 0.00000026 0.00023622 - layer.1.k_cache 0.00385674 0.98219925 - layer.1.v_cache 0.00000079 0.00089904 - layer.2.k_cache 0.00117960 0.48355672 - layer.2.v_cache 0.00000113 0.00132020 - layer.3.k_cache 0.00144752 0.55049584 - layer.3.v_cache 0.00000212 0.00205902 - layer.4.k_cache 0.00316257 1.05016690 - layer.4.v_cache 0.00000306 0.00344120 - layer.4.output 0.00025867 0.11250719 - ------------------------------------------------------------------------------------- - TOTAL 0.00273183 1.29426736 - (elements=874,496) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 874496 -Total Bytes 162912 -BPFP 1.4903 bits/point -EBPFP 2.9807 equivalent bits/point -MSE 1.294267 ----------------------- -------------------------------------------------------- -Time: 0.359s Load: 0.005s, Pack+Encode: 0.157s, Decode+Unpack: 0.197s ----------------------- -------------------------------------------------------- -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 1.2943 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,392B, BPFP=0.4492 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 13,832B, BPFP=1.8316 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 9,460B, BPFP=1.2526 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 14,824B, BPFP=1.9629 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 10,632B, BPFP=1.4078 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 15,064B, BPFP=1.9947 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 11,224B, BPFP=1.4862 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 14,700B, BPFP=1.9465 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 9,472B, BPFP=1.2542 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 14,996B, BPFP=1.9857 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 43,220B, BPFP=1.4307 -⌛️ [2/4] FRONTEND: Frontend time: 0.157s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.202s - -[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, 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 16.17146534 - layer.0.v_cache 0.00000029 0.00026142 - layer.1.k_cache 0.00372036 1.12905987 - layer.1.v_cache 0.00000085 0.00100505 - layer.2.k_cache 0.00115281 0.52953762 - layer.2.v_cache 0.00000145 0.00145363 - layer.3.k_cache 0.00140016 0.61169524 - layer.3.v_cache 0.00000257 0.00248972 - layer.4.k_cache 0.00323529 1.17528443 - layer.4.v_cache 0.00000321 0.00383746 - layer.4.output 0.00024217 0.11854901 - ------------------------------------------------------------------------------------- - TOTAL 0.00296341 1.43573470 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 160816 -BPFP 1.5210 bits/point -EBPFP 3.0421 equivalent bits/point -MSE 1.435735 ----------------------- -------------------------------------------------------- -Time: 0.363s Load: 0.005s, Pack+Encode: 0.157s, Decode+Unpack: 0.202s ----------------------- -------------------------------------------------------- -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 1.4357 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.006s - ------------------------------------------------------------- -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: 3,420B, BPFP=0.4048 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 15,484B, BPFP=1.8329 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 10,980B, BPFP=1.2997 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 17,068B, BPFP=2.0204 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 12,904B, BPFP=1.5275 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 18,056B, BPFP=2.1373 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 13,492B, BPFP=1.5971 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 18,064B, BPFP=2.1383 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 10,800B, BPFP=1.2784 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 18,188B, BPFP=2.1529 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 43,788B, BPFP=1.2958 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.291s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.12290816 - layer.0.v_cache 0.00000028 0.00025646 - layer.1.k_cache 0.00365239 1.17011712 - layer.1.v_cache 0.00000074 0.00087910 - layer.2.k_cache 0.00118996 0.53221720 - layer.2.v_cache 0.00000101 0.00118773 - layer.3.k_cache 0.00141080 0.58369897 - layer.3.v_cache 0.00000191 0.00189559 - layer.4.k_cache 0.00329622 1.15634664 - layer.4.v_cache 0.00000294 0.00325711 - layer.4.output 0.00019151 0.09961808 - ------------------------------------------------------------------------------------- - TOTAL 0.00274804 1.35508831 - (elements=946,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 946176 -Total Bytes 182244 -BPFP 1.5409 bits/point -EBPFP 3.0818 equivalent bits/point -MSE 1.355088 ----------------------- -------------------------------------------------------- -Time: 0.506s Load: 0.006s, Pack+Encode: 0.209s, Decode+Unpack: 0.291s ----------------------- -------------------------------------------------------- -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 1.3551 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,588B, BPFP=0.3840 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 18,260B, BPFP=1.9542 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 12,016B, BPFP=1.2860 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 19,908B, BPFP=2.1306 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 13,816B, BPFP=1.4786 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 20,816B, BPFP=2.2277 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 14,784B, BPFP=1.5822 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 20,200B, BPFP=2.1618 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 11,996B, BPFP=1.2838 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 21,176B, BPFP=2.2663 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 53,752B, BPFP=1.4381 -⌛️ [2/4] FRONTEND: Frontend time: 0.207s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.298s - -[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, 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 15.33009150 - layer.0.v_cache 0.00000028 0.00025661 - layer.1.k_cache 0.00355597 1.22853726 - layer.1.v_cache 0.00000080 0.00087540 - layer.2.k_cache 0.00115478 0.51851184 - layer.2.v_cache 0.00000105 0.00121605 - layer.3.k_cache 0.00135922 0.59718480 - layer.3.v_cache 0.00000200 0.00197747 - layer.4.k_cache 0.00333109 1.11736047 - layer.4.v_cache 0.00000279 0.00319488 - layer.4.output 0.00023199 0.08553246 - ------------------------------------------------------------------------------------- - TOTAL 0.00251192 1.36723829 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 210312 -BPFP 1.6077 bits/point -EBPFP 3.2154 equivalent bits/point -MSE 1.367238 ----------------------- -------------------------------------------------------- -Time: 0.510s Load: 0.005s, Pack+Encode: 0.207s, Decode+Unpack: 0.298s ----------------------- -------------------------------------------------------- -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 1.3672 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 6,544B, BPFP=0.4225 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 26,412B, BPFP=1.7053 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 17,464B, BPFP=1.1276 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 28,092B, BPFP=1.8138 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 20,208B, BPFP=1.3048 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 28,372B, BPFP=1.8319 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,960B, BPFP=1.3533 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,948B, BPFP=1.8045 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,536B, BPFP=1.1322 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,732B, BPFP=1.8551 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 81,296B, BPFP=1.3122 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.41648070 - layer.0.v_cache 0.00000028 0.00024276 - layer.1.k_cache 0.00332455 0.84494952 - layer.1.v_cache 0.00000080 0.00084197 - layer.2.k_cache 0.00116246 0.46119580 - layer.2.v_cache 0.00000109 0.00114329 - layer.3.k_cache 0.00134746 0.52299008 - layer.3.v_cache 0.00000209 0.00192966 - layer.4.k_cache 0.00345892 0.96063390 - layer.4.v_cache 0.00000309 0.00334860 - layer.4.output 0.00018371 0.07507073 - ------------------------------------------------------------------------------------- - TOTAL 0.00249913 1.10814566 - (elements=1,734,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1734656 -Total Bytes 303564 -BPFP 1.4000 bits/point -EBPFP 2.8000 equivalent bits/point -MSE 1.108146 ----------------------- -------------------------------------------------------- -Time: 0.499s Load: 0.008s, Pack+Encode: 0.205s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.1081 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.009s - ------------------------------------------------------------- -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: 5,572B, BPFP=0.3852 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,628B, BPFP=1.7718 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,748B, BPFP=1.1579 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,396B, BPFP=1.8941 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,276B, BPFP=1.3327 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,540B, BPFP=1.9040 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,176B, BPFP=1.3949 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,208B, BPFP=1.8811 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,024B, BPFP=1.1770 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,804B, BPFP=1.9223 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 75,556B, BPFP=1.3059 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.27341157 - layer.0.v_cache 0.00000029 0.00024674 - layer.1.k_cache 0.00330439 0.93620840 - layer.1.v_cache 0.00000076 0.00081802 - layer.2.k_cache 0.00114562 0.44909594 - layer.2.v_cache 0.00000109 0.00112938 - layer.3.k_cache 0.00136171 0.51201346 - layer.3.v_cache 0.00000227 0.00201366 - layer.4.k_cache 0.00342222 0.93851802 - layer.4.v_cache 0.00000313 0.00326266 - layer.4.output 0.00018801 0.06404027 - ------------------------------------------------------------------------------------- - TOTAL 0.00252156 1.09806278 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 289928 -BPFP 1.4318 bits/point -EBPFP 2.8635 equivalent bits/point -MSE 1.098063 ----------------------- -------------------------------------------------------- -Time: 0.500s Load: 0.009s, Pack+Encode: 0.206s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.0981 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.008s - ------------------------------------------------------------- -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: 5,120B, BPFP=0.3636 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,096B, BPFP=1.7824 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,532B, BPFP=1.1741 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,064B, BPFP=1.9222 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,116B, BPFP=1.3577 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,404B, BPFP=1.9463 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 19,788B, BPFP=1.4054 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 26,792B, BPFP=1.9028 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,636B, BPFP=1.1815 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 27,520B, BPFP=1.9545 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 76,008B, BPFP=1.3496 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.299s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.71618874 - layer.0.v_cache 0.00000028 0.00025292 - layer.1.k_cache 0.00323236 0.98154318 - layer.1.v_cache 0.00000082 0.00083594 - layer.2.k_cache 0.00114874 0.49102291 - layer.2.v_cache 0.00000131 0.00124337 - layer.3.k_cache 0.00138397 0.53150409 - layer.3.v_cache 0.00000212 0.00199851 - layer.4.k_cache 0.00346958 1.01345367 - layer.4.v_cache 0.00000311 0.00336645 - layer.4.output 0.00021756 0.09122191 - ------------------------------------------------------------------------------------- - TOTAL 0.00254230 1.07902125 - (elements=1,576,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1576960 -Total Bytes 287076 -BPFP 1.4564 bits/point -EBPFP 2.9127 equivalent bits/point -MSE 1.079021 ----------------------- -------------------------------------------------------- -Time: 0.510s Load: 0.008s, Pack+Encode: 0.203s, Decode+Unpack: 0.299s ----------------------- -------------------------------------------------------- -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 1.0790 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,404B, BPFP=0.4091 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 14,988B, BPFP=1.8014 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 10,660B, BPFP=1.2812 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 16,836B, BPFP=2.0236 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 12,380B, BPFP=1.4880 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 17,672B, BPFP=2.1240 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 12,916B, BPFP=1.5524 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 17,792B, BPFP=2.1385 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 10,352B, BPFP=1.2442 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 17,532B, BPFP=2.1072 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 45,420B, BPFP=1.3648 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.281s - -[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, 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 14.10786790 - layer.0.v_cache 0.00000028 0.00026909 - layer.1.k_cache 0.00351263 1.15987244 - layer.1.v_cache 0.00000086 0.00088949 - layer.2.k_cache 0.00113140 0.48794990 - layer.2.v_cache 0.00000103 0.00118126 - layer.3.k_cache 0.00134002 0.55196416 - layer.3.v_cache 0.00000212 0.00204110 - layer.4.k_cache 0.00331905 1.04197857 - layer.4.v_cache 0.00000293 0.00333477 - layer.4.output 0.00019456 0.09603648 - ------------------------------------------------------------------------------------- - TOTAL 0.00258730 1.26724961 - (elements=931,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 931840 -Total Bytes 179952 -BPFP 1.5449 bits/point -EBPFP 3.0898 equivalent bits/point -MSE 1.267250 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.005s, Pack+Encode: 0.208s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -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 1.2672 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 5,588B, BPFP=0.3763 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,864B, BPFP=1.7419 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,788B, BPFP=1.1307 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,772B, BPFP=1.8704 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,548B, BPFP=1.3165 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,992B, BPFP=1.8852 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,304B, BPFP=1.3675 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,524B, BPFP=1.8537 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 17,064B, BPFP=1.1492 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,376B, BPFP=1.9111 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 79,260B, BPFP=1.3345 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.69047283 - layer.0.v_cache 0.00000027 0.00024768 - layer.1.k_cache 0.00326550 0.90478371 - layer.1.v_cache 0.00000084 0.00087007 - layer.2.k_cache 0.00114506 0.46615048 - layer.2.v_cache 0.00000107 0.00119421 - layer.3.k_cache 0.00140586 0.53742553 - layer.3.v_cache 0.00000237 0.00206793 - layer.4.k_cache 0.00329219 0.99776827 - layer.4.v_cache 0.00000320 0.00342412 - layer.4.output 0.00021339 0.08000063 - ------------------------------------------------------------------------------------- - TOTAL 0.00249392 1.13745767 - (elements=1,662,976) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1662976 -Total Bytes 296080 -BPFP 1.4243 bits/point -EBPFP 2.8487 equivalent bits/point -MSE 1.137458 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -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 1.1375 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,312B, BPFP=0.4386 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 13,656B, BPFP=1.8083 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 9,204B, BPFP=1.2188 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 14,576B, BPFP=1.9301 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 10,548B, BPFP=1.3967 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 14,676B, BPFP=1.9433 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 10,852B, BPFP=1.4370 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 14,284B, BPFP=1.8914 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 9,288B, BPFP=1.2299 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 14,656B, BPFP=1.9407 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 41,808B, BPFP=1.3840 -⌛️ [2/4] FRONTEND: Frontend time: 0.150s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.196s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.40183374 - layer.0.v_cache 0.00000028 0.00026559 - layer.1.k_cache 0.00373147 1.19866788 - layer.1.v_cache 0.00000088 0.00089805 - layer.2.k_cache 0.00131257 0.50821414 - layer.2.v_cache 0.00000108 0.00124478 - layer.3.k_cache 0.00142120 0.59061781 - layer.3.v_cache 0.00000202 0.00196581 - layer.4.k_cache 0.00321287 1.19930578 - layer.4.v_cache 0.00000283 0.00319597 - layer.4.output 0.00027608 0.08605302 - ------------------------------------------------------------------------------------- - TOTAL 0.00274440 1.37503012 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 156860 -BPFP 1.4836 bits/point -EBPFP 2.9672 equivalent bits/point -MSE 1.375030 ----------------------- -------------------------------------------------------- -Time: 0.351s Load: 0.005s, Pack+Encode: 0.150s, Decode+Unpack: 0.196s ----------------------- -------------------------------------------------------- -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 1.3750 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.005s - ------------------------------------------------------------- -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: 3,252B, BPFP=0.4234 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 13,756B, BPFP=1.7911 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 9,452B, BPFP=1.2307 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 14,728B, BPFP=1.9177 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 10,768B, BPFP=1.4021 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 14,848B, BPFP=1.9333 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 11,144B, BPFP=1.4510 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 14,604B, BPFP=1.9016 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 9,468B, BPFP=1.2328 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 14,896B, BPFP=1.9396 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 44,368B, BPFP=1.4443 -⌛️ [2/4] FRONTEND: Frontend time: 0.152s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.264s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 15.20754395 - layer.0.v_cache 0.00000027 0.00026144 - layer.1.k_cache 0.00360089 1.09614461 - layer.1.v_cache 0.00000087 0.00091086 - layer.2.k_cache 0.00116909 0.50144370 - layer.2.v_cache 0.00000119 0.00135269 - layer.3.k_cache 0.00139808 0.57191124 - layer.3.v_cache 0.00000212 0.00207593 - layer.4.k_cache 0.00324009 1.04962145 - layer.4.v_cache 0.00000310 0.00339358 - layer.4.output 0.00020032 0.08426380 - ------------------------------------------------------------------------------------- - TOTAL 0.00276144 1.34083676 - (elements=860,160) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 860160 -Total Bytes 161284 -BPFP 1.5000 bits/point -EBPFP 3.0001 equivalent bits/point -MSE 1.340837 ----------------------- -------------------------------------------------------- -Time: 0.420s Load: 0.005s, Pack+Encode: 0.152s, Decode+Unpack: 0.264s ----------------------- -------------------------------------------------------- -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 1.3408 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.007s - ------------------------------------------------------------- -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: 5,628B, BPFP=0.3857 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 25,724B, BPFP=1.7629 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 16,796B, BPFP=1.1510 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 27,516B, BPFP=1.8857 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 19,504B, BPFP=1.3366 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 27,876B, BPFP=1.9104 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 20,400B, BPFP=1.3980 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 27,292B, BPFP=1.8703 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 16,928B, BPFP=1.1601 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 28,096B, BPFP=1.9254 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 76,704B, BPFP=1.3141 -⌛️ [2/4] FRONTEND: Frontend time: 0.212s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.294s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 12.34733501 - layer.0.v_cache 0.00000028 0.00024172 - layer.1.k_cache 0.00337316 0.85463534 - layer.1.v_cache 0.00000076 0.00085001 - layer.2.k_cache 0.00114524 0.45776250 - layer.2.v_cache 0.00000110 0.00125496 - layer.3.k_cache 0.00139029 0.53189177 - layer.3.v_cache 0.00000222 0.00199769 - layer.4.k_cache 0.00339712 0.99385017 - layer.4.v_cache 0.00000302 0.00340761 - layer.4.output 0.00021100 0.09077687 - ------------------------------------------------------------------------------------- - TOTAL 0.00246523 1.11116673 - (elements=1,634,304) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1634304 -Total Bytes 292464 -BPFP 1.4316 bits/point -EBPFP 2.8633 equivalent bits/point -MSE 1.111167 ----------------------- -------------------------------------------------------- -Time: 0.512s Load: 0.007s, Pack+Encode: 0.212s, Decode+Unpack: 0.294s ----------------------- -------------------------------------------------------- -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 1.1112 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 1.5070 bits/point -Avg EBPFP 3.0140 equivalent bits/point -Avg MSE 1.262504 -Avg Time 0.510s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:862c9bf9c6bacef39e08101facd48ba4da8703c56584b44ee93c217fa0881a86 +size 1119066 diff --git a/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log b/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log index 7d5786bad60c344716a71a9a22ab88700cd2e0a8..98dabdd1e3a80f66a93fbf9039fd98d1476e5157 100644 --- a/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log +++ b/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_hyperprior-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: hyperprior-featurecoding - handler: qwen - checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_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/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 599 -Loaded hyperprior-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 hyperprior-featurecoding -Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_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/lambda0.007/hyperprior-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: 0.026s - ------------------------------------------------------------- -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: 16,688B, BPFP=0.3514 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 76,964B, BPFP=1.6207 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,640B, BPFP=1.0453 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,732B, BPFP=1.7422 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,980B, BPFP=1.2420 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 84,216B, BPFP=1.7734 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,868B, BPFP=1.2818 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 82,764B, BPFP=1.7428 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 51,104B, BPFP=1.0761 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,848B, BPFP=1.7867 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 230,112B, BPFP=1.2114 -⌛️ [2/4] FRONTEND: Frontend time: 0.796s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.761s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.17497052 - layer.0.v_cache 0.00000026 0.00023545 - layer.1.k_cache 0.00289382 0.73259675 - layer.1.v_cache 0.00000075 0.00085051 - layer.2.k_cache 0.00115144 0.44338138 - layer.2.v_cache 0.00000114 0.00126010 - layer.3.k_cache 0.00133317 0.49246356 - layer.3.v_cache 0.00000213 0.00204330 - layer.4.k_cache 0.00354181 0.85875856 - layer.4.v_cache 0.00000324 0.00345658 - layer.4.output 0.00015639 0.05641882 - ------------------------------------------------------------------------------------- - TOTAL 0.00238421 0.92397800 - (elements=5,318,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5318656 -Total Bytes 878916 -BPFP 1.3220 bits/point -EBPFP 2.6440 equivalent bits/point -MSE 0.923978 ----------------------- -------------------------------------------------------- -Time: 1.583s Load: 0.026s, Pack+Encode: 0.796s, Decode+Unpack: 0.761s ----------------------- -------------------------------------------------------- -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 0.9240 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,676B, BPFP=0.3560 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,288B, BPFP=1.6644 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,904B, BPFP=1.0652 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,300B, BPFP=1.8010 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,600B, BPFP=1.2627 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,896B, BPFP=1.8372 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,248B, BPFP=1.3001 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,572B, BPFP=1.8071 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,980B, BPFP=1.0897 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,364B, BPFP=1.8478 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 220,744B, BPFP=1.2533 -⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.719s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.43816642 - layer.0.v_cache 0.00000026 0.00023880 - layer.1.k_cache 0.00287302 0.73506927 - layer.1.v_cache 0.00000078 0.00087477 - layer.2.k_cache 0.00123767 0.44840347 - layer.2.v_cache 0.00000117 0.00133420 - layer.3.k_cache 0.00130179 0.49219548 - layer.3.v_cache 0.00000226 0.00214920 - layer.4.k_cache 0.00355144 0.85887802 - layer.4.v_cache 0.00000325 0.00361115 - layer.4.output 0.00015063 0.06300889 - ------------------------------------------------------------------------------------- - TOTAL 0.00240496 0.94521117 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 838572 -BPFP 1.3603 bits/point -EBPFP 2.7207 equivalent bits/point -MSE 0.945211 ----------------------- -------------------------------------------------------- -Time: 1.153s Load: 0.018s, Pack+Encode: 0.417s, Decode+Unpack: 0.719s ----------------------- -------------------------------------------------------- -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 0.9452 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.021s - ------------------------------------------------------------- -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: 16,816B, BPFP=0.3485 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 78,328B, BPFP=1.6232 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 50,784B, BPFP=1.0524 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 84,248B, BPFP=1.7459 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 60,300B, BPFP=1.2496 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 85,456B, BPFP=1.7709 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 61,696B, BPFP=1.2785 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 83,556B, BPFP=1.7315 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 52,008B, BPFP=1.0778 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 86,196B, BPFP=1.7862 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 236,948B, BPFP=1.2276 -⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.716s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.11373521 - layer.0.v_cache 0.00000027 0.00024476 - layer.1.k_cache 0.00291640 0.71294292 - layer.1.v_cache 0.00000080 0.00091462 - layer.2.k_cache 0.00116459 0.44773808 - layer.2.v_cache 0.00000114 0.00129397 - layer.3.k_cache 0.00131780 0.49178462 - layer.3.v_cache 0.00000210 0.00204452 - layer.4.k_cache 0.00355012 0.83843444 - layer.4.v_cache 0.00000332 0.00369806 - layer.4.output 0.00014433 0.05460753 - ------------------------------------------------------------------------------------- - TOTAL 0.00239596 0.91651866 - (elements=5,404,672) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5404672 -Total Bytes 896336 -BPFP 1.3268 bits/point -EBPFP 2.6535 equivalent bits/point -MSE 0.916519 ----------------------- -------------------------------------------------------- -Time: 1.171s Load: 0.021s, Pack+Encode: 0.435s, Decode+Unpack: 0.716s ----------------------- -------------------------------------------------------- -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 0.9165 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,400B, BPFP=0.3571 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,496B, BPFP=1.6244 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,340B, BPFP=1.0501 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,140B, BPFP=1.7396 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,272B, BPFP=1.2468 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,084B, BPFP=1.7630 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,544B, BPFP=1.2784 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,940B, BPFP=1.7346 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,776B, BPFP=1.0857 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,796B, BPFP=1.7807 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 199,844B, BPFP=1.2391 -⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.608s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.64104353 - layer.0.v_cache 0.00000027 0.00024380 - layer.1.k_cache 0.00284817 0.72327043 - layer.1.v_cache 0.00000082 0.00089735 - layer.2.k_cache 0.00122538 0.44878012 - layer.2.v_cache 0.00000116 0.00129023 - layer.3.k_cache 0.00131111 0.49461781 - layer.3.v_cache 0.00000216 0.00209881 - layer.4.k_cache 0.00356899 0.86391224 - layer.4.v_cache 0.00000330 0.00364308 - layer.4.output 0.00013845 0.05746214 - ------------------------------------------------------------------------------------- - TOTAL 0.00244312 0.95783185 - (elements=4,515,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4515840 -Total Bytes 750632 -BPFP 1.3298 bits/point -EBPFP 2.6596 equivalent bits/point -MSE 0.957832 ----------------------- -------------------------------------------------------- -Time: 1.041s Load: 0.016s, Pack+Encode: 0.418s, Decode+Unpack: 0.608s ----------------------- -------------------------------------------------------- -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 0.9578 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,484B, BPFP=0.3342 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 75,952B, BPFP=1.6392 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,232B, BPFP=1.0625 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,624B, BPFP=1.7831 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,284B, BPFP=1.2579 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 83,736B, BPFP=1.8071 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,160B, BPFP=1.2983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 82,116B, BPFP=1.7722 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,496B, BPFP=1.0898 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,304B, BPFP=1.8194 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 221,948B, BPFP=1.1975 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.716s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.99066398 - layer.0.v_cache 0.00000027 0.00025068 - layer.1.k_cache 0.00294127 0.74805885 - layer.1.v_cache 0.00000079 0.00091963 - layer.2.k_cache 0.00116280 0.45178117 - layer.2.v_cache 0.00000114 0.00130210 - layer.3.k_cache 0.00131002 0.50138965 - layer.3.v_cache 0.00000215 0.00208492 - layer.4.k_cache 0.00366934 0.88731662 - layer.4.v_cache 0.00000328 0.00366694 - layer.4.output 0.00013411 0.05426723 - ------------------------------------------------------------------------------------- - TOTAL 0.00240096 0.98603596 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 864336 -BPFP 1.3324 bits/point -EBPFP 2.6648 equivalent bits/point -MSE 0.986036 ----------------------- -------------------------------------------------------- -Time: 1.155s Load: 0.019s, Pack+Encode: 0.420s, Decode+Unpack: 0.716s ----------------------- -------------------------------------------------------- -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 0.9860 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,824B, BPFP=0.3447 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,968B, BPFP=1.6734 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,044B, BPFP=1.0706 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,004B, BPFP=1.8137 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,148B, BPFP=1.2590 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,868B, BPFP=1.8570 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,424B, BPFP=1.3119 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,148B, BPFP=1.8171 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,600B, BPFP=1.0835 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,192B, BPFP=1.8646 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 210,168B, BPFP=1.2217 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (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, 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: 0.705s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.93995594 - layer.0.v_cache 0.00000026 0.00024203 - layer.1.k_cache 0.00293366 0.75623553 - layer.1.v_cache 0.00000078 0.00088928 - layer.2.k_cache 0.00117059 0.45450928 - layer.2.v_cache 0.00000113 0.00128891 - layer.3.k_cache 0.00132874 0.50155126 - layer.3.v_cache 0.00000215 0.00212211 - layer.4.k_cache 0.00360252 0.89458157 - layer.4.v_cache 0.00000314 0.00345868 - layer.4.output 0.00014135 0.05866923 - ------------------------------------------------------------------------------------- - TOTAL 0.00244837 0.98496511 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 816388 -BPFP 1.3559 bits/point -EBPFP 2.7117 equivalent bits/point -MSE 0.984965 ----------------------- -------------------------------------------------------- -Time: 1.135s Load: 0.017s, Pack+Encode: 0.413s, Decode+Unpack: 0.705s ----------------------- -------------------------------------------------------- -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 0.9850 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,588B, BPFP=0.3595 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,924B, BPFP=1.6001 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,784B, BPFP=1.0544 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,144B, BPFP=1.7287 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,588B, BPFP=1.2467 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,128B, BPFP=1.7530 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,832B, BPFP=1.2774 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,948B, BPFP=1.7239 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,212B, BPFP=1.0896 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,632B, BPFP=1.7654 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 199,240B, BPFP=1.2276 -⌛️ [2/4] FRONTEND: Frontend time: 0.363s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.599s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.94279947 - layer.0.v_cache 0.00000027 0.00024683 - layer.1.k_cache 0.00294942 0.72272412 - layer.1.v_cache 0.00000081 0.00090900 - layer.2.k_cache 0.00116222 0.44931285 - layer.2.v_cache 0.00000115 0.00131968 - layer.3.k_cache 0.00131165 0.49695185 - layer.3.v_cache 0.00000215 0.00215041 - layer.4.k_cache 0.00349495 0.86920079 - layer.4.v_cache 0.00000332 0.00369378 - layer.4.output 0.00014092 0.06043636 - ------------------------------------------------------------------------------------- - TOTAL 0.00246317 0.90936102 - (elements=4,544,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4544512 -Total Bytes 751020 -BPFP 1.3221 bits/point -EBPFP 2.6441 equivalent bits/point -MSE 0.909361 ----------------------- -------------------------------------------------------- -Time: 0.977s Load: 0.016s, Pack+Encode: 0.363s, Decode+Unpack: 0.599s ----------------------- -------------------------------------------------------- -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 0.9094 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,472B, BPFP=0.3424 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,088B, BPFP=1.6397 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,292B, BPFP=1.0467 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 80,456B, BPFP=1.7806 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,028B, BPFP=1.2400 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,728B, BPFP=1.8088 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,780B, BPFP=1.2788 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,796B, BPFP=1.7660 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,384B, BPFP=1.0708 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,816B, BPFP=1.8107 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,100B, BPFP=1.1791 -⌛️ [2/4] FRONTEND: Frontend time: 0.425s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.701s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.49586206 - layer.0.v_cache 0.00000027 0.00024878 - layer.1.k_cache 0.00288490 0.73713844 - layer.1.v_cache 0.00000079 0.00092643 - layer.2.k_cache 0.00118233 0.45298564 - layer.2.v_cache 0.00000115 0.00133839 - layer.3.k_cache 0.00132883 0.50712741 - layer.3.v_cache 0.00000217 0.00207209 - layer.4.k_cache 0.00368366 0.88970091 - layer.4.v_cache 0.00000317 0.00356003 - layer.4.output 0.00013605 0.05773991 - ------------------------------------------------------------------------------------- - TOTAL 0.00247279 0.95156570 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 835940 -BPFP 1.3215 bits/point -EBPFP 2.6430 equivalent bits/point -MSE 0.951566 ----------------------- -------------------------------------------------------- -Time: 1.143s Load: 0.017s, Pack+Encode: 0.425s, Decode+Unpack: 0.701s ----------------------- -------------------------------------------------------- -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 0.9516 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,588B, BPFP=0.3830 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,640B, BPFP=1.5881 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,804B, BPFP=1.0516 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 68,536B, BPFP=1.6838 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,964B, BPFP=1.2275 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 69,956B, BPFP=1.7187 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,912B, BPFP=1.2754 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,168B, BPFP=1.6993 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,112B, BPFP=1.0837 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 70,940B, BPFP=1.7428 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 203,000B, BPFP=1.2468 -⌛️ [2/4] FRONTEND: Frontend time: 0.372s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.601s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.89669090 - layer.0.v_cache 0.00000025 0.00023215 - layer.1.k_cache 0.00290030 0.71538155 - layer.1.v_cache 0.00000077 0.00082595 - layer.2.k_cache 0.00116467 0.44687216 - layer.2.v_cache 0.00000108 0.00119395 - layer.3.k_cache 0.00132360 0.50727988 - layer.3.v_cache 0.00000220 0.00202605 - layer.4.k_cache 0.00367546 0.88039681 - layer.4.v_cache 0.00000314 0.00347901 - layer.4.output 0.00019099 0.06516751 - ------------------------------------------------------------------------------------- - TOTAL 0.00246062 0.90821775 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 750620 -BPFP 1.3172 bits/point -EBPFP 2.6344 equivalent bits/point -MSE 0.908218 ----------------------- -------------------------------------------------------- -Time: 0.990s Load: 0.016s, Pack+Encode: 0.372s, Decode+Unpack: 0.601s ----------------------- -------------------------------------------------------- -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 0.9082 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,496B, BPFP=0.3432 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,404B, BPFP=1.6904 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,484B, BPFP=1.0768 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,696B, BPFP=1.8394 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,300B, BPFP=1.2855 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,276B, BPFP=1.8768 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,776B, BPFP=1.3205 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,660B, BPFP=1.8385 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,944B, BPFP=1.1114 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,740B, BPFP=1.8878 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,580B, BPFP=1.2641 -⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.713s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.08087121 - layer.0.v_cache 0.00000027 0.00024971 - layer.1.k_cache 0.00289258 0.75174228 - layer.1.v_cache 0.00000083 0.00089626 - layer.2.k_cache 0.00117497 0.45129778 - layer.2.v_cache 0.00000126 0.00133598 - layer.3.k_cache 0.00131231 0.49661158 - layer.3.v_cache 0.00000230 0.00215461 - layer.4.k_cache 0.00357793 0.87343177 - layer.4.v_cache 0.00000347 0.00372584 - layer.4.output 0.00014070 0.05877808 - ------------------------------------------------------------------------------------- - TOTAL 0.00237876 0.99267352 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 816356 -BPFP 1.3805 bits/point -EBPFP 2.7609 equivalent bits/point -MSE 0.992674 ----------------------- -------------------------------------------------------- -Time: 1.152s Load: 0.017s, Pack+Encode: 0.422s, Decode+Unpack: 0.713s ----------------------- -------------------------------------------------------- -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 0.9927 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 16,128B, BPFP=0.3529 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 76,064B, BPFP=1.6646 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,576B, BPFP=1.0630 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 81,380B, BPFP=1.7809 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,280B, BPFP=1.2535 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 82,592B, BPFP=1.8074 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,696B, BPFP=1.3064 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 81,404B, BPFP=1.7814 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,060B, BPFP=1.0955 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 83,508B, BPFP=1.8275 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 235,784B, BPFP=1.2900 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.716s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.15196489 - layer.0.v_cache 0.00000027 0.00024853 - layer.1.k_cache 0.00290374 0.76412801 - layer.1.v_cache 0.00000082 0.00088969 - layer.2.k_cache 0.00112985 0.44499557 - layer.2.v_cache 0.00000115 0.00125159 - layer.3.k_cache 0.00131122 0.50847718 - layer.3.v_cache 0.00000220 0.00210881 - layer.4.k_cache 0.00345488 0.85479873 - layer.4.v_cache 0.00000325 0.00359630 - layer.4.output 0.00016267 0.06494947 - ------------------------------------------------------------------------------------- - TOTAL 0.00243901 0.99944694 - (elements=5,117,952) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5117952 -Total Bytes 872472 -BPFP 1.3638 bits/point -EBPFP 2.7276 equivalent bits/point -MSE 0.999447 ----------------------- -------------------------------------------------------- -Time: 1.155s Load: 0.019s, Pack+Encode: 0.420s, Decode+Unpack: 0.716s ----------------------- -------------------------------------------------------- -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 0.9994 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,772B, BPFP=0.3442 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 75,788B, BPFP=1.6539 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,424B, BPFP=1.0567 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 81,516B, BPFP=1.7789 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,344B, BPFP=1.2514 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 83,044B, BPFP=1.8122 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,416B, BPFP=1.2966 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 81,392B, BPFP=1.7762 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,096B, BPFP=1.0932 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 83,896B, BPFP=1.8308 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 229,752B, BPFP=1.2534 -⌛️ [2/4] FRONTEND: Frontend time: 0.439s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.718s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.09487114 - layer.0.v_cache 0.00000026 0.00024594 - layer.1.k_cache 0.00288075 0.75474881 - layer.1.v_cache 0.00000077 0.00089110 - layer.2.k_cache 0.00118336 0.44661419 - layer.2.v_cache 0.00000117 0.00131132 - layer.3.k_cache 0.00130988 0.49555279 - layer.3.v_cache 0.00000211 0.00209362 - layer.4.k_cache 0.00353008 0.85562296 - layer.4.v_cache 0.00000317 0.00363820 - layer.4.output 0.00013834 0.05968491 - ------------------------------------------------------------------------------------- - TOTAL 0.00240333 0.99245212 - (elements=5,132,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5132288 -Total Bytes 866440 -BPFP 1.3506 bits/point -EBPFP 2.7011 equivalent bits/point -MSE 0.992452 ----------------------- -------------------------------------------------------- -Time: 1.175s Load: 0.019s, Pack+Encode: 0.439s, Decode+Unpack: 0.718s ----------------------- -------------------------------------------------------- -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 0.9925 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,264B, BPFP=0.3549 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,960B, BPFP=1.6732 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,792B, BPFP=1.0647 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,728B, BPFP=1.8305 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,712B, BPFP=1.2721 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,496B, BPFP=1.8717 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,884B, BPFP=1.3226 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,556B, BPFP=1.8265 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,020B, BPFP=1.0933 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,896B, BPFP=1.8810 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 217,496B, BPFP=1.2643 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.709s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.31607128 - layer.0.v_cache 0.00000026 0.00023854 - layer.1.k_cache 0.00285290 0.75204186 - layer.1.v_cache 0.00000077 0.00089206 - layer.2.k_cache 0.00116461 0.45566582 - layer.2.v_cache 0.00000115 0.00134889 - layer.3.k_cache 0.00131253 0.50507418 - layer.3.v_cache 0.00000220 0.00213688 - layer.4.k_cache 0.00369396 0.89274970 - layer.4.v_cache 0.00000328 0.00369328 - layer.4.output 0.00015589 0.06255630 - ------------------------------------------------------------------------------------- - TOTAL 0.00240954 0.94143841 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 827804 -BPFP 1.3748 bits/point -EBPFP 2.7497 equivalent bits/point -MSE 0.941438 ----------------------- -------------------------------------------------------- -Time: 1.138s Load: 0.017s, Pack+Encode: 0.412s, Decode+Unpack: 0.709s ----------------------- -------------------------------------------------------- -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 0.9414 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,480B, BPFP=0.3495 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,356B, BPFP=1.6563 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,504B, BPFP=1.0500 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,100B, BPFP=1.7860 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,332B, BPFP=1.2494 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,572B, BPFP=1.8193 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,156B, BPFP=1.2906 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,356B, BPFP=1.7918 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,392B, BPFP=1.0701 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,348B, BPFP=1.8368 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 225,112B, BPFP=1.2707 -⌛️ [2/4] FRONTEND: Frontend time: 0.415s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.03338941 - layer.0.v_cache 0.00000027 0.00024600 - layer.1.k_cache 0.00290669 0.75825906 - layer.1.v_cache 0.00000082 0.00089931 - layer.2.k_cache 0.00115107 0.44562888 - layer.2.v_cache 0.00000117 0.00130093 - layer.3.k_cache 0.00130922 0.51044014 - layer.3.v_cache 0.00000214 0.00211099 - layer.4.k_cache 0.00347357 0.87843949 - layer.4.v_cache 0.00000325 0.00358479 - layer.4.output 0.00015455 0.06542387 - ------------------------------------------------------------------------------------- - TOTAL 0.00246754 0.99257103 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 840708 -BPFP 1.3559 bits/point -EBPFP 2.7118 equivalent bits/point -MSE 0.992571 ----------------------- -------------------------------------------------------- -Time: 1.142s Load: 0.018s, Pack+Encode: 0.415s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9926 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,540B, BPFP=0.3540 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,032B, BPFP=1.6407 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,856B, BPFP=1.0672 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,900B, BPFP=1.7971 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,400B, BPFP=1.2618 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,932B, BPFP=1.8206 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,696B, BPFP=1.3141 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,440B, BPFP=1.7866 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,764B, BPFP=1.0879 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,880B, BPFP=1.8422 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 210,748B, BPFP=1.2001 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.703s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.25704022 - layer.0.v_cache 0.00000026 0.00023231 - layer.1.k_cache 0.00292563 0.74507977 - layer.1.v_cache 0.00000076 0.00088468 - layer.2.k_cache 0.00117950 0.44740429 - layer.2.v_cache 0.00000116 0.00125125 - layer.3.k_cache 0.00131686 0.49907980 - layer.3.v_cache 0.00000208 0.00201085 - layer.4.k_cache 0.00357410 0.87365883 - layer.4.v_cache 0.00000334 0.00348909 - layer.4.output 0.00016206 0.05869959 - ------------------------------------------------------------------------------------- - TOTAL 0.00233876 0.93320925 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 824188 -BPFP 1.3409 bits/point -EBPFP 2.6818 equivalent bits/point -MSE 0.933209 ----------------------- -------------------------------------------------------- -Time: 1.135s Load: 0.018s, Pack+Encode: 0.414s, Decode+Unpack: 0.703s ----------------------- -------------------------------------------------------- -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 0.9332 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,516B, BPFP=0.3377 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 75,856B, BPFP=1.6508 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,492B, BPFP=1.0770 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,876B, BPFP=1.8035 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,324B, BPFP=1.2692 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 83,928B, BPFP=1.8264 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,040B, BPFP=1.3066 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 81,976B, BPFP=1.7839 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,596B, BPFP=1.1011 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,144B, BPFP=1.8311 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 224,368B, BPFP=1.2207 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (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, 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: 0.704s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.48141063 - layer.0.v_cache 0.00000027 0.00024334 - layer.1.k_cache 0.00287336 0.72812340 - layer.1.v_cache 0.00000081 0.00094392 - layer.2.k_cache 0.00118844 0.45971476 - layer.2.v_cache 0.00000120 0.00136044 - layer.3.k_cache 0.00133454 0.49773605 - layer.3.v_cache 0.00000220 0.00214709 - layer.4.k_cache 0.00356152 0.88181001 - layer.4.v_cache 0.00000363 0.00374292 - layer.4.output 0.00015158 0.06235049 - ------------------------------------------------------------------------------------- - TOTAL 0.00239430 0.95047390 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 867116 -BPFP 1.3479 bits/point -EBPFP 2.6957 equivalent bits/point -MSE 0.950474 ----------------------- -------------------------------------------------------- -Time: 1.135s Load: 0.019s, Pack+Encode: 0.413s, Decode+Unpack: 0.704s ----------------------- -------------------------------------------------------- -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 0.9505 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,920B, BPFP=0.3464 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 76,136B, BPFP=1.6569 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,864B, BPFP=1.0634 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,360B, BPFP=1.7923 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,376B, BPFP=1.2704 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 83,604B, BPFP=1.8194 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,100B, BPFP=1.3079 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 81,952B, BPFP=1.7834 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,404B, BPFP=1.0969 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,256B, BPFP=1.8336 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 229,028B, BPFP=1.2460 -⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.722s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.27613257 - layer.0.v_cache 0.00000027 0.00025024 - layer.1.k_cache 0.00293092 0.74122309 - layer.1.v_cache 0.00000080 0.00091775 - layer.2.k_cache 0.00116855 0.45174584 - layer.2.v_cache 0.00000117 0.00133358 - layer.3.k_cache 0.00131991 0.50153502 - layer.3.v_cache 0.00000231 0.00222086 - layer.4.k_cache 0.00358756 0.88710919 - layer.4.v_cache 0.00000356 0.00379914 - layer.4.output 0.00016308 0.06131380 - ------------------------------------------------------------------------------------- - TOTAL 0.00236086 0.93653732 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 871000 -BPFP 1.3539 bits/point -EBPFP 2.7078 equivalent bits/point -MSE 0.936537 ----------------------- -------------------------------------------------------- -Time: 1.179s Load: 0.018s, Pack+Encode: 0.440s, Decode+Unpack: 0.722s ----------------------- -------------------------------------------------------- -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 0.9365 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,628B, BPFP=0.3519 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,444B, BPFP=1.6535 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,204B, BPFP=1.0628 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,868B, BPFP=1.7982 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,952B, BPFP=1.2597 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,756B, BPFP=1.8182 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,744B, BPFP=1.3001 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,332B, BPFP=1.7861 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,184B, BPFP=1.0848 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,592B, BPFP=1.8370 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,668B, BPFP=1.2139 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.695s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.44814509 - layer.0.v_cache 0.00000026 0.00024919 - layer.1.k_cache 0.00293388 0.75418742 - layer.1.v_cache 0.00000079 0.00092300 - layer.2.k_cache 0.00115427 0.44679674 - layer.2.v_cache 0.00000115 0.00128254 - layer.3.k_cache 0.00129772 0.51474931 - layer.3.v_cache 0.00000217 0.00211566 - layer.4.k_cache 0.00353122 0.87008403 - layer.4.v_cache 0.00000331 0.00370133 - layer.4.output 0.00014441 0.05809460 - ------------------------------------------------------------------------------------- - TOTAL 0.00235152 0.94818662 - (elements=4,974,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4974592 -Total Bytes 835372 -BPFP 1.3434 bits/point -EBPFP 2.6868 equivalent bits/point -MSE 0.948187 ----------------------- -------------------------------------------------------- -Time: 1.125s Load: 0.018s, Pack+Encode: 0.412s, Decode+Unpack: 0.695s ----------------------- -------------------------------------------------------- -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 0.9482 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,828B, BPFP=0.3438 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,316B, BPFP=1.6533 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,868B, BPFP=1.0633 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,472B, BPFP=1.7728 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,176B, BPFP=1.2559 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,508B, BPFP=1.8200 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,676B, BPFP=1.3139 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,428B, BPFP=1.7950 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,900B, BPFP=1.0873 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,816B, BPFP=1.8503 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,860B, BPFP=1.2452 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.702s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.91934449 - layer.0.v_cache 0.00000026 0.00024224 - layer.1.k_cache 0.00293885 0.79270052 - layer.1.v_cache 0.00000076 0.00082981 - layer.2.k_cache 0.00113696 0.45330874 - layer.2.v_cache 0.00000113 0.00119040 - layer.3.k_cache 0.00132482 0.51772899 - layer.3.v_cache 0.00000208 0.00198735 - layer.4.k_cache 0.00376518 0.93457937 - layer.4.v_cache 0.00000304 0.00341696 - layer.4.output 0.00014776 0.06316050 - ------------------------------------------------------------------------------------- - TOTAL 0.00241421 0.99128363 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 816848 -BPFP 1.3526 bits/point -EBPFP 2.7052 equivalent bits/point -MSE 0.991284 ----------------------- -------------------------------------------------------- -Time: 1.134s Load: 0.018s, Pack+Encode: 0.414s, Decode+Unpack: 0.702s ----------------------- -------------------------------------------------------- -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 0.9913 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,356B, BPFP=0.3418 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,456B, BPFP=1.6350 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,812B, BPFP=1.0419 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,200B, BPFP=1.7628 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,548B, BPFP=1.2364 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,720B, BPFP=1.7967 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,508B, BPFP=1.2800 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,940B, BPFP=1.7570 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,060B, BPFP=1.0697 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,176B, BPFP=1.8068 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 217,684B, BPFP=1.2113 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (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, 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: 0.692s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.86408253 - layer.0.v_cache 0.00000026 0.00024976 - layer.1.k_cache 0.00294286 0.76301336 - layer.1.v_cache 0.00000076 0.00087412 - layer.2.k_cache 0.00116394 0.44939829 - layer.2.v_cache 0.00000112 0.00126069 - layer.3.k_cache 0.00133268 0.50039938 - layer.3.v_cache 0.00000207 0.00202279 - layer.4.k_cache 0.00360177 0.88886515 - layer.4.v_cache 0.00000315 0.00354083 - layer.4.output 0.00019540 0.06060964 - ------------------------------------------------------------------------------------- - TOTAL 0.00250678 0.97972468 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 834460 -BPFP 1.3267 bits/point -EBPFP 2.6533 equivalent bits/point -MSE 0.979725 ----------------------- -------------------------------------------------------- -Time: 1.115s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.692s ----------------------- -------------------------------------------------------- -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 0.9797 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,540B, BPFP=0.3509 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,928B, BPFP=1.6467 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,832B, BPFP=1.0574 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,636B, BPFP=1.7981 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,560B, BPFP=1.2545 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,008B, BPFP=1.8291 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,224B, BPFP=1.2921 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,416B, BPFP=1.7932 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,668B, BPFP=1.0763 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,492B, BPFP=1.8400 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 219,832B, BPFP=1.2409 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.703s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.06429789 - layer.0.v_cache 0.00000026 0.00023937 - layer.1.k_cache 0.00295457 0.77022883 - layer.1.v_cache 0.00000081 0.00092533 - layer.2.k_cache 0.00115565 0.44565573 - layer.2.v_cache 0.00000119 0.00133758 - layer.3.k_cache 0.00131728 0.49703477 - layer.3.v_cache 0.00000219 0.00214308 - layer.4.k_cache 0.00358690 0.88892528 - layer.4.v_cache 0.00000349 0.00369393 - layer.4.output 0.00015659 0.06116769 - ------------------------------------------------------------------------------------- - TOTAL 0.00232721 0.99422518 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 837136 -BPFP 1.3501 bits/point -EBPFP 2.7003 equivalent bits/point -MSE 0.994225 ----------------------- -------------------------------------------------------- -Time: 1.134s Load: 0.018s, Pack+Encode: 0.414s, Decode+Unpack: 0.703s ----------------------- -------------------------------------------------------- -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 0.9942 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,604B, BPFP=0.3585 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,932B, BPFP=1.6758 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,284B, BPFP=1.0635 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,436B, BPFP=1.8253 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,180B, BPFP=1.2679 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,652B, BPFP=1.8532 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,896B, BPFP=1.3074 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,080B, BPFP=1.8171 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,452B, BPFP=1.0903 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,236B, BPFP=1.8666 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,344B, BPFP=1.2428 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.692s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.91475256 - layer.0.v_cache 0.00000026 0.00025119 - layer.1.k_cache 0.00281368 0.74998241 - layer.1.v_cache 0.00000080 0.00092898 - layer.2.k_cache 0.00120688 0.45606003 - layer.2.v_cache 0.00000119 0.00137002 - layer.3.k_cache 0.00132000 0.50171630 - layer.3.v_cache 0.00000216 0.00215162 - layer.4.k_cache 0.00354241 0.88681515 - layer.4.v_cache 0.00000378 0.00379001 - layer.4.output 0.00015030 0.05949226 - ------------------------------------------------------------------------------------- - TOTAL 0.00236476 0.98255623 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 831096 -BPFP 1.3641 bits/point -EBPFP 2.7281 equivalent bits/point -MSE 0.982556 ----------------------- -------------------------------------------------------- -Time: 1.119s Load: 0.018s, Pack+Encode: 0.408s, Decode+Unpack: 0.692s ----------------------- -------------------------------------------------------- -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 0.9826 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,780B, BPFP=0.3584 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,248B, BPFP=1.6635 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,448B, BPFP=1.0549 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,192B, BPFP=1.7985 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,144B, BPFP=1.2524 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,528B, BPFP=1.8289 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,236B, BPFP=1.2999 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,416B, BPFP=1.8036 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,072B, BPFP=1.0918 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,808B, BPFP=1.8579 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 221,656B, BPFP=1.2585 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.703s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.83390205 - layer.0.v_cache 0.00000027 0.00023686 - layer.1.k_cache 0.00289516 0.74300300 - layer.1.v_cache 0.00000077 0.00087027 - layer.2.k_cache 0.00116570 0.44126648 - layer.2.v_cache 0.00000112 0.00125726 - layer.3.k_cache 0.00132982 0.49719713 - layer.3.v_cache 0.00000218 0.00208275 - layer.4.k_cache 0.00351457 0.86515702 - layer.4.v_cache 0.00000367 0.00374540 - layer.4.output 0.00015182 0.06175562 - ------------------------------------------------------------------------------------- - TOTAL 0.00233825 0.97398148 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 838528 -BPFP 1.3603 bits/point -EBPFP 2.7205 equivalent bits/point -MSE 0.973981 ----------------------- -------------------------------------------------------- -Time: 1.132s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.703s ----------------------- -------------------------------------------------------- -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 0.9740 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,172B, BPFP=0.3517 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,272B, BPFP=1.6754 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,716B, BPFP=1.0598 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,700B, BPFP=1.8245 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,364B, BPFP=1.2603 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,840B, BPFP=1.8509 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,492B, BPFP=1.3096 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,912B, BPFP=1.8062 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,096B, BPFP=1.0918 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,492B, BPFP=1.8660 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 210,952B, BPFP=1.2226 -⌛️ [2/4] FRONTEND: Frontend time: 0.409s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.695s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.02155899 - layer.0.v_cache 0.00000026 0.00024827 - layer.1.k_cache 0.00286757 0.76575477 - layer.1.v_cache 0.00000078 0.00090777 - layer.2.k_cache 0.00117745 0.45344444 - layer.2.v_cache 0.00000112 0.00128074 - layer.3.k_cache 0.00130673 0.50598674 - layer.3.v_cache 0.00000209 0.00204997 - layer.4.k_cache 0.00356837 0.87356748 - layer.4.v_cache 0.00000322 0.00362721 - layer.4.output 0.00014638 0.05925175 - ------------------------------------------------------------------------------------- - TOTAL 0.00241942 0.99038810 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 819008 -BPFP 1.3562 bits/point -EBPFP 2.7124 equivalent bits/point -MSE 0.990388 ----------------------- -------------------------------------------------------- -Time: 1.122s Load: 0.018s, Pack+Encode: 0.409s, Decode+Unpack: 0.695s ----------------------- -------------------------------------------------------- -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 0.9904 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.026s - ------------------------------------------------------------- -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: 17,464B, BPFP=0.3454 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 84,016B, BPFP=1.6617 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 53,824B, BPFP=1.0646 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 90,556B, BPFP=1.7911 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 62,860B, BPFP=1.2433 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 91,908B, BPFP=1.8178 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 66,208B, BPFP=1.3095 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 90,472B, BPFP=1.7894 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 55,224B, BPFP=1.0922 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 93,252B, BPFP=1.8444 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 251,060B, BPFP=1.2414 -⌛️ [2/4] FRONTEND: Frontend time: 0.537s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.807s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.80248838 - layer.0.v_cache 0.00000027 0.00024989 - layer.1.k_cache 0.00291865 0.74209873 - layer.1.v_cache 0.00000080 0.00090122 - layer.2.k_cache 0.00114101 0.44490781 - layer.2.v_cache 0.00000113 0.00124030 - layer.3.k_cache 0.00133612 0.50979197 - layer.3.v_cache 0.00000221 0.00204054 - layer.4.k_cache 0.00365959 0.90457547 - layer.4.v_cache 0.00000309 0.00349301 - layer.4.output 0.00017803 0.06346853 - ------------------------------------------------------------------------------------- - TOTAL 0.00236502 0.97611868 - (elements=5,662,720) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5662720 -Total Bytes 956844 -BPFP 1.3518 bits/point -EBPFP 2.7036 equivalent bits/point -MSE 0.976119 ----------------------- -------------------------------------------------------- -Time: 1.370s Load: 0.026s, Pack+Encode: 0.537s, Decode+Unpack: 0.807s ----------------------- -------------------------------------------------------- -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 0.9761 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,312B, BPFP=0.3438 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,516B, BPFP=1.6504 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,204B, BPFP=1.0597 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,540B, BPFP=1.7857 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,940B, BPFP=1.2558 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,748B, BPFP=1.8128 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,724B, BPFP=1.2959 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,988B, BPFP=1.7733 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,260B, BPFP=1.0834 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,388B, BPFP=1.8271 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,356B, BPFP=1.1974 -⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.696s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.06357969 - layer.0.v_cache 0.00000026 0.00025087 - layer.1.k_cache 0.00289751 0.74061532 - layer.1.v_cache 0.00000077 0.00090976 - layer.2.k_cache 0.00114660 0.45441524 - layer.2.v_cache 0.00000117 0.00132518 - layer.3.k_cache 0.00133042 0.49610730 - layer.3.v_cache 0.00000210 0.00207981 - layer.4.k_cache 0.00358099 0.89293574 - layer.4.v_cache 0.00000335 0.00370951 - layer.4.output 0.00013932 0.05727432 - ------------------------------------------------------------------------------------- - TOTAL 0.00252269 0.99178755 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 831976 -BPFP 1.3341 bits/point -EBPFP 2.6682 equivalent bits/point -MSE 0.991788 ----------------------- -------------------------------------------------------- -Time: 1.146s Load: 0.018s, Pack+Encode: 0.433s, Decode+Unpack: 0.696s ----------------------- -------------------------------------------------------- -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 0.9918 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,384B, BPFP=0.3556 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,140B, BPFP=1.6105 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,440B, BPFP=1.0492 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,836B, BPFP=1.7266 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,008B, BPFP=1.2364 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,856B, BPFP=1.7518 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,696B, BPFP=1.2781 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,860B, BPFP=1.7272 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,664B, BPFP=1.0795 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,524B, BPFP=1.7683 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 199,040B, BPFP=1.2302 -⌛️ [2/4] FRONTEND: Frontend time: 0.363s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.590s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.75189054 - layer.0.v_cache 0.00000027 0.00023459 - layer.1.k_cache 0.00293329 0.70167976 - layer.1.v_cache 0.00000080 0.00088640 - layer.2.k_cache 0.00117467 0.44329221 - layer.2.v_cache 0.00000111 0.00127367 - layer.3.k_cache 0.00131319 0.49459795 - layer.3.v_cache 0.00000216 0.00207329 - layer.4.k_cache 0.00359311 0.84440767 - layer.4.v_cache 0.00000344 0.00363018 - layer.4.output 0.00015753 0.05920485 - ------------------------------------------------------------------------------------- - TOTAL 0.00248390 0.96291326 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 748448 -BPFP 1.3217 bits/point -EBPFP 2.6434 equivalent bits/point -MSE 0.962913 ----------------------- -------------------------------------------------------- -Time: 0.970s Load: 0.017s, Pack+Encode: 0.363s, Decode+Unpack: 0.590s ----------------------- -------------------------------------------------------- -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 0.9629 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,472B, BPFP=0.3473 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,012B, BPFP=1.6391 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,156B, BPFP=1.0586 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,628B, BPFP=1.7876 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,692B, BPFP=1.2503 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,180B, BPFP=1.8225 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,692B, BPFP=1.2952 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,488B, BPFP=1.7845 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,244B, BPFP=1.0831 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,636B, BPFP=1.8327 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 218,080B, BPFP=1.2240 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.707s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.90898185 - layer.0.v_cache 0.00000027 0.00025119 - layer.1.k_cache 0.00288004 0.76063117 - layer.1.v_cache 0.00000079 0.00092590 - layer.2.k_cache 0.00117070 0.45483412 - layer.2.v_cache 0.00000116 0.00136108 - layer.3.k_cache 0.00131678 0.50218850 - layer.3.v_cache 0.00000214 0.00214746 - layer.4.k_cache 0.00357349 0.90193203 - layer.4.v_cache 0.00000347 0.00373882 - layer.4.output 0.00013697 0.05740777 - ------------------------------------------------------------------------------------- - TOTAL 0.00237019 0.98333023 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 837280 -BPFP 1.3426 bits/point -EBPFP 2.6852 equivalent bits/point -MSE 0.983330 ----------------------- -------------------------------------------------------- -Time: 1.135s Load: 0.017s, Pack+Encode: 0.412s, Decode+Unpack: 0.707s ----------------------- -------------------------------------------------------- -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 0.9833 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,344B, BPFP=0.3475 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,556B, BPFP=1.6430 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,772B, BPFP=1.0591 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,212B, BPFP=1.7937 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,852B, BPFP=1.2648 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,992B, BPFP=1.8341 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,596B, BPFP=1.3043 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,416B, BPFP=1.7984 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,960B, BPFP=1.0861 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,548B, BPFP=1.8466 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 217,888B, BPFP=1.2335 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.709s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.63300144 - layer.0.v_cache 0.00000026 0.00023893 - layer.1.k_cache 0.00288937 0.74978408 - layer.1.v_cache 0.00000077 0.00090345 - layer.2.k_cache 0.00116899 0.45101823 - layer.2.v_cache 0.00000115 0.00132544 - layer.3.k_cache 0.00131620 0.49817089 - layer.3.v_cache 0.00000216 0.00210142 - layer.4.k_cache 0.00353981 0.87774154 - layer.4.v_cache 0.00000338 0.00371847 - layer.4.output 0.00014567 0.05858885 - ------------------------------------------------------------------------------------- - TOTAL 0.00236310 0.96088281 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 835136 -BPFP 1.3508 bits/point -EBPFP 2.7017 equivalent bits/point -MSE 0.960883 ----------------------- -------------------------------------------------------- -Time: 1.140s Load: 0.017s, Pack+Encode: 0.414s, Decode+Unpack: 0.709s ----------------------- -------------------------------------------------------- -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 0.9609 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,928B, BPFP=0.3534 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,636B, BPFP=1.6959 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,560B, BPFP=1.0786 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,640B, BPFP=1.8381 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,292B, BPFP=1.2853 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,488B, BPFP=1.8818 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,036B, BPFP=1.3266 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,704B, BPFP=1.8396 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,740B, BPFP=1.1065 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,168B, BPFP=1.8979 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,624B, BPFP=1.2762 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.707s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.06849254 - layer.0.v_cache 0.00000027 0.00024964 - layer.1.k_cache 0.00287412 0.77851257 - layer.1.v_cache 0.00000081 0.00090407 - layer.2.k_cache 0.00119273 0.45620630 - layer.2.v_cache 0.00000117 0.00134153 - layer.3.k_cache 0.00130297 0.50206419 - layer.3.v_cache 0.00000221 0.00216907 - layer.4.k_cache 0.00352071 0.88598466 - layer.4.v_cache 0.00000330 0.00379257 - layer.4.output 0.00014370 0.05811590 - ------------------------------------------------------------------------------------- - TOTAL 0.00241545 0.99515577 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 819816 -BPFP 1.3863 bits/point -EBPFP 2.7726 equivalent bits/point -MSE 0.995156 ----------------------- -------------------------------------------------------- -Time: 1.138s Load: 0.017s, Pack+Encode: 0.414s, Decode+Unpack: 0.707s ----------------------- -------------------------------------------------------- -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 0.9952 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,692B, BPFP=0.3437 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,540B, BPFP=1.6734 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,612B, BPFP=1.0669 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,688B, BPFP=1.8172 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,860B, BPFP=1.2598 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,412B, BPFP=1.8575 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,216B, BPFP=1.3149 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,444B, BPFP=1.8115 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,944B, BPFP=1.0981 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,956B, BPFP=1.8702 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,660B, BPFP=1.2670 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (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, 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: 0.693s - -[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, 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 10.68607390 - layer.0.v_cache 0.00000026 0.00024748 - layer.1.k_cache 0.00290221 0.78073732 - layer.1.v_cache 0.00000079 0.00091604 - layer.2.k_cache 0.00116170 0.45489136 - layer.2.v_cache 0.00000115 0.00133460 - layer.3.k_cache 0.00132769 0.50944259 - layer.3.v_cache 0.00000211 0.00210225 - layer.4.k_cache 0.00362300 0.91393111 - layer.4.v_cache 0.00000352 0.00369629 - layer.4.output 0.00014243 0.06470095 - ------------------------------------------------------------------------------------- - TOTAL 0.00237320 0.97229834 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 820024 -BPFP 1.3701 bits/point -EBPFP 2.7401 equivalent bits/point -MSE 0.972298 ----------------------- -------------------------------------------------------- -Time: 1.118s Load: 0.017s, Pack+Encode: 0.407s, Decode+Unpack: 0.693s ----------------------- -------------------------------------------------------- -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 0.9723 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,864B, BPFP=0.3519 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,316B, BPFP=1.6884 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,372B, BPFP=1.0741 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,432B, BPFP=1.8331 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,980B, BPFP=1.2779 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,156B, BPFP=1.8740 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,916B, BPFP=1.3238 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,580B, BPFP=1.8366 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,768B, BPFP=1.1072 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,876B, BPFP=1.8910 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,828B, BPFP=1.2833 -⌛️ [2/4] FRONTEND: Frontend time: 0.426s (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, 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: 0.702s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.14876968 - layer.0.v_cache 0.00000026 0.00025261 - layer.1.k_cache 0.00288475 0.77152927 - layer.1.v_cache 0.00000079 0.00090300 - layer.2.k_cache 0.00119186 0.45759152 - layer.2.v_cache 0.00000115 0.00131103 - layer.3.k_cache 0.00132245 0.49951509 - layer.3.v_cache 0.00000226 0.00217184 - layer.4.k_cache 0.00352035 0.87894528 - layer.4.v_cache 0.00000339 0.00368857 - layer.4.output 0.00014978 0.06401505 - ------------------------------------------------------------------------------------- - TOTAL 0.00240252 1.00148129 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 819088 -BPFP 1.3851 bits/point -EBPFP 2.7702 equivalent bits/point -MSE 1.001481 ----------------------- -------------------------------------------------------- -Time: 1.145s Load: 0.016s, Pack+Encode: 0.426s, Decode+Unpack: 0.702s ----------------------- -------------------------------------------------------- -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 1.0015 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,532B, BPFP=0.3451 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,272B, BPFP=1.6924 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,848B, BPFP=1.0650 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,688B, BPFP=1.8210 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,580B, BPFP=1.2723 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,100B, BPFP=1.8546 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,740B, BPFP=1.3236 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,852B, BPFP=1.8249 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,780B, BPFP=1.1108 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,444B, BPFP=1.8865 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,196B, BPFP=1.2775 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.71720575 - layer.0.v_cache 0.00000027 0.00024775 - layer.1.k_cache 0.00291730 0.76055862 - layer.1.v_cache 0.00000079 0.00087781 - layer.2.k_cache 0.00116725 0.45502437 - layer.2.v_cache 0.00000113 0.00127348 - layer.3.k_cache 0.00130327 0.50340739 - layer.3.v_cache 0.00000218 0.00207649 - layer.4.k_cache 0.00359671 0.86891123 - layer.4.v_cache 0.00000327 0.00373041 - layer.4.output 0.00014627 0.06383664 - ------------------------------------------------------------------------------------- - TOTAL 0.00241763 0.96918999 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 813032 -BPFP 1.3790 bits/point -EBPFP 2.7581 equivalent bits/point -MSE 0.969190 ----------------------- -------------------------------------------------------- -Time: 1.136s Load: 0.016s, Pack+Encode: 0.412s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9692 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,972B, BPFP=0.3419 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 77,028B, BPFP=1.6487 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,648B, BPFP=1.0627 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 83,288B, BPFP=1.7827 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 59,040B, BPFP=1.2637 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 84,608B, BPFP=1.8110 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,576B, BPFP=1.2966 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 82,960B, BPFP=1.7757 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 51,024B, BPFP=1.0921 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,916B, BPFP=1.8176 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 234,744B, BPFP=1.2561 -⌛️ [2/4] FRONTEND: Frontend time: 0.410s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.66194884 - layer.0.v_cache 0.00000027 0.00024728 - layer.1.k_cache 0.00290279 0.76642247 - layer.1.v_cache 0.00000079 0.00091750 - layer.2.k_cache 0.00116885 0.45501019 - layer.2.v_cache 0.00000114 0.00134664 - layer.3.k_cache 0.00131452 0.49604505 - layer.3.v_cache 0.00000216 0.00213120 - layer.4.k_cache 0.00360833 0.86587775 - layer.4.v_cache 0.00000353 0.00368399 - layer.4.output 0.00014951 0.05820739 - ------------------------------------------------------------------------------------- - TOTAL 0.00246232 0.96331861 - (elements=5,232,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5232640 -Total Bytes 883804 -BPFP 1.3512 bits/point -EBPFP 2.7024 equivalent bits/point -MSE 0.963319 ----------------------- -------------------------------------------------------- -Time: 1.138s Load: 0.019s, Pack+Encode: 0.410s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9633 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,768B, BPFP=0.3444 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,080B, BPFP=1.6810 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,844B, BPFP=1.0691 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,808B, BPFP=1.8379 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,588B, BPFP=1.2730 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,868B, BPFP=1.8626 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,212B, BPFP=1.3109 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,732B, BPFP=1.8128 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,936B, BPFP=1.0946 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,368B, BPFP=1.8743 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,720B, BPFP=1.2519 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.701s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.54874942 - layer.0.v_cache 0.00000027 0.00024658 - layer.1.k_cache 0.00287179 0.74966221 - layer.1.v_cache 0.00000079 0.00092103 - layer.2.k_cache 0.00118919 0.45186626 - layer.2.v_cache 0.00000114 0.00132265 - layer.3.k_cache 0.00130718 0.49680599 - layer.3.v_cache 0.00000215 0.00210288 - layer.4.k_cache 0.00355403 0.87184494 - layer.4.v_cache 0.00000335 0.00360523 - layer.4.output 0.00013864 0.05963963 - ------------------------------------------------------------------------------------- - TOTAL 0.00243638 0.95469184 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 821924 -BPFP 1.3691 bits/point -EBPFP 2.7383 equivalent bits/point -MSE 0.954692 ----------------------- -------------------------------------------------------- -Time: 1.130s Load: 0.016s, Pack+Encode: 0.412s, Decode+Unpack: 0.701s ----------------------- -------------------------------------------------------- -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 0.9547 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.020s - ------------------------------------------------------------- -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: 15,236B, BPFP=0.3553 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,496B, BPFP=1.6674 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,836B, BPFP=1.0689 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,916B, BPFP=1.8171 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,476B, BPFP=1.2704 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,820B, BPFP=1.8615 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,180B, BPFP=1.3102 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,728B, BPFP=1.8127 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,524B, BPFP=1.0850 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,900B, BPFP=1.8633 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,684B, BPFP=1.2458 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.30193491 - layer.0.v_cache 0.00000026 0.00024084 - layer.1.k_cache 0.00289851 0.75224309 - layer.1.v_cache 0.00000079 0.00088940 - layer.2.k_cache 0.00115963 0.45891013 - layer.2.v_cache 0.00000116 0.00134962 - layer.3.k_cache 0.00131431 0.50261162 - layer.3.v_cache 0.00000219 0.00216577 - layer.4.k_cache 0.00369786 0.89094348 - layer.4.v_cache 0.00000335 0.00368959 - layer.4.output 0.00014933 0.06359622 - ------------------------------------------------------------------------------------- - TOTAL 0.00240113 0.94066881 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 818796 -BPFP 1.3639 bits/point -EBPFP 2.7279 equivalent bits/point -MSE 0.940669 ----------------------- -------------------------------------------------------- -Time: 1.148s Load: 0.020s, Pack+Encode: 0.420s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9407 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,256B, BPFP=0.3516 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,644B, BPFP=1.6741 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,512B, BPFP=1.0719 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,836B, BPFP=1.8168 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,320B, BPFP=1.2749 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,736B, BPFP=1.8606 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,896B, BPFP=1.3112 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,056B, BPFP=1.8219 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,300B, BPFP=1.0901 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,292B, BPFP=1.8734 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 218,696B, BPFP=1.2600 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.718s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.78789653 - layer.0.v_cache 0.00000027 0.00024972 - layer.1.k_cache 0.00292607 0.76979313 - layer.1.v_cache 0.00000081 0.00088649 - layer.2.k_cache 0.00125336 0.45720459 - layer.2.v_cache 0.00000116 0.00132835 - layer.3.k_cache 0.00131849 0.50401684 - layer.3.v_cache 0.00000220 0.00212745 - layer.4.k_cache 0.00356446 0.86485646 - layer.4.v_cache 0.00000343 0.00366171 - layer.4.output 0.00016600 0.06198732 - ------------------------------------------------------------------------------------- - TOTAL 0.00239722 0.97428361 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 832544 -BPFP 1.3705 bits/point -EBPFP 2.7409 equivalent bits/point -MSE 0.974284 ----------------------- -------------------------------------------------------- -Time: 1.148s Load: 0.017s, Pack+Encode: 0.414s, Decode+Unpack: 0.718s ----------------------- -------------------------------------------------------- -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 0.9743 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,736B, BPFP=0.3396 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 76,620B, BPFP=1.6536 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,244B, BPFP=1.0628 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,876B, BPFP=1.7886 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,648B, BPFP=1.2657 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 84,064B, BPFP=1.8142 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,156B, BPFP=1.2983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 82,296B, BPFP=1.7761 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,804B, BPFP=1.0964 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,536B, BPFP=1.8244 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 225,196B, BPFP=1.2150 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.702s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.93044892 - layer.0.v_cache 0.00000028 0.00025363 - layer.1.k_cache 0.00287711 0.73933748 - layer.1.v_cache 0.00000077 0.00090671 - layer.2.k_cache 0.00119138 0.45530317 - layer.2.v_cache 0.00000112 0.00130821 - layer.3.k_cache 0.00131510 0.48983744 - layer.3.v_cache 0.00000211 0.00208342 - layer.4.k_cache 0.00352384 0.87614053 - layer.4.v_cache 0.00000335 0.00366630 - layer.4.output 0.00013081 0.05434853 - ------------------------------------------------------------------------------------- - TOTAL 0.00234480 0.97976285 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 870176 -BPFP 1.3414 bits/point -EBPFP 2.6828 equivalent bits/point -MSE 0.979763 ----------------------- -------------------------------------------------------- -Time: 1.134s Load: 0.019s, Pack+Encode: 0.414s, Decode+Unpack: 0.702s ----------------------- -------------------------------------------------------- -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 0.9798 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,188B, BPFP=0.3731 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,996B, BPFP=1.5968 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 43,008B, BPFP=1.0566 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,136B, BPFP=1.7231 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,604B, BPFP=1.2432 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,316B, BPFP=1.7521 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 52,136B, BPFP=1.2809 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,220B, BPFP=1.7251 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,700B, BPFP=1.0982 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,680B, BPFP=1.7610 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 206,080B, BPFP=1.2657 -⌛️ [2/4] FRONTEND: Frontend time: 0.371s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.594s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.02431886 - layer.0.v_cache 0.00000027 0.00024478 - layer.1.k_cache 0.00289381 0.71210816 - layer.1.v_cache 0.00000083 0.00091140 - layer.2.k_cache 0.00119129 0.45092913 - layer.2.v_cache 0.00000117 0.00134717 - layer.3.k_cache 0.00131631 0.49969516 - layer.3.v_cache 0.00000257 0.00218570 - layer.4.k_cache 0.00356231 0.85788751 - layer.4.v_cache 0.00000346 0.00376213 - layer.4.output 0.00015646 0.06474282 - ------------------------------------------------------------------------------------- - TOTAL 0.00245518 0.91516866 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 760064 -BPFP 1.3338 bits/point -EBPFP 2.6676 equivalent bits/point -MSE 0.915169 ----------------------- -------------------------------------------------------- -Time: 0.981s Load: 0.016s, Pack+Encode: 0.371s, Decode+Unpack: 0.594s ----------------------- -------------------------------------------------------- -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 0.9152 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,252B, BPFP=0.3484 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,536B, BPFP=1.6570 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,480B, BPFP=1.0618 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,100B, BPFP=1.8069 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,148B, BPFP=1.2598 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,508B, BPFP=1.8391 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,104B, BPFP=1.3045 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,840B, BPFP=1.8010 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,456B, BPFP=1.0841 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,876B, BPFP=1.8475 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,684B, BPFP=1.2375 -⌛️ [2/4] FRONTEND: Frontend time: 0.415s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.702s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.45384172 - layer.0.v_cache 0.00000026 0.00024104 - layer.1.k_cache 0.00288341 0.74378508 - layer.1.v_cache 0.00000082 0.00090626 - layer.2.k_cache 0.00117177 0.44883960 - layer.2.v_cache 0.00000117 0.00131761 - layer.3.k_cache 0.00130255 0.49691759 - layer.3.v_cache 0.00000215 0.00209456 - layer.4.k_cache 0.00351771 0.86350380 - layer.4.v_cache 0.00000337 0.00364070 - layer.4.output 0.00013439 0.05469188 - ------------------------------------------------------------------------------------- - TOTAL 0.00244940 0.94527539 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 829984 -BPFP 1.3543 bits/point -EBPFP 2.7085 equivalent bits/point -MSE 0.945275 ----------------------- -------------------------------------------------------- -Time: 1.136s Load: 0.018s, Pack+Encode: 0.415s, Decode+Unpack: 0.702s ----------------------- -------------------------------------------------------- -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 0.9453 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,420B, BPFP=0.3512 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,152B, BPFP=1.6662 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,696B, BPFP=1.0636 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,580B, BPFP=1.8126 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,584B, BPFP=1.2660 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,852B, BPFP=1.8416 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,268B, BPFP=1.3044 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,148B, BPFP=1.8028 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,952B, BPFP=1.0922 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,384B, BPFP=1.8537 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 219,692B, BPFP=1.2510 -⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.706s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.65127964 - layer.0.v_cache 0.00000027 0.00024255 - layer.1.k_cache 0.00288980 0.73911835 - layer.1.v_cache 0.00000078 0.00090970 - layer.2.k_cache 0.00118800 0.44703087 - layer.2.v_cache 0.00000116 0.00132169 - layer.3.k_cache 0.00131625 0.49987215 - layer.3.v_cache 0.00000215 0.00213192 - layer.4.k_cache 0.00354145 0.86873900 - layer.4.v_cache 0.00000355 0.00371043 - layer.4.output 0.00016614 0.06062435 - ------------------------------------------------------------------------------------- - TOTAL 0.00267732 0.96120383 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 836728 -BPFP 1.3613 bits/point -EBPFP 2.7226 equivalent bits/point -MSE 0.961204 ----------------------- -------------------------------------------------------- -Time: 1.140s Load: 0.018s, Pack+Encode: 0.416s, Decode+Unpack: 0.706s ----------------------- -------------------------------------------------------- -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 0.9612 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,852B, BPFP=0.3484 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,072B, BPFP=1.6909 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,920B, BPFP=1.0773 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,956B, BPFP=1.8289 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,408B, BPFP=1.2765 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,656B, BPFP=1.8688 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,392B, BPFP=1.3230 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,788B, BPFP=1.8250 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,940B, BPFP=1.1013 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,296B, BPFP=1.8838 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 209,468B, BPFP=1.2286 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (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, 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: 0.707s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.01152080 - layer.0.v_cache 0.00000026 0.00024868 - layer.1.k_cache 0.00292822 0.79393853 - layer.1.v_cache 0.00000083 0.00092145 - layer.2.k_cache 0.00115284 0.45802779 - layer.2.v_cache 0.00000118 0.00132312 - layer.3.k_cache 0.00132958 0.51149839 - layer.3.v_cache 0.00000212 0.00210334 - layer.4.k_cache 0.00351398 0.88904927 - layer.4.v_cache 0.00000364 0.00369886 - layer.4.output 0.00014544 0.05996067 - ------------------------------------------------------------------------------------- - TOTAL 0.00234040 0.99372664 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 815748 -BPFP 1.3670 bits/point -EBPFP 2.7340 equivalent bits/point -MSE 0.993727 ----------------------- -------------------------------------------------------- -Time: 1.131s Load: 0.017s, Pack+Encode: 0.407s, Decode+Unpack: 0.707s ----------------------- -------------------------------------------------------- -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 0.9937 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.020s - ------------------------------------------------------------- -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: 15,432B, BPFP=0.3387 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,676B, BPFP=1.6388 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,356B, BPFP=1.0612 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 81,144B, BPFP=1.7807 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,216B, BPFP=1.2556 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 82,412B, BPFP=1.8085 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,980B, BPFP=1.2943 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 81,124B, BPFP=1.7803 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 49,284B, BPFP=1.0815 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 83,220B, BPFP=1.8263 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 227,256B, BPFP=1.2468 -⌛️ [2/4] FRONTEND: Frontend time: 0.426s (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, 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: 0.704s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.12919017 - layer.0.v_cache 0.00000027 0.00024521 - layer.1.k_cache 0.00289550 0.77379651 - layer.1.v_cache 0.00000079 0.00091383 - layer.2.k_cache 0.00117641 0.45811124 - layer.2.v_cache 0.00000113 0.00132738 - layer.3.k_cache 0.00131994 0.50523402 - layer.3.v_cache 0.00000217 0.00216130 - layer.4.k_cache 0.00362269 0.89805886 - layer.4.v_cache 0.00000343 0.00376218 - layer.4.output 0.00016041 0.06062388 - ------------------------------------------------------------------------------------- - TOTAL 0.00246536 1.00109259 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 859100 -BPFP 1.3467 bits/point -EBPFP 2.6933 equivalent bits/point -MSE 1.001093 ----------------------- -------------------------------------------------------- -Time: 1.151s Load: 0.020s, Pack+Encode: 0.426s, Decode+Unpack: 0.704s ----------------------- -------------------------------------------------------- -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 1.0011 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,104B, BPFP=0.3565 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,572B, BPFP=1.6893 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,796B, BPFP=1.0809 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,676B, BPFP=1.8334 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,348B, BPFP=1.2828 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,512B, BPFP=1.8767 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,280B, BPFP=1.3284 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,548B, BPFP=1.8303 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,012B, BPFP=1.1096 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,684B, BPFP=1.8808 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,508B, BPFP=1.2598 -⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.719s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.47408421 - layer.0.v_cache 0.00000026 0.00023983 - layer.1.k_cache 0.00291335 0.77086227 - layer.1.v_cache 0.00000080 0.00091900 - layer.2.k_cache 0.00116981 0.45833599 - layer.2.v_cache 0.00000115 0.00131908 - layer.3.k_cache 0.00132745 0.50056245 - layer.3.v_cache 0.00000216 0.00209271 - layer.4.k_cache 0.00359183 0.90336927 - layer.4.v_cache 0.00000354 0.00364205 - layer.4.output 0.00014916 0.06161282 - ------------------------------------------------------------------------------------- - TOTAL 0.00234816 0.95441987 - (elements=4,745,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4745216 -Total Bytes 818040 -BPFP 1.3791 bits/point -EBPFP 2.7583 equivalent bits/point -MSE 0.954420 ----------------------- -------------------------------------------------------- -Time: 1.164s Load: 0.016s, Pack+Encode: 0.428s, Decode+Unpack: 0.719s ----------------------- -------------------------------------------------------- -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 0.9544 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,756B, BPFP=0.3568 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,540B, BPFP=1.6427 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,252B, BPFP=1.0474 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,080B, BPFP=1.7681 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,820B, BPFP=1.2414 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,232B, BPFP=1.8168 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,000B, BPFP=1.2908 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,032B, BPFP=1.7897 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,284B, BPFP=1.0707 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,068B, BPFP=1.8358 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 224,476B, BPFP=1.2708 -⌛️ [2/4] FRONTEND: Frontend time: 0.439s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.718s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.84030585 - layer.0.v_cache 0.00000026 0.00024457 - layer.1.k_cache 0.00293293 0.77204749 - layer.1.v_cache 0.00000076 0.00084854 - layer.2.k_cache 0.00115776 0.43926294 - layer.2.v_cache 0.00000114 0.00127957 - layer.3.k_cache 0.00132774 0.50370873 - layer.3.v_cache 0.00000222 0.00210780 - layer.4.k_cache 0.00355356 0.87392729 - layer.4.v_cache 0.00000324 0.00357864 - layer.4.output 0.00016540 0.06507686 - ------------------------------------------------------------------------------------- - TOTAL 0.00236485 0.97840135 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 836540 -BPFP 1.3531 bits/point -EBPFP 2.7062 equivalent bits/point -MSE 0.978401 ----------------------- -------------------------------------------------------- -Time: 1.175s Load: 0.017s, Pack+Encode: 0.439s, Decode+Unpack: 0.718s ----------------------- -------------------------------------------------------- -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 0.9784 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,596B, BPFP=0.3542 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,892B, BPFP=1.6554 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,772B, BPFP=1.0622 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,164B, BPFP=1.7979 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,320B, BPFP=1.2564 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,488B, BPFP=1.8279 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,428B, BPFP=1.3042 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,136B, BPFP=1.7972 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,984B, BPFP=1.0898 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,400B, BPFP=1.8487 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 220,220B, BPFP=1.2503 -⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.706s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.75221501 - layer.0.v_cache 0.00000026 0.00024246 - layer.1.k_cache 0.00287155 0.73145600 - layer.1.v_cache 0.00000078 0.00088707 - layer.2.k_cache 0.00117748 0.44063222 - layer.2.v_cache 0.00000114 0.00126957 - layer.3.k_cache 0.00132414 0.49054625 - layer.3.v_cache 0.00000215 0.00207070 - layer.4.k_cache 0.00352424 0.87061718 - layer.4.v_cache 0.00000349 0.00360480 - layer.4.output 0.00015726 0.06049038 - ------------------------------------------------------------------------------------- - TOTAL 0.00238109 0.96682163 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 836400 -BPFP 1.3568 bits/point -EBPFP 2.7136 equivalent bits/point -MSE 0.966822 ----------------------- -------------------------------------------------------- -Time: 1.142s Load: 0.018s, Pack+Encode: 0.417s, Decode+Unpack: 0.706s ----------------------- -------------------------------------------------------- -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 0.9668 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,376B, BPFP=0.3544 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,880B, BPFP=1.6796 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,364B, BPFP=1.0685 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,204B, BPFP=1.8253 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,020B, BPFP=1.2680 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,484B, BPFP=1.8548 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,904B, BPFP=1.3114 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,716B, BPFP=1.8141 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,492B, BPFP=1.0945 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,860B, BPFP=1.8635 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,896B, BPFP=1.2496 -⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.711s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.36925729 - layer.0.v_cache 0.00000026 0.00024630 - layer.1.k_cache 0.00285639 0.77510201 - layer.1.v_cache 0.00000078 0.00090750 - layer.2.k_cache 0.00115969 0.45259243 - layer.2.v_cache 0.00000117 0.00130745 - layer.3.k_cache 0.00131371 0.49804228 - layer.3.v_cache 0.00000219 0.00209335 - layer.4.k_cache 0.00358480 0.87965389 - layer.4.v_cache 0.00000371 0.00364890 - layer.4.output 0.00015316 0.06320274 - ------------------------------------------------------------------------------------- - TOTAL 0.00248089 0.94540445 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 830196 -BPFP 1.3666 bits/point -EBPFP 2.7332 equivalent bits/point -MSE 0.945404 ----------------------- -------------------------------------------------------- -Time: 1.162s Load: 0.017s, Pack+Encode: 0.434s, Decode+Unpack: 0.711s ----------------------- -------------------------------------------------------- -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 0.9454 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 13,896B, BPFP=0.3436 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,048B, BPFP=1.6082 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,648B, BPFP=1.0544 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 70,072B, BPFP=1.7324 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,388B, BPFP=1.2457 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,296B, BPFP=1.7627 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 52,008B, BPFP=1.2858 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 70,300B, BPFP=1.7380 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 44,156B, BPFP=1.0917 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 72,012B, BPFP=1.7804 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 203,928B, BPFP=1.2604 -⌛️ [2/4] FRONTEND: Frontend time: 0.376s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.593s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.62788410 - layer.0.v_cache 0.00000026 0.00024232 - layer.1.k_cache 0.00292725 0.73313701 - layer.1.v_cache 0.00000082 0.00089397 - layer.2.k_cache 0.00117417 0.44475633 - layer.2.v_cache 0.00000116 0.00132920 - layer.3.k_cache 0.00131215 0.49983655 - layer.3.v_cache 0.00000217 0.00213889 - layer.4.k_cache 0.00352248 0.86508642 - layer.4.v_cache 0.00000354 0.00372238 - layer.4.output 0.00015546 0.06275185 - ------------------------------------------------------------------------------------- - TOTAL 0.00240343 0.95928818 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 755752 -BPFP 1.3346 bits/point -EBPFP 2.6692 equivalent bits/point -MSE 0.959288 ----------------------- -------------------------------------------------------- -Time: 0.985s Load: 0.016s, Pack+Encode: 0.376s, Decode+Unpack: 0.593s ----------------------- -------------------------------------------------------- -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 0.9593 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,464B, BPFP=0.3522 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,944B, BPFP=1.6614 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,744B, BPFP=1.0647 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,392B, BPFP=1.8083 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,700B, BPFP=1.2687 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,720B, BPFP=1.8386 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,496B, BPFP=1.3096 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,120B, BPFP=1.8021 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,836B, BPFP=1.0896 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,324B, BPFP=1.8523 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 217,376B, BPFP=1.2378 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.65081840 - layer.0.v_cache 0.00000027 0.00024583 - layer.1.k_cache 0.00286131 0.73618492 - layer.1.v_cache 0.00000077 0.00090422 - layer.2.k_cache 0.00115688 0.45547855 - layer.2.v_cache 0.00000135 0.00133901 - layer.3.k_cache 0.00129955 0.48913410 - layer.3.v_cache 0.00000214 0.00211915 - layer.4.k_cache 0.00368945 0.87217112 - layer.4.v_cache 0.00000337 0.00368624 - layer.4.output 0.00015649 0.06184354 - ------------------------------------------------------------------------------------- - TOTAL 0.00237842 0.96138969 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 834116 -BPFP 1.3570 bits/point -EBPFP 2.7141 equivalent bits/point -MSE 0.961390 ----------------------- -------------------------------------------------------- -Time: 1.134s Load: 0.018s, Pack+Encode: 0.408s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9614 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,664B, BPFP=0.3514 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,444B, BPFP=1.6882 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,188B, BPFP=1.0590 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 75,664B, BPFP=1.8133 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 52,888B, BPFP=1.2674 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 77,408B, BPFP=1.8551 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,260B, BPFP=1.3243 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 75,788B, BPFP=1.8162 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,024B, BPFP=1.1030 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 78,236B, BPFP=1.8749 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 207,576B, BPFP=1.2436 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.692s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.36301314 - layer.0.v_cache 0.00000027 0.00025113 - layer.1.k_cache 0.00290854 0.75013827 - layer.1.v_cache 0.00000077 0.00087377 - layer.2.k_cache 0.00114310 0.44140789 - layer.2.v_cache 0.00000112 0.00126158 - layer.3.k_cache 0.00131875 0.49759847 - layer.3.v_cache 0.00000210 0.00203494 - layer.4.k_cache 0.00360283 0.86541374 - layer.4.v_cache 0.00000318 0.00352293 - layer.4.output 0.00014710 0.06164471 - ------------------------------------------------------------------------------------- - TOTAL 0.00242508 1.01229248 - (elements=4,673,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4673536 -Total Bytes 798140 -BPFP 1.3662 bits/point -EBPFP 2.7325 equivalent bits/point -MSE 1.012292 ----------------------- -------------------------------------------------------- -Time: 1.113s Load: 0.017s, Pack+Encode: 0.404s, Decode+Unpack: 0.692s ----------------------- -------------------------------------------------------- -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 1.0123 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,012B, BPFP=0.3501 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,328B, BPFP=1.6868 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,956B, BPFP=1.0717 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,468B, BPFP=1.8299 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,444B, BPFP=1.2697 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,636B, BPFP=1.8572 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,284B, BPFP=1.3126 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,008B, BPFP=1.8192 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,916B, BPFP=1.0941 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,496B, BPFP=1.8772 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 218,476B, BPFP=1.2738 -⌛️ [2/4] FRONTEND: Frontend time: 0.410s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.705s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.42914529 - layer.0.v_cache 0.00000026 0.00024594 - layer.1.k_cache 0.00291837 0.76533595 - layer.1.v_cache 0.00000081 0.00092696 - layer.2.k_cache 0.00116761 0.45346302 - layer.2.v_cache 0.00000118 0.00134730 - layer.3.k_cache 0.00133299 0.50947184 - layer.3.v_cache 0.00000217 0.00217957 - layer.4.k_cache 0.00352932 0.87924868 - layer.4.v_cache 0.00000340 0.00368091 - layer.4.output 0.00015509 0.06238653 - ------------------------------------------------------------------------------------- - TOTAL 0.00241467 1.02104226 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 826024 -BPFP 1.3760 bits/point -EBPFP 2.7519 equivalent bits/point -MSE 1.021042 ----------------------- -------------------------------------------------------- -Time: 1.132s Load: 0.017s, Pack+Encode: 0.410s, Decode+Unpack: 0.705s ----------------------- -------------------------------------------------------- -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 1.0210 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,432B, BPFP=0.3480 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 69,280B, BPFP=1.6705 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,852B, BPFP=1.0815 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 75,504B, BPFP=1.8206 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,624B, BPFP=1.2930 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 77,196B, BPFP=1.8614 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,224B, BPFP=1.3316 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 75,652B, BPFP=1.8242 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 45,844B, BPFP=1.1054 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 77,832B, BPFP=1.8767 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 209,060B, BPFP=1.2602 -⌛️ [2/4] FRONTEND: Frontend time: 0.405s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.702s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.65203933 - layer.0.v_cache 0.00000026 0.00024266 - layer.1.k_cache 0.00295515 0.77450901 - layer.1.v_cache 0.00000079 0.00088686 - layer.2.k_cache 0.00117180 0.45952644 - layer.2.v_cache 0.00000115 0.00131593 - layer.3.k_cache 0.00132196 0.50316295 - layer.3.v_cache 0.00000215 0.00213629 - layer.4.k_cache 0.00355989 0.86800742 - layer.4.v_cache 0.00000355 0.00368466 - layer.4.output 0.00015612 0.06242429 - ------------------------------------------------------------------------------------- - TOTAL 0.00234089 1.03680062 - (elements=4,644,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4644864 -Total Bytes 798500 -BPFP 1.3753 bits/point -EBPFP 2.7506 equivalent bits/point -MSE 1.036801 ----------------------- -------------------------------------------------------- -Time: 1.123s Load: 0.016s, Pack+Encode: 0.405s, Decode+Unpack: 0.702s ----------------------- -------------------------------------------------------- -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 1.0368 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,016B, BPFP=0.3534 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,752B, BPFP=1.6884 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,820B, BPFP=1.0782 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,228B, BPFP=1.8408 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,432B, BPFP=1.2809 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,632B, BPFP=1.8739 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,272B, BPFP=1.3242 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,948B, BPFP=1.8342 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,260B, BPFP=1.1121 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,352B, BPFP=1.8908 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,200B, BPFP=1.2719 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.700s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.70470383 - layer.0.v_cache 0.00000026 0.00024252 - layer.1.k_cache 0.00288168 0.77586466 - layer.1.v_cache 0.00000079 0.00091686 - layer.2.k_cache 0.00118772 0.46075242 - layer.2.v_cache 0.00000117 0.00137895 - layer.3.k_cache 0.00132341 0.50989229 - layer.3.v_cache 0.00000218 0.00217908 - layer.4.k_cache 0.00354949 0.89221201 - layer.4.v_cache 0.00000378 0.00384386 - layer.4.output 0.00018345 0.06354609 - ------------------------------------------------------------------------------------- - TOTAL 0.00237789 0.97186934 - (elements=4,759,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4759552 -Total Bytes 822912 -BPFP 1.3832 bits/point -EBPFP 2.7664 equivalent bits/point -MSE 0.971869 ----------------------- -------------------------------------------------------- -Time: 1.129s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.700s ----------------------- -------------------------------------------------------- -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 0.9719 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,156B, BPFP=0.3467 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,880B, BPFP=1.5889 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,100B, BPFP=1.0311 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,816B, BPFP=1.7098 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,892B, BPFP=1.2219 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,832B, BPFP=1.7347 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,268B, BPFP=1.2556 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,432B, BPFP=1.7004 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,340B, BPFP=1.0614 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,132B, BPFP=1.7421 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 195,400B, BPFP=1.1964 -⌛️ [2/4] FRONTEND: Frontend time: 0.357s (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, 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: 0.609s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.34243133 - layer.0.v_cache 0.00000027 0.00024613 - layer.1.k_cache 0.00287626 0.65069054 - layer.1.v_cache 0.00000082 0.00091227 - layer.2.k_cache 0.00122126 0.43895088 - layer.2.v_cache 0.00000114 0.00130552 - layer.3.k_cache 0.00133159 0.48200276 - layer.3.v_cache 0.00000213 0.00207382 - layer.4.k_cache 0.00352266 0.84325320 - layer.4.v_cache 0.00000325 0.00361473 - layer.4.output 0.00013715 0.04906865 - ------------------------------------------------------------------------------------- - TOTAL 0.00242081 0.85441113 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 742248 -BPFP 1.2984 bits/point -EBPFP 2.5969 equivalent bits/point -MSE 0.854411 ----------------------- -------------------------------------------------------- -Time: 0.982s Load: 0.017s, Pack+Encode: 0.357s, Decode+Unpack: 0.609s ----------------------- -------------------------------------------------------- -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 0.8544 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 15,192B, BPFP=0.3543 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,064B, BPFP=1.6806 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,680B, BPFP=1.0653 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,172B, BPFP=1.8230 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,460B, BPFP=1.2701 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,476B, BPFP=1.8535 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,972B, BPFP=1.3053 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,788B, BPFP=1.8141 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,112B, BPFP=1.0987 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,348B, BPFP=1.8738 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 211,088B, BPFP=1.2307 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.47039267 - layer.0.v_cache 0.00000026 0.00024268 - layer.1.k_cache 0.00283722 0.74617515 - layer.1.v_cache 0.00000078 0.00090438 - layer.2.k_cache 0.00117890 0.45279609 - layer.2.v_cache 0.00000113 0.00128285 - layer.3.k_cache 0.00131429 0.49899930 - layer.3.v_cache 0.00000212 0.00208966 - layer.4.k_cache 0.00357066 0.88549513 - layer.4.v_cache 0.00000341 0.00364831 - layer.4.output 0.00013949 0.05780895 - ------------------------------------------------------------------------------------- - TOTAL 0.00233536 0.94951872 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 817352 -BPFP 1.3615 bits/point -EBPFP 2.7231 equivalent bits/point -MSE 0.949519 ----------------------- -------------------------------------------------------- -Time: 1.133s Load: 0.017s, Pack+Encode: 0.408s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9495 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,876B, BPFP=0.3504 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,996B, BPFP=1.6551 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,048B, BPFP=1.0383 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 80,848B, BPFP=1.7843 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,224B, BPFP=1.2408 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,992B, BPFP=1.8095 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,288B, BPFP=1.2864 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,544B, BPFP=1.7775 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 49,004B, BPFP=1.0815 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 82,760B, BPFP=1.8264 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 227,596B, BPFP=1.2557 -⌛️ [2/4] FRONTEND: Frontend time: 0.427s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.705s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.35713429 - layer.0.v_cache 0.00000027 0.00025197 - layer.1.k_cache 0.00288960 0.73656296 - layer.1.v_cache 0.00000085 0.00092612 - layer.2.k_cache 0.00116959 0.44924970 - layer.2.v_cache 0.00000115 0.00132476 - layer.3.k_cache 0.00132243 0.50389224 - layer.3.v_cache 0.00000218 0.00213993 - layer.4.k_cache 0.00355194 0.88428364 - layer.4.v_cache 0.00000329 0.00374530 - layer.4.output 0.00015235 0.05674500 - ------------------------------------------------------------------------------------- - TOTAL 0.00245037 0.94046364 - (elements=5,074,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5074944 -Total Bytes 855176 -BPFP 1.3481 bits/point -EBPFP 2.6962 equivalent bits/point -MSE 0.940464 ----------------------- -------------------------------------------------------- -Time: 1.151s Load: 0.019s, Pack+Encode: 0.427s, Decode+Unpack: 0.705s ----------------------- -------------------------------------------------------- -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 0.9405 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,832B, BPFP=0.3544 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,992B, BPFP=1.6961 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,324B, BPFP=1.0829 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,648B, BPFP=1.8312 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,848B, BPFP=1.2865 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,432B, BPFP=1.8739 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,716B, BPFP=1.3311 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,748B, BPFP=1.8336 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,560B, BPFP=1.1124 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,160B, BPFP=1.8912 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,000B, BPFP=1.2901 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.699s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.98811550 - layer.0.v_cache 0.00000027 0.00024056 - layer.1.k_cache 0.00290162 0.75310430 - layer.1.v_cache 0.00000083 0.00091775 - layer.2.k_cache 0.00115956 0.44900312 - layer.2.v_cache 0.00000120 0.00132247 - layer.3.k_cache 0.00131802 0.50829486 - layer.3.v_cache 0.00000223 0.00210787 - layer.4.k_cache 0.00351161 0.86577884 - layer.4.v_cache 0.00000355 0.00366566 - layer.4.output 0.00016630 0.06556349 - ------------------------------------------------------------------------------------- - TOTAL 0.00247233 0.98820035 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 814260 -BPFP 1.3896 bits/point -EBPFP 2.7791 equivalent bits/point -MSE 0.988200 ----------------------- -------------------------------------------------------- -Time: 1.129s Load: 0.017s, Pack+Encode: 0.412s, Decode+Unpack: 0.699s ----------------------- -------------------------------------------------------- -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 0.9882 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,200B, BPFP=0.3544 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,436B, BPFP=1.6333 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,004B, BPFP=1.0484 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,784B, BPFP=1.7418 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,056B, BPFP=1.2494 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,020B, BPFP=1.7727 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,596B, BPFP=1.2878 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,944B, BPFP=1.7458 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,468B, BPFP=1.0850 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,684B, BPFP=1.7892 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 204,264B, BPFP=1.2746 -⌛️ [2/4] FRONTEND: Frontend time: 0.353s (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, 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: 0.608s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.37968937 - layer.0.v_cache 0.00000027 0.00023922 - layer.1.k_cache 0.00295685 0.73234085 - layer.1.v_cache 0.00000081 0.00084338 - layer.2.k_cache 0.00117427 0.44362248 - layer.2.v_cache 0.00000111 0.00127190 - layer.3.k_cache 0.00131811 0.49852467 - layer.3.v_cache 0.00000214 0.00204779 - layer.4.k_cache 0.00360037 0.85110298 - layer.4.v_cache 0.00000322 0.00359015 - layer.4.output 0.00015721 0.06074550 - ------------------------------------------------------------------------------------- - TOTAL 0.00238240 0.93973249 - (elements=4,487,168) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4487168 -Total Bytes 753456 -BPFP 1.3433 bits/point -EBPFP 2.6866 equivalent bits/point -MSE 0.939732 ----------------------- -------------------------------------------------------- -Time: 0.977s Load: 0.017s, Pack+Encode: 0.353s, Decode+Unpack: 0.608s ----------------------- -------------------------------------------------------- -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 0.9397 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,592B, BPFP=0.3465 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,712B, BPFP=1.6791 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,884B, BPFP=1.0658 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,976B, BPFP=1.8279 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,656B, BPFP=1.2741 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,492B, BPFP=1.8639 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,552B, BPFP=1.3191 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,784B, BPFP=1.8233 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,652B, BPFP=1.1078 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,280B, BPFP=1.8826 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,720B, BPFP=1.2747 -⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.698s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.61671629 - layer.0.v_cache 0.00000027 0.00025288 - layer.1.k_cache 0.00282437 0.74852518 - layer.1.v_cache 0.00000081 0.00092450 - layer.2.k_cache 0.00117800 0.45773812 - layer.2.v_cache 0.00000117 0.00134093 - layer.3.k_cache 0.00131145 0.51532256 - layer.3.v_cache 0.00000228 0.00219189 - layer.4.k_cache 0.00354228 0.88947609 - layer.4.v_cache 0.00000348 0.00382578 - layer.4.output 0.00016143 0.06445572 - ------------------------------------------------------------------------------------- - TOTAL 0.00241866 0.96386694 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 812300 -BPFP 1.3778 bits/point -EBPFP 2.7556 equivalent bits/point -MSE 0.963867 ----------------------- -------------------------------------------------------- -Time: 1.132s Load: 0.016s, Pack+Encode: 0.418s, Decode+Unpack: 0.698s ----------------------- -------------------------------------------------------- -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 0.9639 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,840B, BPFP=0.3513 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,324B, BPFP=1.6885 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,088B, BPFP=1.0674 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,848B, BPFP=1.8430 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,164B, BPFP=1.2823 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,080B, BPFP=1.8722 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,016B, BPFP=1.3261 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,448B, BPFP=1.8335 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,968B, BPFP=1.1119 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,828B, BPFP=1.8899 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,496B, BPFP=1.2636 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (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, 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: 0.715s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.88076172 - layer.0.v_cache 0.00000026 0.00024935 - layer.1.k_cache 0.00286502 0.76027046 - layer.1.v_cache 0.00000078 0.00090881 - layer.2.k_cache 0.00115867 0.46048478 - layer.2.v_cache 0.00000116 0.00130820 - layer.3.k_cache 0.00132713 0.50964087 - layer.3.v_cache 0.00000212 0.00211343 - layer.4.k_cache 0.00352212 0.87849713 - layer.4.v_cache 0.00000405 0.00381200 - layer.4.output 0.00014449 0.06261738 - ------------------------------------------------------------------------------------- - TOTAL 0.00233809 0.98203688 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 816100 -BPFP 1.3800 bits/point -EBPFP 2.7601 equivalent bits/point -MSE 0.982037 ----------------------- -------------------------------------------------------- -Time: 1.145s Load: 0.018s, Pack+Encode: 0.413s, Decode+Unpack: 0.715s ----------------------- -------------------------------------------------------- -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 0.9820 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 12,980B, BPFP=0.3169 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,408B, BPFP=1.5725 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,028B, BPFP=1.0017 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,416B, BPFP=1.6947 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,252B, BPFP=1.2024 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,680B, BPFP=1.7256 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 50,480B, BPFP=1.2324 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,408B, BPFP=1.6945 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 42,600B, BPFP=1.0400 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,244B, BPFP=1.7394 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 198,988B, BPFP=1.2145 -⌛️ [2/4] FRONTEND: Frontend time: 0.366s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.605s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.48141174 - layer.0.v_cache 0.00000026 0.00023965 - layer.1.k_cache 0.00291374 0.65436597 - layer.1.v_cache 0.00000082 0.00088658 - layer.2.k_cache 0.00121384 0.43350964 - layer.2.v_cache 0.00000117 0.00132430 - layer.3.k_cache 0.00132315 0.49105544 - layer.3.v_cache 0.00000219 0.00213620 - layer.4.k_cache 0.00353222 0.82572603 - layer.4.v_cache 0.00000326 0.00371738 - layer.4.output 0.00015660 0.05364517 - ------------------------------------------------------------------------------------- - TOTAL 0.00244527 0.86492526 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 740484 -BPFP 1.2913 bits/point -EBPFP 2.5826 equivalent bits/point -MSE 0.864925 ----------------------- -------------------------------------------------------- -Time: 0.987s Load: 0.016s, Pack+Encode: 0.366s, Decode+Unpack: 0.605s ----------------------- -------------------------------------------------------- -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 0.8649 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,292B, BPFP=0.3524 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,596B, BPFP=1.6730 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,584B, BPFP=1.0736 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,132B, BPFP=1.8237 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,044B, BPFP=1.2685 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,468B, BPFP=1.8544 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,844B, BPFP=1.3100 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,904B, BPFP=1.8184 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,656B, BPFP=1.0983 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,132B, BPFP=1.8697 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,240B, BPFP=1.2401 -⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.721s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.31752685 - layer.0.v_cache 0.00000026 0.00024261 - layer.1.k_cache 0.00286983 0.77894750 - layer.1.v_cache 0.00000081 0.00091676 - layer.2.k_cache 0.00115809 0.44531754 - layer.2.v_cache 0.00000117 0.00133325 - layer.3.k_cache 0.00131372 0.50813379 - layer.3.v_cache 0.00000240 0.00215226 - layer.4.k_cache 0.00359509 0.88173691 - layer.4.v_cache 0.00000368 0.00378202 - layer.4.output 0.00014345 0.05885646 - ------------------------------------------------------------------------------------- - TOTAL 0.00238822 0.94110824 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 828892 -BPFP 1.3645 bits/point -EBPFP 2.7289 equivalent bits/point -MSE 0.941108 ----------------------- -------------------------------------------------------- -Time: 1.163s Load: 0.018s, Pack+Encode: 0.424s, Decode+Unpack: 0.721s ----------------------- -------------------------------------------------------- -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 0.9411 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 13,832B, BPFP=0.3441 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 65,040B, BPFP=1.6182 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,920B, BPFP=1.0430 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,916B, BPFP=1.7396 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,748B, BPFP=1.2378 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,944B, BPFP=1.7651 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,324B, BPFP=1.2770 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,836B, BPFP=1.7376 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,600B, BPFP=1.0848 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,540B, BPFP=1.7800 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 200,104B, BPFP=1.2447 -⌛️ [2/4] FRONTEND: Frontend time: 0.385s (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, 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: 0.592s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.97583708 - layer.0.v_cache 0.00000027 0.00025449 - layer.1.k_cache 0.00292133 0.75670361 - layer.1.v_cache 0.00000080 0.00091251 - layer.2.k_cache 0.00120652 0.46115035 - layer.2.v_cache 0.00000113 0.00131702 - layer.3.k_cache 0.00133526 0.49652532 - layer.3.v_cache 0.00000215 0.00209463 - layer.4.k_cache 0.00353655 0.88895373 - layer.4.v_cache 0.00000336 0.00368072 - layer.4.output 0.00016938 0.06174529 - ------------------------------------------------------------------------------------- - TOTAL 0.00239216 0.98817218 - (elements=4,501,504) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4501504 -Total Bytes 747804 -BPFP 1.3290 bits/point -EBPFP 2.6580 equivalent bits/point -MSE 0.988172 ----------------------- -------------------------------------------------------- -Time: 0.995s Load: 0.018s, Pack+Encode: 0.385s, Decode+Unpack: 0.592s ----------------------- -------------------------------------------------------- -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 0.9882 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,916B, BPFP=0.3615 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,752B, BPFP=1.6750 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,776B, BPFP=1.0623 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,768B, BPFP=1.8116 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,872B, BPFP=1.2689 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,052B, BPFP=1.8408 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,560B, BPFP=1.3072 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,740B, BPFP=1.8110 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,036B, BPFP=1.0909 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,848B, BPFP=1.8588 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,764B, BPFP=1.2307 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.61748984 - layer.0.v_cache 0.00000027 0.00024475 - layer.1.k_cache 0.00284490 0.73100157 - layer.1.v_cache 0.00000080 0.00092078 - layer.2.k_cache 0.00115895 0.44559381 - layer.2.v_cache 0.00000119 0.00134337 - layer.3.k_cache 0.00130300 0.49448244 - layer.3.v_cache 0.00000215 0.00212867 - layer.4.k_cache 0.00352154 0.85153331 - layer.4.v_cache 0.00000376 0.00377002 - layer.4.output 0.00013131 0.05782477 - ------------------------------------------------------------------------------------- - TOTAL 0.00239946 0.95570054 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 837084 -BPFP 1.3579 bits/point -EBPFP 2.7158 equivalent bits/point -MSE 0.955701 ----------------------- -------------------------------------------------------- -Time: 1.136s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9557 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.016s - ------------------------------------------------------------- -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: 14,024B, BPFP=0.3557 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,912B, BPFP=1.6465 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,664B, BPFP=1.0568 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,784B, BPFP=1.7701 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,404B, BPFP=1.2531 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,928B, BPFP=1.7991 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 50,996B, BPFP=1.2935 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,776B, BPFP=1.7699 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,068B, BPFP=1.0924 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,472B, BPFP=1.8129 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 202,820B, BPFP=1.2861 -⌛️ [2/4] FRONTEND: Frontend time: 0.378s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.600s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.53923926 - layer.0.v_cache 0.00000027 0.00024691 - layer.1.k_cache 0.00291045 0.74161217 - layer.1.v_cache 0.00000082 0.00090449 - layer.2.k_cache 0.00116968 0.44591195 - layer.2.v_cache 0.00000115 0.00129815 - layer.3.k_cache 0.00131855 0.51341461 - layer.3.v_cache 0.00000224 0.00215418 - layer.4.k_cache 0.00354502 0.87381051 - layer.4.v_cache 0.00000324 0.00367904 - layer.4.output 0.00015550 0.06442279 - ------------------------------------------------------------------------------------- - TOTAL 0.00235900 0.95571160 - (elements=4,415,488) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4415488 -Total Bytes 748848 -BPFP 1.3568 bits/point -EBPFP 2.7135 equivalent bits/point -MSE 0.955712 ----------------------- -------------------------------------------------------- -Time: 0.994s Load: 0.016s, Pack+Encode: 0.378s, Decode+Unpack: 0.600s ----------------------- -------------------------------------------------------- -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 0.9557 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,152B, BPFP=0.3492 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,460B, BPFP=1.6699 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,880B, BPFP=1.0573 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,580B, BPFP=1.8109 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,296B, BPFP=1.2513 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,168B, BPFP=1.8475 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,684B, BPFP=1.3063 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,716B, BPFP=1.8141 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,640B, BPFP=1.0749 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,880B, BPFP=1.8639 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 218,804B, BPFP=1.2606 -⌛️ [2/4] FRONTEND: Frontend time: 0.415s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.703s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.29213766 - layer.0.v_cache 0.00000026 0.00024920 - layer.1.k_cache 0.00294413 0.76421687 - layer.1.v_cache 0.00000079 0.00090859 - layer.2.k_cache 0.00117757 0.45092647 - layer.2.v_cache 0.00000116 0.00133843 - layer.3.k_cache 0.00130963 0.49989999 - layer.3.v_cache 0.00000218 0.00219701 - layer.4.k_cache 0.00367572 0.89208390 - layer.4.v_cache 0.00000325 0.00367370 - layer.4.output 0.00014293 0.05906840 - ------------------------------------------------------------------------------------- - TOTAL 0.00248846 1.01027896 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 828260 -BPFP 1.3634 bits/point -EBPFP 2.7268 equivalent bits/point -MSE 1.010279 ----------------------- -------------------------------------------------------- -Time: 1.136s Load: 0.018s, Pack+Encode: 0.415s, Decode+Unpack: 0.703s ----------------------- -------------------------------------------------------- -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 1.0103 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,508B, BPFP=0.3413 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 75,084B, BPFP=1.6524 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,812B, BPFP=1.0522 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 81,468B, BPFP=1.7929 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,408B, BPFP=1.2634 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 82,540B, BPFP=1.8165 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,156B, BPFP=1.3018 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,968B, BPFP=1.7819 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 49,652B, BPFP=1.0927 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 83,144B, BPFP=1.8298 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 219,560B, BPFP=1.2080 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (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, 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: 0.693s - -[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, 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 10.03719053 - layer.0.v_cache 0.00000027 0.00024508 - layer.1.k_cache 0.00283684 0.75421641 - layer.1.v_cache 0.00000080 0.00093334 - layer.2.k_cache 0.00116829 0.45621957 - layer.2.v_cache 0.00000120 0.00136048 - layer.3.k_cache 0.00130356 0.49600602 - layer.3.v_cache 0.00000219 0.00214394 - layer.4.k_cache 0.00355311 0.87046612 - layer.4.v_cache 0.00000352 0.00372366 - layer.4.output 0.00013754 0.05445884 - ------------------------------------------------------------------------------------- - TOTAL 0.00241478 0.91716718 - (elements=5,089,280) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5089280 -Total Bytes 852300 -BPFP 1.3398 bits/point -EBPFP 2.6795 equivalent bits/point -MSE 0.917167 ----------------------- -------------------------------------------------------- -Time: 1.125s Load: 0.019s, Pack+Encode: 0.413s, Decode+Unpack: 0.693s ----------------------- -------------------------------------------------------- -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 0.9172 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,328B, BPFP=0.3585 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,012B, BPFP=1.6844 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,608B, BPFP=1.0668 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,260B, BPFP=1.8306 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,384B, BPFP=1.2721 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,716B, BPFP=1.8646 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,108B, BPFP=1.3124 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,748B, BPFP=1.8186 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,972B, BPFP=1.0987 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,420B, BPFP=1.8811 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,260B, BPFP=1.2529 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.694s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.42357302 - layer.0.v_cache 0.00000026 0.00024966 - layer.1.k_cache 0.00284576 0.76982208 - layer.1.v_cache 0.00000078 0.00091921 - layer.2.k_cache 0.00117801 0.45649674 - layer.2.v_cache 0.00000115 0.00130222 - layer.3.k_cache 0.00132187 0.49947978 - layer.3.v_cache 0.00000218 0.00216971 - layer.4.k_cache 0.00348374 0.88182379 - layer.4.v_cache 0.00000337 0.00382814 - layer.4.output 0.00014793 0.06008313 - ------------------------------------------------------------------------------------- - TOTAL 0.00241951 0.94857120 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 820816 -BPFP 1.3714 bits/point -EBPFP 2.7428 equivalent bits/point -MSE 0.948571 ----------------------- -------------------------------------------------------- -Time: 1.120s Load: 0.018s, Pack+Encode: 0.408s, Decode+Unpack: 0.694s ----------------------- -------------------------------------------------------- -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 0.9486 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,452B, BPFP=0.3530 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,588B, BPFP=1.6582 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,568B, BPFP=1.0638 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,124B, BPFP=1.8075 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,336B, BPFP=1.2641 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,216B, BPFP=1.8324 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,200B, BPFP=1.3067 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,772B, BPFP=1.7994 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,568B, BPFP=1.0866 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,920B, BPFP=1.8485 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 209,720B, BPFP=1.1977 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.710s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.70354975 - layer.0.v_cache 0.00000027 0.00024807 - layer.1.k_cache 0.00293320 0.76020251 - layer.1.v_cache 0.00000079 0.00093924 - layer.2.k_cache 0.00117175 0.46294706 - layer.2.v_cache 0.00000114 0.00134151 - layer.3.k_cache 0.00133100 0.50485073 - layer.3.v_cache 0.00000223 0.00211041 - layer.4.k_cache 0.00354789 0.90082510 - layer.4.v_cache 0.00000344 0.00368463 - layer.4.output 0.00015145 0.05930864 - ------------------------------------------------------------------------------------- - TOTAL 0.00242100 0.96985240 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 823464 -BPFP 1.3436 bits/point -EBPFP 2.6873 equivalent bits/point -MSE 0.969852 ----------------------- -------------------------------------------------------- -Time: 1.131s Load: 0.017s, Pack+Encode: 0.404s, Decode+Unpack: 0.710s ----------------------- -------------------------------------------------------- -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 0.9699 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,084B, BPFP=0.3527 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,932B, BPFP=1.6259 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,888B, BPFP=1.0489 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,744B, BPFP=1.7464 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,872B, BPFP=1.2488 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,048B, BPFP=1.7790 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,480B, BPFP=1.2891 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,856B, BPFP=1.7492 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,120B, BPFP=1.0797 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,424B, BPFP=1.7885 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 201,068B, BPFP=1.2587 -⌛️ [2/4] FRONTEND: Frontend time: 0.357s (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, 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: 0.595s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.39945788 - layer.0.v_cache 0.00000026 0.00024747 - layer.1.k_cache 0.00299662 0.76026163 - layer.1.v_cache 0.00000081 0.00089136 - layer.2.k_cache 0.00115233 0.45649744 - layer.2.v_cache 0.00000114 0.00133312 - layer.3.k_cache 0.00133293 0.50842774 - layer.3.v_cache 0.00000220 0.00214887 - layer.4.k_cache 0.00362032 0.89274519 - layer.4.v_cache 0.00000325 0.00363855 - layer.4.output 0.00015781 0.06490176 - ------------------------------------------------------------------------------------- - TOTAL 0.00240738 0.94894688 - (elements=4,472,832) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4472832 -Total Bytes 748516 -BPFP 1.3388 bits/point -EBPFP 2.6776 equivalent bits/point -MSE 0.948947 ----------------------- -------------------------------------------------------- -Time: 0.969s Load: 0.017s, Pack+Encode: 0.357s, Decode+Unpack: 0.595s ----------------------- -------------------------------------------------------- -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 0.9489 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.020s - ------------------------------------------------------------- -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: 15,616B, BPFP=0.3476 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,100B, BPFP=1.6493 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,132B, BPFP=1.0491 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 80,424B, BPFP=1.7901 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,184B, BPFP=1.2505 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,468B, BPFP=1.8133 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,848B, BPFP=1.2876 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,952B, BPFP=1.7796 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,836B, BPFP=1.0870 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 82,084B, BPFP=1.8270 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 220,728B, BPFP=1.2282 -⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.707s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.12536030 - layer.0.v_cache 0.00000027 0.00024729 - layer.1.k_cache 0.00294291 0.73921695 - layer.1.v_cache 0.00000079 0.00092504 - layer.2.k_cache 0.00116231 0.44715981 - layer.2.v_cache 0.00000118 0.00131120 - layer.3.k_cache 0.00133045 0.49435573 - layer.3.v_cache 0.00000222 0.00213482 - layer.4.k_cache 0.00357138 0.88170256 - layer.4.v_cache 0.00000341 0.00368128 - layer.4.output 0.00014419 0.06058401 - ------------------------------------------------------------------------------------- - TOTAL 0.00242500 0.92417364 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 844372 -BPFP 1.3424 bits/point -EBPFP 2.6848 equivalent bits/point -MSE 0.924174 ----------------------- -------------------------------------------------------- -Time: 1.145s Load: 0.020s, Pack+Encode: 0.419s, Decode+Unpack: 0.707s ----------------------- -------------------------------------------------------- -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 0.9242 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 14,392B, BPFP=0.3481 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 69,140B, BPFP=1.6723 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,492B, BPFP=1.0761 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 75,432B, BPFP=1.8245 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,020B, BPFP=1.2824 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 76,540B, BPFP=1.8513 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,064B, BPFP=1.3318 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 75,072B, BPFP=1.8158 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 45,716B, BPFP=1.1057 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 76,980B, BPFP=1.8619 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 202,228B, BPFP=1.2228 -⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.705s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.53308370 - layer.0.v_cache 0.00000026 0.00024520 - layer.1.k_cache 0.00287105 0.74355394 - layer.1.v_cache 0.00000078 0.00091680 - layer.2.k_cache 0.00117289 0.45880599 - layer.2.v_cache 0.00000112 0.00131402 - layer.3.k_cache 0.00130124 0.50230738 - layer.3.v_cache 0.00000213 0.00207568 - layer.4.k_cache 0.00357998 0.89228164 - layer.4.v_cache 0.00000342 0.00368515 - layer.4.output 0.00014651 0.05858838 - ------------------------------------------------------------------------------------- - TOTAL 0.00235363 0.95518736 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 788076 -BPFP 1.3615 bits/point -EBPFP 2.7231 equivalent bits/point -MSE 0.955187 ----------------------- -------------------------------------------------------- -Time: 1.159s Load: 0.017s, Pack+Encode: 0.436s, Decode+Unpack: 0.705s ----------------------- -------------------------------------------------------- -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 0.9552 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.020s - ------------------------------------------------------------- -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: 15,720B, BPFP=0.3450 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 75,248B, BPFP=1.6513 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 48,108B, BPFP=1.0557 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 81,252B, BPFP=1.7831 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 57,416B, BPFP=1.2600 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 82,380B, BPFP=1.8078 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 59,588B, BPFP=1.3077 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,884B, BPFP=1.7750 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 49,560B, BPFP=1.0876 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 83,072B, BPFP=1.8230 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 225,388B, BPFP=1.2365 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.694s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.09221521 - layer.0.v_cache 0.00000027 0.00025464 - layer.1.k_cache 0.00294478 0.75687331 - layer.1.v_cache 0.00000080 0.00092509 - layer.2.k_cache 0.00115317 0.46111520 - layer.2.v_cache 0.00000113 0.00131550 - layer.3.k_cache 0.00130758 0.51474565 - layer.3.v_cache 0.00000212 0.00212590 - layer.4.k_cache 0.00357902 0.89819130 - layer.4.v_cache 0.00000333 0.00372700 - layer.4.output 0.00014592 0.06288905 - ------------------------------------------------------------------------------------- - TOTAL 0.00245352 0.99878893 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 858616 -BPFP 1.3459 bits/point -EBPFP 2.6918 equivalent bits/point -MSE 0.998789 ----------------------- -------------------------------------------------------- -Time: 1.125s Load: 0.020s, Pack+Encode: 0.411s, Decode+Unpack: 0.694s ----------------------- -------------------------------------------------------- -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 0.9988 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,696B, BPFP=0.3586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,904B, BPFP=1.6654 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,952B, BPFP=1.0497 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,064B, BPFP=1.8061 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,288B, BPFP=1.2630 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,248B, BPFP=1.8332 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,072B, BPFP=1.3037 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,888B, BPFP=1.8021 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,608B, BPFP=1.0875 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,212B, BPFP=1.8552 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,252B, BPFP=1.2350 -⌛️ [2/4] FRONTEND: Frontend time: 0.431s (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, 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: 0.700s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.01731542 - layer.0.v_cache 0.00000027 0.00025184 - layer.1.k_cache 0.00288942 0.74966038 - layer.1.v_cache 0.00000084 0.00091664 - layer.2.k_cache 0.00116702 0.45311862 - layer.2.v_cache 0.00000122 0.00131481 - layer.3.k_cache 0.00131477 0.49633463 - layer.3.v_cache 0.00000216 0.00210856 - layer.4.k_cache 0.00350203 0.87282674 - layer.4.v_cache 0.00000356 0.00372239 - layer.4.output 0.00013713 0.06060513 - ------------------------------------------------------------------------------------- - TOTAL 0.00228853 0.98857075 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 830184 -BPFP 1.3546 bits/point -EBPFP 2.7092 equivalent bits/point -MSE 0.988571 ----------------------- -------------------------------------------------------- -Time: 1.148s Load: 0.018s, Pack+Encode: 0.431s, Decode+Unpack: 0.700s ----------------------- -------------------------------------------------------- -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 0.9886 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 16,040B, BPFP=0.3481 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 76,772B, BPFP=1.6661 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 49,060B, BPFP=1.0647 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 82,672B, BPFP=1.7941 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 58,260B, BPFP=1.2643 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 83,532B, BPFP=1.8128 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 60,500B, BPFP=1.3129 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 82,252B, BPFP=1.7850 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 50,864B, BPFP=1.1038 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 84,656B, BPFP=1.8372 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 232,136B, BPFP=1.2594 -⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.721s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.38312853 - layer.0.v_cache 0.00000028 0.00025382 - layer.1.k_cache 0.00285409 0.72835804 - layer.1.v_cache 0.00000079 0.00092312 - layer.2.k_cache 0.00117169 0.45452109 - layer.2.v_cache 0.00000113 0.00131504 - layer.3.k_cache 0.00130395 0.50401086 - layer.3.v_cache 0.00000220 0.00218731 - layer.4.k_cache 0.00353751 0.86841914 - layer.4.v_cache 0.00000341 0.00386257 - layer.4.output 0.00014250 0.06060520 - ------------------------------------------------------------------------------------- - TOTAL 0.00234766 0.94210002 - (elements=5,160,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5160960 -Total Bytes 876744 -BPFP 1.3590 bits/point -EBPFP 2.7181 equivalent bits/point -MSE 0.942100 ----------------------- -------------------------------------------------------- -Time: 1.173s Load: 0.019s, Pack+Encode: 0.432s, Decode+Unpack: 0.721s ----------------------- -------------------------------------------------------- -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 0.9421 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 13,012B, BPFP=0.3458 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 63,164B, BPFP=1.6785 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 40,204B, BPFP=1.0683 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 67,984B, BPFP=1.8065 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 47,792B, BPFP=1.2700 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 69,228B, BPFP=1.8396 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 49,432B, BPFP=1.3136 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 67,848B, BPFP=1.8029 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 41,600B, BPFP=1.1054 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 69,732B, BPFP=1.8530 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 194,136B, BPFP=1.2897 -⌛️ [2/4] FRONTEND: Frontend time: 0.376s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.592s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.36438154 - layer.0.v_cache 0.00000027 0.00024675 - layer.1.k_cache 0.00289989 0.78007030 - layer.1.v_cache 0.00000083 0.00089028 - layer.2.k_cache 0.00119190 0.45948522 - layer.2.v_cache 0.00000115 0.00131142 - layer.3.k_cache 0.00131407 0.50827577 - layer.3.v_cache 0.00000220 0.00213716 - layer.4.k_cache 0.00351532 0.89601488 - layer.4.v_cache 0.00000333 0.00370283 - layer.4.output 0.00015801 0.06933202 - ------------------------------------------------------------------------------------- - TOTAL 0.00242567 1.02098887 - (elements=4,214,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4214784 -Total Bytes 724132 -BPFP 1.3745 bits/point -EBPFP 2.7489 equivalent bits/point -MSE 1.020989 ----------------------- -------------------------------------------------------- -Time: 0.984s Load: 0.016s, Pack+Encode: 0.376s, Decode+Unpack: 0.592s ----------------------- -------------------------------------------------------- -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 1.0210 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,540B, BPFP=0.3592 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,632B, BPFP=1.6788 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,188B, BPFP=1.0676 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,044B, BPFP=1.8270 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,128B, BPFP=1.2742 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,380B, BPFP=1.8579 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,028B, BPFP=1.3181 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,096B, BPFP=1.8282 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,316B, BPFP=1.0937 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,252B, BPFP=1.8781 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 217,408B, BPFP=1.2563 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.709s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.85975566 - layer.0.v_cache 0.00000026 0.00024477 - layer.1.k_cache 0.00288373 0.78078429 - layer.1.v_cache 0.00000083 0.00090655 - layer.2.k_cache 0.00117924 0.45815367 - layer.2.v_cache 0.00000113 0.00130231 - layer.3.k_cache 0.00131114 0.50347467 - layer.3.v_cache 0.00000214 0.00213537 - layer.4.k_cache 0.00355340 0.89907458 - layer.4.v_cache 0.00000355 0.00371349 - layer.4.output 0.00015048 0.06204492 - ------------------------------------------------------------------------------------- - TOTAL 0.00237458 0.98269465 - (elements=4,845,568) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4845568 -Total Bytes 831012 -BPFP 1.3720 bits/point -EBPFP 2.7440 equivalent bits/point -MSE 0.982695 ----------------------- -------------------------------------------------------- -Time: 1.141s Load: 0.019s, Pack+Encode: 0.414s, Decode+Unpack: 0.709s ----------------------- -------------------------------------------------------- -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 0.9827 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,280B, BPFP=0.3511 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,072B, BPFP=1.6790 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,308B, BPFP=1.0641 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,048B, BPFP=1.8164 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,820B, BPFP=1.2597 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,580B, BPFP=1.8516 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,140B, BPFP=1.3130 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,180B, BPFP=1.8194 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,800B, BPFP=1.0983 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,484B, BPFP=1.8723 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 223,568B, BPFP=1.2843 -⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.720s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.10329805 - layer.0.v_cache 0.00000026 0.00024357 - layer.1.k_cache 0.00294777 0.76928145 - layer.1.v_cache 0.00000081 0.00092573 - layer.2.k_cache 0.00121079 0.45116703 - layer.2.v_cache 0.00000125 0.00133621 - layer.3.k_cache 0.00131075 0.51410801 - layer.3.v_cache 0.00000218 0.00215245 - layer.4.k_cache 0.00357831 0.86522244 - layer.4.v_cache 0.00000337 0.00376335 - layer.4.output 0.00014951 0.06283867 - ------------------------------------------------------------------------------------- - TOTAL 0.00238048 0.99734664 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 838280 -BPFP 1.3759 bits/point -EBPFP 2.7517 equivalent bits/point -MSE 0.997347 ----------------------- -------------------------------------------------------- -Time: 1.156s Load: 0.019s, Pack+Encode: 0.417s, Decode+Unpack: 0.720s ----------------------- -------------------------------------------------------- -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 0.9973 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.021s - ------------------------------------------------------------- -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: 13,436B, BPFP=0.3280 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,492B, BPFP=1.5745 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,340B, BPFP=1.0093 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,540B, BPFP=1.6978 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 49,196B, BPFP=1.2011 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,572B, BPFP=1.7229 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 50,652B, BPFP=1.2366 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,288B, BPFP=1.6916 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 42,708B, BPFP=1.0427 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,060B, BPFP=1.7349 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 195,048B, BPFP=1.1905 -⌛️ [2/4] FRONTEND: Frontend time: 0.365s (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, 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: 0.597s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.19740295 - layer.0.v_cache 0.00000027 0.00023541 - layer.1.k_cache 0.00288080 0.66635656 - layer.1.v_cache 0.00000084 0.00090899 - layer.2.k_cache 0.00117023 0.42466865 - layer.2.v_cache 0.00000114 0.00128069 - layer.3.k_cache 0.00131996 0.47600517 - layer.3.v_cache 0.00000215 0.00204581 - layer.4.k_cache 0.00348745 0.82100878 - layer.4.v_cache 0.00000346 0.00357180 - layer.4.output 0.00014756 0.05205284 - ------------------------------------------------------------------------------------- - TOTAL 0.00240337 0.84297830 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 737332 -BPFP 1.2858 bits/point -EBPFP 2.5716 equivalent bits/point -MSE 0.842978 ----------------------- -------------------------------------------------------- -Time: 0.983s Load: 0.021s, Pack+Encode: 0.365s, Decode+Unpack: 0.597s ----------------------- -------------------------------------------------------- -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 0.8430 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 14,540B, BPFP=0.3463 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,940B, BPFP=1.6897 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,120B, BPFP=1.0747 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,708B, BPFP=1.8271 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,100B, BPFP=1.2648 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,480B, BPFP=1.8693 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,796B, BPFP=1.3290 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,764B, BPFP=1.8284 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,672B, BPFP=1.1117 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,048B, BPFP=1.8828 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,384B, BPFP=1.2706 -⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.722s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.80417093 - layer.0.v_cache 0.00000026 0.00025159 - layer.1.k_cache 0.00293126 0.75515831 - layer.1.v_cache 0.00000079 0.00091156 - layer.2.k_cache 0.00115822 0.44613345 - layer.2.v_cache 0.00000116 0.00130184 - layer.3.k_cache 0.00132993 0.50527331 - layer.3.v_cache 0.00000214 0.00209743 - layer.4.k_cache 0.00358323 0.88798290 - layer.4.v_cache 0.00000336 0.00365415 - layer.4.output 0.00014415 0.06131893 - ------------------------------------------------------------------------------------- - TOTAL 0.00240751 1.04658651 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 810552 -BPFP 1.3790 bits/point -EBPFP 2.7580 equivalent bits/point -MSE 1.046587 ----------------------- -------------------------------------------------------- -Time: 1.159s Load: 0.019s, Pack+Encode: 0.417s, Decode+Unpack: 0.722s ----------------------- -------------------------------------------------------- -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 1.0466 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,680B, BPFP=0.3497 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,512B, BPFP=1.6795 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,252B, BPFP=1.0778 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,988B, BPFP=1.8337 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,460B, BPFP=1.2733 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,764B, BPFP=1.8760 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,652B, BPFP=1.3256 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,788B, BPFP=1.8290 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,488B, BPFP=1.1073 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 78,756B, BPFP=1.8759 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 210,428B, BPFP=1.2530 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.713s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.63617911 - layer.0.v_cache 0.00000026 0.00024798 - layer.1.k_cache 0.00293683 0.73951828 - layer.1.v_cache 0.00000081 0.00091158 - layer.2.k_cache 0.00118071 0.45640262 - layer.2.v_cache 0.00000117 0.00133859 - layer.3.k_cache 0.00133336 0.50515835 - layer.3.v_cache 0.00000214 0.00208457 - layer.4.k_cache 0.00361319 0.89302761 - layer.4.v_cache 0.00000321 0.00356907 - layer.4.output 0.00015382 0.06567988 - ------------------------------------------------------------------------------------- - TOTAL 0.00253689 0.96436838 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 807768 -BPFP 1.3743 bits/point -EBPFP 2.7486 equivalent bits/point -MSE 0.964368 ----------------------- -------------------------------------------------------- -Time: 1.142s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.713s ----------------------- -------------------------------------------------------- -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 0.9644 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 15,576B, BPFP=0.3457 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,124B, BPFP=1.6452 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,488B, BPFP=1.0540 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,980B, BPFP=1.7751 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 56,220B, BPFP=1.2478 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,584B, BPFP=1.8107 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,156B, BPFP=1.2907 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,104B, BPFP=1.7779 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,676B, BPFP=1.0803 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 82,372B, BPFP=1.8282 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 222,852B, BPFP=1.2365 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.717s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.44766582 - layer.0.v_cache 0.00000026 0.00024866 - layer.1.k_cache 0.00288394 0.74720105 - layer.1.v_cache 0.00000077 0.00091384 - layer.2.k_cache 0.00114480 0.44629726 - layer.2.v_cache 0.00000114 0.00133709 - layer.3.k_cache 0.00130173 0.50001413 - layer.3.v_cache 0.00000215 0.00217596 - layer.4.k_cache 0.00352615 0.87768433 - layer.4.v_cache 0.00000336 0.00387855 - layer.4.output 0.00014232 0.05825824 - ------------------------------------------------------------------------------------- - TOTAL 0.00238507 0.94717498 - (elements=5,046,272) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5046272 -Total Bytes 847132 -BPFP 1.3430 bits/point -EBPFP 2.6860 equivalent bits/point -MSE 0.947175 ----------------------- -------------------------------------------------------- -Time: 1.156s Load: 0.019s, Pack+Encode: 0.420s, Decode+Unpack: 0.717s ----------------------- -------------------------------------------------------- -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 0.9472 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,876B, BPFP=0.3598 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 69,884B, BPFP=1.6903 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,728B, BPFP=1.0818 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 75,272B, BPFP=1.8206 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,624B, BPFP=1.2970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 77,228B, BPFP=1.8679 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,400B, BPFP=1.3400 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 75,796B, BPFP=1.8333 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,120B, BPFP=1.1155 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 77,852B, BPFP=1.8830 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,972B, BPFP=1.2999 -⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.720s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.53157048 - layer.0.v_cache 0.00000027 0.00024614 - layer.1.k_cache 0.00290123 0.73894158 - layer.1.v_cache 0.00000082 0.00090145 - layer.2.k_cache 0.00116542 0.45796657 - layer.2.v_cache 0.00000118 0.00135306 - layer.3.k_cache 0.00132756 0.50343299 - layer.3.v_cache 0.00000224 0.00217969 - layer.4.k_cache 0.00351721 0.86657668 - layer.4.v_cache 0.00000335 0.00379591 - layer.4.output 0.00015064 0.06426832 - ------------------------------------------------------------------------------------- - TOTAL 0.00241998 0.95457413 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 805752 -BPFP 1.3921 bits/point -EBPFP 2.7841 equivalent bits/point -MSE 0.954574 ----------------------- -------------------------------------------------------- -Time: 1.156s Load: 0.018s, Pack+Encode: 0.418s, Decode+Unpack: 0.720s ----------------------- -------------------------------------------------------- -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 0.9546 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,808B, BPFP=0.3538 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,828B, BPFP=1.6922 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,468B, BPFP=1.0863 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 76,828B, BPFP=1.8355 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,512B, BPFP=1.2785 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,596B, BPFP=1.8778 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,716B, BPFP=1.3311 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,640B, BPFP=1.8310 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,592B, BPFP=1.1131 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 78,812B, BPFP=1.8829 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,200B, BPFP=1.2854 -⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.711s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.07417097 - layer.0.v_cache 0.00000027 0.00024466 - layer.1.k_cache 0.00291079 0.74629603 - layer.1.v_cache 0.00000081 0.00092660 - layer.2.k_cache 0.00118771 0.44909145 - layer.2.v_cache 0.00000120 0.00134677 - layer.3.k_cache 0.00132021 0.50697863 - layer.3.v_cache 0.00000219 0.00216776 - layer.4.k_cache 0.00351552 0.87818587 - layer.4.v_cache 0.00000327 0.00370724 - layer.4.output 0.00014986 0.06576770 - ------------------------------------------------------------------------------------- - TOTAL 0.00241031 0.99472763 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 813000 -BPFP 1.3874 bits/point -EBPFP 2.7748 equivalent bits/point -MSE 0.994728 ----------------------- -------------------------------------------------------- -Time: 1.166s Load: 0.018s, Pack+Encode: 0.437s, Decode+Unpack: 0.711s ----------------------- -------------------------------------------------------- -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 0.9947 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 14,852B, BPFP=0.3516 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,020B, BPFP=1.6813 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,260B, BPFP=1.0715 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,204B, BPFP=1.8277 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,868B, BPFP=1.2753 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,736B, BPFP=1.8640 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,812B, BPFP=1.3213 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,140B, BPFP=1.8262 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,796B, BPFP=1.1079 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,364B, BPFP=1.8789 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 214,348B, BPFP=1.2686 -⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.718s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.11438654 - layer.0.v_cache 0.00000027 0.00024436 - layer.1.k_cache 0.00294190 0.76483783 - layer.1.v_cache 0.00000076 0.00086892 - layer.2.k_cache 0.00116180 0.45262003 - layer.2.v_cache 0.00000113 0.00125489 - layer.3.k_cache 0.00131353 0.50646270 - layer.3.v_cache 0.00000217 0.00205242 - layer.4.k_cache 0.00362690 0.89597991 - layer.4.v_cache 0.00000322 0.00355848 - layer.4.output 0.00015895 0.06577807 - ------------------------------------------------------------------------------------- - TOTAL 0.00250451 1.00038417 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 814400 -BPFP 1.3772 bits/point -EBPFP 2.7543 equivalent bits/point -MSE 1.000384 ----------------------- -------------------------------------------------------- -Time: 1.159s Load: 0.019s, Pack+Encode: 0.422s, Decode+Unpack: 0.718s ----------------------- -------------------------------------------------------- -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 1.0004 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,264B, BPFP=0.3387 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,552B, BPFP=1.6753 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,200B, BPFP=1.0733 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,096B, BPFP=1.8307 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,300B, BPFP=1.2657 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,364B, BPFP=1.8608 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,616B, BPFP=1.3207 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 76,592B, BPFP=1.8188 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,396B, BPFP=1.1017 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 78,808B, BPFP=1.8714 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 208,104B, BPFP=1.2354 -⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.731s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.12284874 - layer.0.v_cache 0.00000026 0.00024931 - layer.1.k_cache 0.00292509 0.77670279 - layer.1.v_cache 0.00000078 0.00090386 - layer.2.k_cache 0.00118155 0.45710281 - layer.2.v_cache 0.00000112 0.00127905 - layer.3.k_cache 0.00133849 0.50767095 - layer.3.v_cache 0.00000208 0.00203343 - layer.4.k_cache 0.00360187 0.89105271 - layer.4.v_cache 0.00000304 0.00350939 - layer.4.output 0.00015018 0.06326039 - ------------------------------------------------------------------------------------- - TOTAL 0.00234762 1.00117104 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 804292 -BPFP 1.3642 bits/point -EBPFP 2.7284 equivalent bits/point -MSE 1.001171 ----------------------- -------------------------------------------------------- -Time: 1.172s Load: 0.018s, Pack+Encode: 0.423s, Decode+Unpack: 0.731s ----------------------- -------------------------------------------------------- -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 1.0012 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,672B, BPFP=0.3442 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,400B, BPFP=1.6751 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,776B, BPFP=1.0739 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,732B, BPFP=1.8237 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,484B, BPFP=1.2782 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,208B, BPFP=1.8583 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,612B, BPFP=1.3282 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,920B, BPFP=1.8281 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,084B, BPFP=1.1046 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,176B, BPFP=1.8810 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 213,716B, BPFP=1.2535 -⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.718s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.31866070 - layer.0.v_cache 0.00000026 0.00024747 - layer.1.k_cache 0.00286224 0.76153060 - layer.1.v_cache 0.00000078 0.00088777 - layer.2.k_cache 0.00116467 0.46074665 - layer.2.v_cache 0.00000115 0.00130250 - layer.3.k_cache 0.00135416 0.51050213 - layer.3.v_cache 0.00000234 0.00218305 - layer.4.k_cache 0.00356469 0.90686915 - layer.4.v_cache 0.00000325 0.00365909 - layer.4.output 0.00015110 0.06432716 - ------------------------------------------------------------------------------------- - TOTAL 0.00233011 1.01599270 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 818780 -BPFP 1.3721 bits/point -EBPFP 2.7442 equivalent bits/point -MSE 1.015993 ----------------------- -------------------------------------------------------- -Time: 1.173s Load: 0.018s, Pack+Encode: 0.436s, Decode+Unpack: 0.718s ----------------------- -------------------------------------------------------- -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 1.0160 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.017s - ------------------------------------------------------------- -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: 13,408B, BPFP=0.3446 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,092B, BPFP=1.6471 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 41,204B, BPFP=1.0589 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,240B, BPFP=1.7794 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 48,812B, BPFP=1.2544 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 70,400B, BPFP=1.8092 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 50,148B, BPFP=1.2888 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 68,936B, BPFP=1.7716 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 42,048B, BPFP=1.0806 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 70,908B, BPFP=1.8223 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 196,080B, BPFP=1.2598 -⌛️ [2/4] FRONTEND: Frontend time: 0.377s (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, 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: 0.601s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.50461538 - layer.0.v_cache 0.00000027 0.00024544 - layer.1.k_cache 0.00290015 0.73254736 - layer.1.v_cache 0.00000081 0.00088853 - layer.2.k_cache 0.00116609 0.44874839 - layer.2.v_cache 0.00000115 0.00129719 - layer.3.k_cache 0.00131777 0.49220346 - layer.3.v_cache 0.00000216 0.00212357 - layer.4.k_cache 0.00349415 0.85938564 - layer.4.v_cache 0.00000336 0.00365811 - layer.4.output 0.00015143 0.05773174 - ------------------------------------------------------------------------------------- - TOTAL 0.00240374 0.94833143 - (elements=4,358,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4358144 -Total Bytes 735276 -BPFP 1.3497 bits/point -EBPFP 2.6994 equivalent bits/point -MSE 0.948331 ----------------------- -------------------------------------------------------- -Time: 0.996s Load: 0.017s, Pack+Encode: 0.377s, Decode+Unpack: 0.601s ----------------------- -------------------------------------------------------- -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 0.9483 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.020s - ------------------------------------------------------------- -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: 15,396B, BPFP=0.3427 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,616B, BPFP=1.6385 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,016B, BPFP=1.0465 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,568B, BPFP=1.7710 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,772B, BPFP=1.2414 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,356B, BPFP=1.8108 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,048B, BPFP=1.2920 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,124B, BPFP=1.7834 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,752B, BPFP=1.0851 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 82,388B, BPFP=1.8338 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 223,788B, BPFP=1.2453 -⌛️ [2/4] FRONTEND: Frontend time: 0.426s (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, 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: 0.716s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.17766621 - layer.0.v_cache 0.00000026 0.00023733 - layer.1.k_cache 0.00288319 0.73932554 - layer.1.v_cache 0.00000079 0.00087539 - layer.2.k_cache 0.00116994 0.44637822 - layer.2.v_cache 0.00000119 0.00132743 - layer.3.k_cache 0.00131288 0.49691190 - layer.3.v_cache 0.00000224 0.00215315 - layer.4.k_cache 0.00358885 0.87670933 - layer.4.v_cache 0.00000339 0.00366355 - layer.4.output 0.00014945 0.06124667 - ------------------------------------------------------------------------------------- - TOTAL 0.00231701 0.92787391 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 845824 -BPFP 1.3447 bits/point -EBPFP 2.6895 equivalent bits/point -MSE 0.927874 ----------------------- -------------------------------------------------------- -Time: 1.162s Load: 0.020s, Pack+Encode: 0.426s, Decode+Unpack: 0.716s ----------------------- -------------------------------------------------------- -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 0.9279 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 14,564B, BPFP=0.3417 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 71,096B, BPFP=1.6680 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,640B, BPFP=1.0708 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,300B, BPFP=1.8135 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,000B, BPFP=1.2669 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,144B, BPFP=1.8568 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,156B, BPFP=1.3175 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,496B, BPFP=1.8181 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,652B, BPFP=1.0945 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,848B, BPFP=1.8733 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 209,484B, BPFP=1.2287 -⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.729s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.16496575 - layer.0.v_cache 0.00000026 0.00024893 - layer.1.k_cache 0.00290993 0.76526012 - layer.1.v_cache 0.00000078 0.00089169 - layer.2.k_cache 0.00115179 0.45854448 - layer.2.v_cache 0.00000114 0.00134151 - layer.3.k_cache 0.00134347 0.51194653 - layer.3.v_cache 0.00000212 0.00215101 - layer.4.k_cache 0.00362086 0.92595043 - layer.4.v_cache 0.00000365 0.00364644 - layer.4.output 0.00013924 0.06086103 - ------------------------------------------------------------------------------------- - TOTAL 0.00234853 1.00559936 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 811380 -BPFP 1.3597 bits/point -EBPFP 2.7194 equivalent bits/point -MSE 1.005599 ----------------------- -------------------------------------------------------- -Time: 1.189s Load: 0.019s, Pack+Encode: 0.441s, Decode+Unpack: 0.729s ----------------------- -------------------------------------------------------- -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 1.0056 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 13,956B, BPFP=0.3418 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 64,728B, BPFP=1.5852 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 42,248B, BPFP=1.0347 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 69,876B, BPFP=1.7113 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 50,416B, BPFP=1.2347 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 71,256B, BPFP=1.7451 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 51,588B, BPFP=1.2634 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 69,912B, BPFP=1.7122 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 43,972B, BPFP=1.0769 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 71,672B, BPFP=1.7553 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 203,308B, BPFP=1.2448 -⌛️ [2/4] FRONTEND: Frontend time: 0.383s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.605s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.01916083 - layer.0.v_cache 0.00000027 0.00023722 - layer.1.k_cache 0.00284013 0.67086974 - layer.1.v_cache 0.00000084 0.00089194 - layer.2.k_cache 0.00117936 0.44148015 - layer.2.v_cache 0.00000120 0.00133083 - layer.3.k_cache 0.00130692 0.47761320 - layer.3.v_cache 0.00000226 0.00212624 - layer.4.k_cache 0.00361670 0.83319886 - layer.4.v_cache 0.00000371 0.00374639 - layer.4.output 0.00016871 0.05666013 - ------------------------------------------------------------------------------------- - TOTAL 0.00247411 0.83409257 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 752932 -BPFP 1.3171 bits/point -EBPFP 2.6343 equivalent bits/point -MSE 0.834093 ----------------------- -------------------------------------------------------- -Time: 1.006s Load: 0.018s, Pack+Encode: 0.383s, Decode+Unpack: 0.605s ----------------------- -------------------------------------------------------- -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 0.8341 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.019s - ------------------------------------------------------------- -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: 15,908B, BPFP=0.3521 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 74,876B, BPFP=1.6571 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 47,240B, BPFP=1.0455 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,992B, BPFP=1.7704 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,896B, BPFP=1.2371 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,580B, BPFP=1.8055 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 58,028B, BPFP=1.2843 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 80,096B, BPFP=1.7727 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,508B, BPFP=1.0736 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 82,128B, BPFP=1.8176 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 230,480B, BPFP=1.2752 -⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.719s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.63961386 - layer.0.v_cache 0.00000027 0.00025097 - layer.1.k_cache 0.00292436 0.74043019 - layer.1.v_cache 0.00000085 0.00091387 - layer.2.k_cache 0.00118067 0.44759540 - layer.2.v_cache 0.00000120 0.00132329 - layer.3.k_cache 0.00130453 0.50459039 - layer.3.v_cache 0.00000223 0.00216530 - layer.4.k_cache 0.00350570 0.86315589 - layer.4.v_cache 0.00000328 0.00367561 - layer.4.output 0.00016368 0.06367274 - ------------------------------------------------------------------------------------- - TOTAL 0.00238518 0.88988612 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 854732 -BPFP 1.3512 bits/point -EBPFP 2.7024 equivalent bits/point -MSE 0.889886 ----------------------- -------------------------------------------------------- -Time: 1.174s Load: 0.019s, Pack+Encode: 0.435s, Decode+Unpack: 0.719s ----------------------- -------------------------------------------------------- -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 0.8899 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.023s - ------------------------------------------------------------- -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: 15,688B, BPFP=0.3563 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 73,788B, BPFP=1.6758 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,964B, BPFP=1.0666 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 79,816B, BPFP=1.8127 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,896B, BPFP=1.2694 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 81,124B, BPFP=1.8424 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,652B, BPFP=1.3093 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 79,984B, BPFP=1.8165 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,084B, BPFP=1.0920 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,760B, BPFP=1.8568 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 226,664B, BPFP=1.2869 -⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.717s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.69593776 - layer.0.v_cache 0.00000027 0.00025624 - layer.1.k_cache 0.00288154 0.75716746 - layer.1.v_cache 0.00000082 0.00092969 - layer.2.k_cache 0.00115445 0.44941663 - layer.2.v_cache 0.00000117 0.00133763 - layer.3.k_cache 0.00131257 0.50646556 - layer.3.v_cache 0.00000224 0.00221470 - layer.4.k_cache 0.00356118 0.85665477 - layer.4.v_cache 0.00000339 0.00369789 - layer.4.output 0.00014496 0.06276905 - ------------------------------------------------------------------------------------- - TOTAL 0.00246697 0.96608247 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 847420 -BPFP 1.3747 bits/point -EBPFP 2.7494 equivalent bits/point -MSE 0.966082 ----------------------- -------------------------------------------------------- -Time: 1.175s Load: 0.023s, Pack+Encode: 0.434s, Decode+Unpack: 0.717s ----------------------- -------------------------------------------------------- -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 0.9661 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,172B, BPFP=0.3449 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 68,820B, BPFP=1.6749 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,288B, BPFP=1.0779 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 74,720B, BPFP=1.8185 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,124B, BPFP=1.2929 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 76,040B, BPFP=1.8507 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 54,792B, BPFP=1.3335 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 74,576B, BPFP=1.8150 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 45,752B, BPFP=1.1135 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 76,760B, BPFP=1.8682 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 205,164B, BPFP=1.2483 -⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.708s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 9.91986788 - layer.0.v_cache 0.00000026 0.00024190 - layer.1.k_cache 0.00287206 0.74751163 - layer.1.v_cache 0.00000077 0.00087409 - layer.2.k_cache 0.00116438 0.44982596 - layer.2.v_cache 0.00000114 0.00129765 - layer.3.k_cache 0.00132116 0.49081117 - layer.3.v_cache 0.00000214 0.00207685 - layer.4.k_cache 0.00354096 0.86261528 - layer.4.v_cache 0.00000337 0.00367133 - layer.4.output 0.00013764 0.05553902 - ------------------------------------------------------------------------------------- - TOTAL 0.00235925 0.90721070 - (elements=4,601,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4601856 -Total Bytes 788208 -BPFP 1.3702 bits/point -EBPFP 2.7405 equivalent bits/point -MSE 0.907211 ----------------------- -------------------------------------------------------- -Time: 1.149s Load: 0.018s, Pack+Encode: 0.423s, Decode+Unpack: 0.708s ----------------------- -------------------------------------------------------- -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 0.9072 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,696B, BPFP=0.3490 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,944B, BPFP=1.6847 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,108B, BPFP=1.0711 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,200B, BPFP=1.8332 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,004B, BPFP=1.2824 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,956B, BPFP=1.8749 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,504B, BPFP=1.3180 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,116B, BPFP=1.8312 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,744B, BPFP=1.1100 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,744B, BPFP=1.8936 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 210,824B, BPFP=1.2516 -⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.713s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.84594207 - layer.0.v_cache 0.00000026 0.00024221 - layer.1.k_cache 0.00285945 0.73882941 - layer.1.v_cache 0.00000078 0.00090333 - layer.2.k_cache 0.00114467 0.45382612 - layer.2.v_cache 0.00000113 0.00131309 - layer.3.k_cache 0.00130198 0.50847911 - layer.3.v_cache 0.00000219 0.00216744 - layer.4.k_cache 0.00350851 0.88486080 - layer.4.v_cache 0.00000338 0.00379133 - layer.4.output 0.00014624 0.06056550 - ------------------------------------------------------------------------------------- - TOTAL 0.00238323 0.97732978 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 810840 -BPFP 1.3753 bits/point -EBPFP 2.7506 equivalent bits/point -MSE 0.977330 ----------------------- -------------------------------------------------------- -Time: 1.152s Load: 0.017s, Pack+Encode: 0.422s, Decode+Unpack: 0.713s ----------------------- -------------------------------------------------------- -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 0.9773 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,680B, BPFP=0.3500 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,684B, BPFP=1.6224 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,536B, BPFP=1.0388 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,496B, BPFP=1.7521 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 55,544B, BPFP=1.2398 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 80,016B, BPFP=1.7861 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 57,712B, BPFP=1.2882 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,844B, BPFP=1.7599 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 48,332B, BPFP=1.0788 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 81,056B, BPFP=1.8093 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 215,788B, BPFP=1.2042 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (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, 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: 0.715s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.49518555 - layer.0.v_cache 0.00000028 0.00024357 - layer.1.k_cache 0.00283737 0.72327345 - layer.1.v_cache 0.00000078 0.00082636 - layer.2.k_cache 0.00114986 0.44520408 - layer.2.v_cache 0.00000111 0.00121284 - layer.3.k_cache 0.00128401 0.49234506 - layer.3.v_cache 0.00000216 0.00200738 - layer.4.k_cache 0.00352053 0.84965628 - layer.4.v_cache 0.00000313 0.00348231 - layer.4.output 0.00014459 0.06029020 - ------------------------------------------------------------------------------------- - TOTAL 0.00227848 0.94675698 - (elements=5,017,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5017600 -Total Bytes 830688 -BPFP 1.3244 bits/point -EBPFP 2.6489 equivalent bits/point -MSE 0.946757 ----------------------- -------------------------------------------------------- -Time: 1.146s Load: 0.018s, Pack+Encode: 0.413s, Decode+Unpack: 0.715s ----------------------- -------------------------------------------------------- -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 0.9468 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 14,472B, BPFP=0.3500 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 69,120B, BPFP=1.6718 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 44,680B, BPFP=1.0807 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 75,176B, BPFP=1.8183 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 53,244B, BPFP=1.2878 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 76,872B, BPFP=1.8593 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,272B, BPFP=1.3369 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 75,740B, BPFP=1.8319 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 45,816B, BPFP=1.1082 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 77,408B, BPFP=1.8723 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 208,296B, BPFP=1.2595 -⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.711s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.21554140 - layer.0.v_cache 0.00000026 0.00024213 - layer.1.k_cache 0.00293942 0.74410231 - layer.1.v_cache 0.00000088 0.00091458 - layer.2.k_cache 0.00118806 0.45873608 - layer.2.v_cache 0.00000115 0.00131962 - layer.3.k_cache 0.00133183 0.51023123 - layer.3.v_cache 0.00000219 0.00217791 - layer.4.k_cache 0.00365354 0.89390937 - layer.4.v_cache 0.00000330 0.00369060 - layer.4.output 0.00014070 0.06250288 - ------------------------------------------------------------------------------------- - TOTAL 0.00236644 1.00577691 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 796096 -BPFP 1.3754 bits/point -EBPFP 2.7508 equivalent bits/point -MSE 1.005777 ----------------------- -------------------------------------------------------- -Time: 1.152s Load: 0.017s, Pack+Encode: 0.424s, Decode+Unpack: 0.711s ----------------------- -------------------------------------------------------- -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 1.0058 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.021s - ------------------------------------------------------------- -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: 17,872B, BPFP=0.3332 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 86,252B, BPFP=1.6082 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 55,864B, BPFP=1.0416 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 92,584B, BPFP=1.7263 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 66,084B, BPFP=1.2322 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 94,368B, BPFP=1.7595 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 68,832B, BPFP=1.2834 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 93,204B, BPFP=1.7378 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 57,788B, BPFP=1.0775 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 95,976B, BPFP=1.7895 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 251,680B, BPFP=1.1732 -⌛️ [2/4] FRONTEND: Frontend time: 0.479s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.806s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.38527488 - layer.0.v_cache 0.00000028 0.00024134 - layer.1.k_cache 0.00282601 0.71782661 - layer.1.v_cache 0.00000075 0.00081604 - layer.2.k_cache 0.00117249 0.43237378 - layer.2.v_cache 0.00000110 0.00117085 - layer.3.k_cache 0.00129145 0.48678625 - layer.3.v_cache 0.00000206 0.00196039 - layer.4.k_cache 0.00360397 0.84794518 - layer.4.v_cache 0.00000307 0.00340096 - layer.4.output 0.00013808 0.05511435 - ------------------------------------------------------------------------------------- - TOTAL 0.00228159 0.93558955 - (elements=6,006,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 6006784 -Total Bytes 980504 -BPFP 1.3059 bits/point -EBPFP 2.6117 equivalent bits/point -MSE 0.935590 ----------------------- -------------------------------------------------------- -Time: 1.306s Load: 0.021s, Pack+Encode: 0.479s, Decode+Unpack: 0.806s ----------------------- -------------------------------------------------------- -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 0.9356 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 15,228B, BPFP=0.3541 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 72,308B, BPFP=1.6813 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 46,224B, BPFP=1.0748 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 78,392B, BPFP=1.8227 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,788B, BPFP=1.2739 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 79,872B, BPFP=1.8571 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 56,664B, BPFP=1.3175 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 78,508B, BPFP=1.8254 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 47,364B, BPFP=1.1013 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 80,748B, BPFP=1.8775 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 216,992B, BPFP=1.2613 -⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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: 0.695s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 11.02158247 - layer.0.v_cache 0.00000026 0.00024429 - layer.1.k_cache 0.00291391 0.77367583 - layer.1.v_cache 0.00000078 0.00090249 - layer.2.k_cache 0.00115383 0.45145471 - layer.2.v_cache 0.00000117 0.00131936 - layer.3.k_cache 0.00133386 0.49822208 - layer.3.v_cache 0.00000214 0.00212540 - layer.4.k_cache 0.00353874 0.88353311 - layer.4.v_cache 0.00000342 0.00373451 - layer.4.output 0.00014590 0.06552020 - ------------------------------------------------------------------------------------- - TOTAL 0.00237300 0.99277679 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 827088 -BPFP 1.3736 bits/point -EBPFP 2.7473 equivalent bits/point -MSE 0.992777 ----------------------- -------------------------------------------------------- -Time: 1.134s Load: 0.018s, Pack+Encode: 0.422s, Decode+Unpack: 0.695s ----------------------- -------------------------------------------------------- -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 0.9928 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-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: 0.018s - ------------------------------------------------------------- -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: 14,472B, BPFP=0.3437 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 70,984B, BPFP=1.6856 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 45,180B, BPFP=1.0729 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 77,156B, BPFP=1.8322 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 54,052B, BPFP=1.2835 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 78,612B, BPFP=1.8667 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 55,488B, BPFP=1.3176 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 77,004B, BPFP=1.8286 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 46,572B, BPFP=1.1059 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 79,228B, BPFP=1.8814 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 207,660B, BPFP=1.2328 -⌛️ [2/4] FRONTEND: Frontend time: 0.426s (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, 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: 0.706s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - 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 10.86223271 - layer.0.v_cache 0.00000027 0.00024824 - layer.1.k_cache 0.00292153 0.74404954 - layer.1.v_cache 0.00000078 0.00091246 - layer.2.k_cache 0.00117184 0.45961931 - layer.2.v_cache 0.00000114 0.00132250 - layer.3.k_cache 0.00131754 0.49724887 - layer.3.v_cache 0.00000217 0.00216945 - layer.4.k_cache 0.00353394 0.87804313 - layer.4.v_cache 0.00000342 0.00371855 - layer.4.output 0.00014204 0.05889927 - ------------------------------------------------------------------------------------- - TOTAL 0.00249574 0.97751156 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 806408 -BPFP 1.3678 bits/point -EBPFP 2.7356 equivalent bits/point -MSE 0.977512 ----------------------- -------------------------------------------------------- -Time: 1.150s Load: 0.018s, Pack+Encode: 0.426s, Decode+Unpack: 0.706s ----------------------- -------------------------------------------------------- -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 0.9775 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.007/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 1.3544 bits/point -Avg EBPFP 2.7089 equivalent bits/point -Avg MSE 0.963442 -Avg Time 1.128s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:b25e1f848ff9e679e7ceaf4e80f0e167565feea2be11acb1823c3a9d06b0d043 +size 1125050