Experiment: dtufc_elic-featurecoding_falconmamba_individual Log file: output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/dtufc_elic-featurecoding_falconmamba_individual.log DTUFCCodecConfig: arch: elic-featurecoding handler: falconmamba checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.001_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar transform_type: kmeans transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json bit_depth: 8 device: cuda:0 Loading checkpoint: codec_weights/elic_hybrid/elic2022-official_lambda0.001_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar Checkpoint epoch: 286 Loaded elic-featurecoding (1-channel) on cuda:0 Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.0.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.0.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.1.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.1.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.2.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.2.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.3.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.3.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.4.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.4.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json: torch.Size([256]) Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json Loaded per-key mappings: model=falconmamba Keys: ['layer.0.conv_state', 'layer.0.ssm_state', 'layer.1.conv_state', 'layer.1.ssm_state', 'layer.2.conv_state', 'layer.2.ssm_state', 'layer.3.conv_state', 'layer.3.ssm_state', 'layer.4.conv_state', 'layer.4.ssm_state', 'layer.4.output'] ---------------- -------------------------------------------------------------------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding Checkpoint codec_weights/elic_hybrid/elic2022-official_lambda0.001_epochs600_lr0.0001_bs60_patch256-256_checkpoint_best.pth.tar Transform type kmeans Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json Input ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande Output output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande ---------------- -------------------------------------------------------------------------------------------------------------------- Files found: 100 ---------------------------------------------------------------------- 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst (1/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 109, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,296B, BPFP=0.8726 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,100B, BPFP=0.4944 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,444B, BPFP=0.1692 ⌛️ [2/4] FRONTEND: Frontend time: 2.993s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.257s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014359 layer.1.conv_state 0.00050886 1.47216272 layer.2.ssm_state 0.00000001 0.00001407 layer.2.conv_state 0.00013788 0.27173716 layer.3.ssm_state 0.00000001 0.00001561 layer.3.conv_state 0.00007003 0.25975606 layer.4.ssm_state 0.00000003 0.00001772 layer.4.conv_state 0.00023930 0.49407208 layer.4.output 0.00000273 0.00152911 ------------------------------------------------------------------------------------- TOTAL 0.00002952 0.07539796 (elements=1,265,664) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1265664 Total Bytes 109776 BPFP 0.6939 bits/point EBPFP 1.3280 equivalent bits/point MSE 0.075398 ---------------------- -------------------------------------------------------- Time: 5.257s Load: 0.007s, Pack+Encode: 2.993s, Decode+Unpack: 2.257s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0754 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample1-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst (2/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,024B, BPFP=0.8560 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,148B, BPFP=0.4973 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 16,400B, BPFP=0.3051 ⌛️ [2/4] FRONTEND: Frontend time: 2.483s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.232s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014314 layer.1.conv_state 0.00050781 1.47695649 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00013544 0.27436072 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00006839 0.25967985 layer.4.ssm_state 0.00000001 0.00001746 layer.4.conv_state 0.00024756 0.48414370 layer.4.output 0.00000285 0.00160613 ------------------------------------------------------------------------------------- TOTAL 0.00002999 0.07632529 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 116488 BPFP 0.7460 bits/point EBPFP 1.3869 equivalent bits/point MSE 0.076325 ---------------------- -------------------------------------------------------- Time: 4.723s Load: 0.008s, Pack+Encode: 2.483s, Decode+Unpack: 2.232s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0763 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample10-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst (3/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,668B, BPFP=0.8342 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,044B, BPFP=0.4910 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,936B, BPFP=0.1980 ⌛️ [2/4] FRONTEND: Frontend time: 2.485s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.238s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000014 0.00014383 layer.1.conv_state 0.00050623 1.47632885 layer.2.ssm_state 0.00000001 0.00001405 layer.2.conv_state 0.00013817 0.27431309 layer.3.ssm_state 0.00000001 0.00001573 layer.3.conv_state 0.00007075 0.26068786 layer.4.ssm_state 0.00000001 0.00001734 layer.4.conv_state 0.00027926 0.48887450 layer.4.output 0.00000298 0.00163494 ------------------------------------------------------------------------------------- TOTAL 0.00003162 0.07822593 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 109584 BPFP 0.7182 bits/point EBPFP 1.3713 equivalent bits/point MSE 0.078226 ---------------------- -------------------------------------------------------- Time: 4.732s Load: 0.008s, Pack+Encode: 2.485s, Decode+Unpack: 2.238s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0782 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample100-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst (4/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,364B, BPFP=0.8767 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,772B, BPFP=0.4744 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,160B, BPFP=0.1180 ⌛️ [2/4] FRONTEND: Frontend time: 2.502s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.245s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014388 layer.1.conv_state 0.00051199 1.47469997 layer.2.ssm_state 0.00000001 0.00001414 layer.2.conv_state 0.00013382 0.27375716 layer.3.ssm_state 0.00000001 0.00001546 layer.3.conv_state 0.00007055 0.25990590 layer.4.ssm_state 0.00000005 0.00001754 layer.4.conv_state 0.00025245 0.48909780 layer.4.output 0.00000298 0.00158646 ------------------------------------------------------------------------------------- TOTAL 0.00003058 0.07712245 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 106236 BPFP 0.6871 bits/point EBPFP 1.3343 equivalent bits/point MSE 0.077122 ---------------------- -------------------------------------------------------- Time: 4.753s Load: 0.007s, Pack+Encode: 2.502s, Decode+Unpack: 2.245s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0771 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample103-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst (5/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,948B, BPFP=0.8513 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,028B, BPFP=0.4900 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,572B, BPFP=0.1246 ⌛️ [2/4] FRONTEND: Frontend time: 2.487s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.229s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014295 layer.1.conv_state 0.00050971 1.47028482 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00014652 0.27140945 layer.3.ssm_state 0.00000001 0.00001533 layer.3.conv_state 0.00006897 0.26063794 layer.4.ssm_state 0.00000002 0.00001770 layer.4.conv_state 0.00023369 0.49358171 layer.4.output 0.00000300 0.00155851 ------------------------------------------------------------------------------------- TOTAL 0.00003021 0.07682271 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 106496 BPFP 0.6865 bits/point EBPFP 1.3306 equivalent bits/point MSE 0.076823 ---------------------- -------------------------------------------------------- Time: 4.725s Load: 0.009s, Pack+Encode: 2.487s, Decode+Unpack: 2.229s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0768 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample104-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst (6/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 15,092B, BPFP=0.9211 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,656B, BPFP=0.4673 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,780B, BPFP=0.9229 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,516B, BPFP=0.1260 ⌛️ [2/4] FRONTEND: Frontend time: 2.489s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.249s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000947 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014268 layer.1.conv_state 0.00050648 1.47145355 layer.2.ssm_state 0.00000001 0.00001414 layer.2.conv_state 0.00011943 0.27101368 layer.3.ssm_state 0.00000001 0.00001545 layer.3.conv_state 0.00007205 0.25742859 layer.4.ssm_state 0.00000003 0.00001774 layer.4.conv_state 0.00022891 0.49289456 layer.4.output 0.00000312 0.00158128 ------------------------------------------------------------------------------------- TOTAL 0.00002958 0.07724742 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 107180 BPFP 0.6955 bits/point EBPFP 1.3487 equivalent bits/point MSE 0.077247 ---------------------- -------------------------------------------------------- Time: 4.747s Load: 0.009s, Pack+Encode: 2.489s, Decode+Unpack: 2.249s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample105-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst (7/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,364B, BPFP=0.8767 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,856B, BPFP=0.4795 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,168B, BPFP=0.2598 ⌛️ [2/4] FRONTEND: Frontend time: 2.481s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.229s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014324 layer.1.conv_state 0.00050697 1.47438848 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00013442 0.27321979 layer.3.ssm_state 0.00000001 0.00001546 layer.3.conv_state 0.00006950 0.25980356 layer.4.ssm_state 0.00000001 0.00001746 layer.4.conv_state 0.00023016 0.48816651 layer.4.output 0.00000331 0.00165223 ------------------------------------------------------------------------------------- TOTAL 0.00003020 0.07785166 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 113300 BPFP 0.7401 bits/point EBPFP 1.3942 equivalent bits/point MSE 0.077852 ---------------------- -------------------------------------------------------- Time: 4.719s Load: 0.009s, Pack+Encode: 2.481s, Decode+Unpack: 2.229s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample108-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst (8/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,224B, BPFP=0.8682 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,460B, BPFP=0.5164 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,940B, BPFP=0.1316 ⌛️ [2/4] FRONTEND: Frontend time: 2.475s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014361 layer.1.conv_state 0.00051140 1.47315085 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00014293 0.27589226 layer.3.ssm_state 0.00000001 0.00001576 layer.3.conv_state 0.00006973 0.26257491 layer.4.ssm_state 0.00000001 0.00001754 layer.4.conv_state 0.00023680 0.49292612 layer.4.output 0.00000289 0.00156498 ------------------------------------------------------------------------------------- TOTAL 0.00003023 0.07705287 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 107532 BPFP 0.6931 bits/point EBPFP 1.3416 equivalent bits/point MSE 0.077053 ---------------------- -------------------------------------------------------- Time: 4.712s Load: 0.009s, Pack+Encode: 2.475s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0771 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample110-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst (9/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.010s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,016B, BPFP=0.8555 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,404B, BPFP=0.5129 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,744B, BPFP=0.9141 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,584B, BPFP=1.1953 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,784B, BPFP=0.9238 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,912B, BPFP=0.9551 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,668B, BPFP=0.1644 ⌛️ [2/4] FRONTEND: Frontend time: 2.473s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.226s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000014 0.00014407 layer.1.conv_state 0.00051350 1.47650433 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00014131 0.27547196 layer.3.ssm_state 0.00000001 0.00001578 layer.3.conv_state 0.00006915 0.26250032 layer.4.ssm_state 0.00000002 0.00001762 layer.4.conv_state 0.00025593 0.49518964 layer.4.output 0.00000302 0.00158581 ------------------------------------------------------------------------------------- TOTAL 0.00003077 0.07719524 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 109008 BPFP 0.7027 bits/point EBPFP 1.3494 equivalent bits/point MSE 0.077195 ---------------------- -------------------------------------------------------- Time: 4.709s Load: 0.010s, Pack+Encode: 2.473s, Decode+Unpack: 2.226s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample111-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst (10/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,308B, BPFP=0.8123 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,484B, BPFP=0.5178 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,768B, BPFP=0.1564 ⌛️ [2/4] FRONTEND: Frontend time: 2.473s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.227s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000015 0.00014356 layer.1.conv_state 0.00050852 1.47770774 layer.2.ssm_state 0.00000001 0.00001405 layer.2.conv_state 0.00013698 0.27466825 layer.3.ssm_state 0.00000001 0.00001532 layer.3.conv_state 0.00006971 0.25979102 layer.4.ssm_state 0.00000001 0.00001742 layer.4.conv_state 0.00027123 0.48271418 layer.4.output 0.00000318 0.00165996 ------------------------------------------------------------------------------------- TOTAL 0.00003157 0.07834854 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 107492 BPFP 0.7069 bits/point EBPFP 1.3627 equivalent bits/point MSE 0.078349 ---------------------- -------------------------------------------------------- Time: 4.708s Load: 0.007s, Pack+Encode: 2.473s, Decode+Unpack: 2.227s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0783 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample112-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst (11/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,248B, BPFP=0.8696 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,028B, BPFP=0.4900 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,620B, BPFP=1.1975 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,920B, BPFP=0.9570 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,308B, BPFP=0.1069 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.225s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014334 layer.1.conv_state 0.00049980 1.47731829 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00011977 0.27422187 layer.3.ssm_state 0.00000001 0.00001582 layer.3.conv_state 0.00007203 0.26046222 layer.4.ssm_state 0.00000001 0.00001786 layer.4.conv_state 0.00024562 0.50253111 layer.4.output 0.00000308 0.00163812 ------------------------------------------------------------------------------------- TOTAL 0.00003020 0.07887085 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 105564 BPFP 0.6942 bits/point EBPFP 1.3535 equivalent bits/point MSE 0.078871 ---------------------- -------------------------------------------------------- Time: 4.703s Load: 0.009s, Pack+Encode: 2.469s, Decode+Unpack: 2.225s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0789 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample113-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst (12/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 95, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,780B, BPFP=0.8411 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,800B, BPFP=0.9277 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,932B, BPFP=0.4841 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,620B, BPFP=1.1975 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,780B, BPFP=0.2011 ⌛️ [2/4] FRONTEND: Frontend time: 2.468s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014379 layer.1.conv_state 0.00051224 1.47483134 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00013107 0.27424899 layer.3.ssm_state 0.00000001 0.00001547 layer.3.conv_state 0.00006945 0.26081264 layer.4.ssm_state 0.00000001 0.00001741 layer.4.conv_state 0.00027531 0.48354250 layer.4.output 0.00000313 0.00170143 ------------------------------------------------------------------------------------- TOTAL 0.00003178 0.07884265 (elements=1,208,320) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1208320 Total Bytes 109436 BPFP 0.7245 bits/point EBPFP 1.3843 equivalent bits/point MSE 0.078843 ---------------------- -------------------------------------------------------- Time: 4.708s Load: 0.007s, Pack+Encode: 2.468s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0788 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample117-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst (13/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 94, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,824B, BPFP=0.8438 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,916B, BPFP=0.4832 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,780B, BPFP=0.9229 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,588B, BPFP=0.1577 ⌛️ [2/4] FRONTEND: Frontend time: 2.467s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 94, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 94, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000009 0.00014185 layer.1.conv_state 0.00049706 1.47436810 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00015153 0.27299774 layer.3.ssm_state 0.00000001 0.00001503 layer.3.conv_state 0.00007501 0.25954360 layer.4.ssm_state 0.00000002 0.00001745 layer.4.conv_state 0.00026177 0.48921746 layer.4.output 0.00000322 0.00168362 ------------------------------------------------------------------------------------- TOTAL 0.00003183 0.07917233 (elements=1,204,224) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1204224 Total Bytes 107304 BPFP 0.7129 bits/point EBPFP 1.3753 equivalent bits/point MSE 0.079172 ---------------------- -------------------------------------------------------- Time: 4.704s Load: 0.009s, Pack+Encode: 2.467s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0792 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample121-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst (14/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 12,928B, BPFP=0.7891 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,828B, BPFP=0.9346 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,620B, BPFP=0.4651 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 16,596B, BPFP=0.3241 ⌛️ [2/4] FRONTEND: Frontend time: 2.486s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.223s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014446 layer.1.conv_state 0.00051295 1.47292101 layer.2.ssm_state 0.00000001 0.00001400 layer.2.conv_state 0.00015130 0.27406931 layer.3.ssm_state 0.00000001 0.00001515 layer.3.conv_state 0.00007484 0.26179194 layer.4.ssm_state 0.00000001 0.00001740 layer.4.conv_state 0.00021889 0.48565990 layer.4.output 0.00000110 0.00162406 ------------------------------------------------------------------------------------- TOTAL 0.00002983 0.07755805 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 115064 BPFP 0.7491 bits/point EBPFP 1.3902 equivalent bits/point MSE 0.077558 ---------------------- -------------------------------------------------------- Time: 4.716s Load: 0.007s, Pack+Encode: 2.486s, Decode+Unpack: 2.223s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0776 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample123-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst (15/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,664B, BPFP=0.8340 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,160B, BPFP=0.4980 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 11,324B, BPFP=0.2257 ⌛️ [2/4] FRONTEND: Frontend time: 2.457s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.223s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014317 layer.1.conv_state 0.00051116 1.47526252 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00011994 0.27332222 layer.3.ssm_state 0.00000001 0.00001551 layer.3.conv_state 0.00007252 0.25936332 layer.4.ssm_state 0.00000001 0.00001743 layer.4.conv_state 0.00026041 0.48574349 layer.4.output 0.00000302 0.00164405 ------------------------------------------------------------------------------------- TOTAL 0.00003081 0.07805401 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 111088 BPFP 0.7281 bits/point EBPFP 1.3819 equivalent bits/point MSE 0.078054 ---------------------- -------------------------------------------------------- Time: 4.686s Load: 0.007s, Pack+Encode: 2.457s, Decode+Unpack: 2.223s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0781 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample124-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst (16/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,116B, BPFP=0.8616 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,160B, BPFP=0.4980 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,576B, BPFP=1.1948 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,356B, BPFP=0.1827 ⌛️ [2/4] FRONTEND: Frontend time: 2.487s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.239s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014343 layer.1.conv_state 0.00051289 1.47830045 layer.2.ssm_state 0.00000001 0.00001395 layer.2.conv_state 0.00013364 0.27306056 layer.3.ssm_state 0.00000001 0.00001528 layer.3.conv_state 0.00007097 0.25904626 layer.4.ssm_state 0.00000004 0.00001741 layer.4.conv_state 0.00024017 0.48480055 layer.4.output 0.00000292 0.00159411 ------------------------------------------------------------------------------------- TOTAL 0.00003044 0.07756838 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 109544 BPFP 0.7132 bits/point EBPFP 1.3654 equivalent bits/point MSE 0.077568 ---------------------- -------------------------------------------------------- Time: 4.735s Load: 0.009s, Pack+Encode: 2.487s, Decode+Unpack: 2.239s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0776 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample125-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst (17/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 12,412B, BPFP=0.7576 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,004B, BPFP=0.4885 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,740B, BPFP=0.9131 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,920B, BPFP=0.9570 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 12,920B, BPFP=0.2575 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014372 layer.1.conv_state 0.00050368 1.47830188 layer.2.ssm_state 0.00000001 0.00001401 layer.2.conv_state 0.00013069 0.27425933 layer.3.ssm_state 0.00000001 0.00001527 layer.3.conv_state 0.00007150 0.26105660 layer.4.ssm_state 0.00000001 0.00001744 layer.4.conv_state 0.00023706 0.48987368 layer.4.output 0.00000305 0.00164855 ------------------------------------------------------------------------------------- TOTAL 0.00003027 0.07831860 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 111240 BPFP 0.7291 bits/point EBPFP 1.3735 equivalent bits/point MSE 0.078319 ---------------------- -------------------------------------------------------- Time: 4.716s Load: 0.009s, Pack+Encode: 2.479s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0783 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample126-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst (18/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,968B, BPFP=0.8525 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,804B, BPFP=0.9287 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,616B, BPFP=0.4648 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,820B, BPFP=0.0914 ⌛️ [2/4] FRONTEND: Frontend time: 2.474s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.241s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014141 layer.1.conv_state 0.00051308 1.47408092 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013634 0.27245706 layer.3.ssm_state 0.00000001 0.00001536 layer.3.conv_state 0.00007240 0.26060808 layer.4.ssm_state 0.00000002 0.00001741 layer.4.conv_state 0.00022022 0.49113494 layer.4.output 0.00000307 0.00154385 ------------------------------------------------------------------------------------- TOTAL 0.00002979 0.07688003 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 104336 BPFP 0.6725 bits/point EBPFP 1.3140 equivalent bits/point MSE 0.076880 ---------------------- -------------------------------------------------------- Time: 4.724s Load: 0.009s, Pack+Encode: 2.474s, Decode+Unpack: 2.241s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0769 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample13-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst (19/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,288B, BPFP=0.8721 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,868B, BPFP=0.4802 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,912B, BPFP=0.9551 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,504B, BPFP=0.1108 ⌛️ [2/4] FRONTEND: Frontend time: 2.464s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014341 layer.1.conv_state 0.00049676 1.47586894 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00011509 0.27406305 layer.3.ssm_state 0.00000001 0.00001570 layer.3.conv_state 0.00006928 0.26028523 layer.4.ssm_state 0.00000002 0.00001780 layer.4.conv_state 0.00024590 0.50211430 layer.4.output 0.00000308 0.00164147 ------------------------------------------------------------------------------------- TOTAL 0.00002993 0.07881261 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 105608 BPFP 0.6945 bits/point EBPFP 1.3528 equivalent bits/point MSE 0.078813 ---------------------- -------------------------------------------------------- Time: 4.700s Load: 0.008s, Pack+Encode: 2.464s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0788 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample131-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst (20/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,364B, BPFP=0.8767 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,712B, BPFP=1.2031 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,828B, BPFP=0.9346 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,020B, BPFP=0.4895 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 11,916B, BPFP=0.2375 ⌛️ [2/4] FRONTEND: Frontend time: 2.480s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000018 0.00014671 layer.1.conv_state 0.00050366 1.47491479 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00012362 0.27294192 layer.3.ssm_state 0.00000001 0.00001514 layer.3.conv_state 0.00007017 0.25983563 layer.4.ssm_state 0.00000002 0.00001771 layer.4.conv_state 0.00024945 0.49941644 layer.4.output 0.00000327 0.00165018 ------------------------------------------------------------------------------------- TOTAL 0.00003044 0.07841659 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 112256 BPFP 0.7357 bits/point EBPFP 1.3934 equivalent bits/point MSE 0.078417 ---------------------- -------------------------------------------------------- Time: 4.721s Load: 0.008s, Pack+Encode: 2.480s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0784 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample137-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst (21/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,956B, BPFP=0.8518 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,904B, BPFP=0.4824 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,716B, BPFP=0.1538 ⌛️ [2/4] FRONTEND: Frontend time: 2.470s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.230s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000028 0.00014400 layer.1.conv_state 0.00050590 1.47319388 layer.2.ssm_state 0.00000001 0.00001415 layer.2.conv_state 0.00013813 0.27291256 layer.3.ssm_state 0.00000001 0.00001540 layer.3.conv_state 0.00006868 0.25926831 layer.4.ssm_state 0.00000003 0.00001763 layer.4.conv_state 0.00022950 0.49582261 layer.4.output 0.00000311 0.00166794 ------------------------------------------------------------------------------------- TOTAL 0.00003028 0.07826347 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 107524 BPFP 0.7047 bits/point EBPFP 1.3589 equivalent bits/point MSE 0.078263 ---------------------- -------------------------------------------------------- Time: 4.710s Load: 0.009s, Pack+Encode: 2.470s, Decode+Unpack: 2.230s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0783 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample139-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst (22/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 106, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,300B, BPFP=0.8728 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,800B, BPFP=0.4761 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,028B, BPFP=0.1295 ⌛️ [2/4] FRONTEND: Frontend time: 2.481s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014175 layer.1.conv_state 0.00051424 1.47024107 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00013151 0.27148780 layer.3.ssm_state 0.00000001 0.00001505 layer.3.conv_state 0.00006895 0.25877589 layer.4.ssm_state 0.00000001 0.00001753 layer.4.conv_state 0.00023185 0.48358735 layer.4.output 0.00000293 0.00153316 ------------------------------------------------------------------------------------- TOTAL 0.00002960 0.07576680 (elements=1,253,376) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1253376 Total Bytes 107048 BPFP 0.6833 bits/point EBPFP 1.3217 equivalent bits/point MSE 0.075767 ---------------------- -------------------------------------------------------- Time: 4.717s Load: 0.008s, Pack+Encode: 2.481s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0758 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample14-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst (23/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,520B, BPFP=0.8252 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,904B, BPFP=0.4824 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,268B, BPFP=0.1378 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014258 layer.1.conv_state 0.00051274 1.47327113 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00014729 0.27352223 layer.3.ssm_state 0.00000001 0.00001555 layer.3.conv_state 0.00006977 0.26066276 layer.4.ssm_state 0.00000001 0.00001747 layer.4.conv_state 0.00021673 0.49079007 layer.4.output 0.00000307 0.00154971 ------------------------------------------------------------------------------------- TOTAL 0.00002991 0.07688125 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 106632 BPFP 0.6873 bits/point EBPFP 1.3278 equivalent bits/point MSE 0.076881 ---------------------- -------------------------------------------------------- Time: 4.713s Load: 0.009s, Pack+Encode: 2.469s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0769 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample149-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst (24/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 12,588B, BPFP=0.7683 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,072B, BPFP=0.4927 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,212B, BPFP=0.1572 ⌛️ [2/4] FRONTEND: Frontend time: 2.472s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014348 layer.1.conv_state 0.00050701 1.47995472 layer.2.ssm_state 0.00000001 0.00001415 layer.2.conv_state 0.00014784 0.27406290 layer.3.ssm_state 0.00000001 0.00001556 layer.3.conv_state 0.00006861 0.26095951 layer.4.ssm_state 0.00000001 0.00001771 layer.4.conv_state 0.00022948 0.49981722 layer.4.output 0.00000294 0.00161532 ------------------------------------------------------------------------------------- TOTAL 0.00003012 0.07759134 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 106808 BPFP 0.6908 bits/point EBPFP 1.3284 equivalent bits/point MSE 0.077591 ---------------------- -------------------------------------------------------- Time: 4.709s Load: 0.009s, Pack+Encode: 2.472s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0776 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample15-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst (25/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 11,660B, BPFP=0.7117 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,184B, BPFP=0.4995 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,740B, BPFP=0.0890 ⌛️ [2/4] FRONTEND: Frontend time: 2.472s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.225s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014185 layer.1.conv_state 0.00050709 1.47332728 layer.2.ssm_state 0.00000001 0.00001398 layer.2.conv_state 0.00011582 0.27356815 layer.3.ssm_state 0.00000001 0.00001524 layer.3.conv_state 0.00007623 0.26002151 layer.4.ssm_state 0.00000002 0.00001734 layer.4.conv_state 0.00025645 0.48774758 layer.4.output 0.00000303 0.00152057 ------------------------------------------------------------------------------------- TOTAL 0.00003006 0.07652909 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 102536 BPFP 0.6588 bits/point EBPFP 1.2871 equivalent bits/point MSE 0.076529 ---------------------- -------------------------------------------------------- Time: 4.705s Load: 0.008s, Pack+Encode: 2.472s, Decode+Unpack: 2.225s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0765 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample153-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst (26/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,388B, BPFP=0.8171 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,600B, BPFP=0.4639 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,744B, BPFP=0.9141 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,884B, BPFP=0.9482 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,572B, BPFP=0.1479 ⌛️ [2/4] FRONTEND: Frontend time: 2.470s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.229s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014364 layer.1.conv_state 0.00050979 1.47336793 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00012394 0.27351108 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00006896 0.26012453 layer.4.ssm_state 0.00000002 0.00001801 layer.4.conv_state 0.00026177 0.50294524 layer.4.output 0.00000303 0.00162325 ------------------------------------------------------------------------------------- TOTAL 0.00003064 0.07797131 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 106460 BPFP 0.6931 bits/point EBPFP 1.3369 equivalent bits/point MSE 0.077971 ---------------------- -------------------------------------------------------- Time: 4.707s Load: 0.008s, Pack+Encode: 2.470s, Decode+Unpack: 2.229s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0780 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample16-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst (27/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,916B, BPFP=0.8494 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,792B, BPFP=0.9258 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,544B, BPFP=0.5215 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 16,040B, BPFP=0.3133 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.217s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014434 layer.1.conv_state 0.00051051 1.47346306 layer.2.ssm_state 0.00000001 0.00001413 layer.2.conv_state 0.00012244 0.27265954 layer.3.ssm_state 0.00000001 0.00001556 layer.3.conv_state 0.00006890 0.25961640 layer.4.ssm_state 0.00000001 0.00001740 layer.4.conv_state 0.00024994 0.48309016 layer.4.output 0.00000315 0.00161319 ------------------------------------------------------------------------------------- TOTAL 0.00003034 0.07740479 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 116452 BPFP 0.7582 bits/point EBPFP 1.4119 equivalent bits/point MSE 0.077405 ---------------------- -------------------------------------------------------- Time: 4.695s Load: 0.009s, Pack+Encode: 2.469s, Decode+Unpack: 2.217s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0774 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample168-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst (28/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,356B, BPFP=0.8762 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,784B, BPFP=0.4751 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,820B, BPFP=0.1083 ⌛️ [2/4] FRONTEND: Frontend time: 2.470s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.225s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014576 layer.1.conv_state 0.00051452 1.47493100 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00013443 0.27328300 layer.3.ssm_state 0.00000001 0.00001543 layer.3.conv_state 0.00007113 0.25961727 layer.4.ssm_state 0.00000002 0.00001754 layer.4.conv_state 0.00021975 0.49108234 layer.4.output 0.00000261 0.00152689 ------------------------------------------------------------------------------------- TOTAL 0.00002940 0.07639725 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 105880 BPFP 0.6780 bits/point EBPFP 1.3188 equivalent bits/point MSE 0.076397 ---------------------- -------------------------------------------------------- Time: 4.704s Load: 0.009s, Pack+Encode: 2.470s, Decode+Unpack: 2.225s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0764 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample179-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst (29/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,404B, BPFP=0.8792 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,712B, BPFP=1.2031 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,704B, BPFP=0.4702 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,932B, BPFP=0.1103 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014575 layer.1.conv_state 0.00051408 1.47434163 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00013299 0.27327842 layer.3.ssm_state 0.00000001 0.00001550 layer.3.conv_state 0.00006990 0.25972480 layer.4.ssm_state 0.00000003 0.00001761 layer.4.conv_state 0.00024394 0.49691844 layer.4.output 0.00000276 0.00154793 ------------------------------------------------------------------------------------- TOTAL 0.00003002 0.07654483 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 105996 BPFP 0.6788 bits/point EBPFP 1.3195 equivalent bits/point MSE 0.076545 ---------------------- -------------------------------------------------------- Time: 4.709s Load: 0.008s, Pack+Encode: 2.469s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0765 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample180-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst (30/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 98, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,148B, BPFP=0.8025 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,548B, BPFP=0.5217 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 15,404B, BPFP=0.3070 ⌛️ [2/4] FRONTEND: Frontend time: 2.478s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.226s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 98, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000029 0.00014350 layer.1.conv_state 0.00052074 1.47363210 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00013395 0.27548620 layer.3.ssm_state 0.00000001 0.00001532 layer.3.conv_state 0.00007102 0.26136476 layer.4.ssm_state 0.00000002 0.00001714 layer.4.conv_state 0.00024289 0.48068601 layer.4.output 0.00000300 0.00167183 ------------------------------------------------------------------------------------- TOTAL 0.00003095 0.07799541 (elements=1,220,608) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1220608 Total Bytes 115020 BPFP 0.7539 bits/point EBPFP 1.4067 equivalent bits/point MSE 0.077995 ---------------------- -------------------------------------------------------- Time: 4.712s Load: 0.008s, Pack+Encode: 2.478s, Decode+Unpack: 2.226s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0780 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample183-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst (31/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,852B, BPFP=0.9065 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,784B, BPFP=0.4751 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 10,380B, BPFP=0.2048 ⌛️ [2/4] FRONTEND: Frontend time: 2.494s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014406 layer.1.conv_state 0.00050627 1.47624385 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00013355 0.27403793 layer.3.ssm_state 0.00000001 0.00001545 layer.3.conv_state 0.00006878 0.25958627 layer.4.ssm_state 0.00000005 0.00001739 layer.4.conv_state 0.00026388 0.48316619 layer.4.output 0.00000306 0.00162577 ------------------------------------------------------------------------------------- TOTAL 0.00003096 0.07777490 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 110960 BPFP 0.7248 bits/point EBPFP 1.3818 equivalent bits/point MSE 0.077775 ---------------------- -------------------------------------------------------- Time: 4.736s Load: 0.008s, Pack+Encode: 2.494s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0778 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample185-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst (32/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,176B, BPFP=0.8652 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,936B, BPFP=0.4844 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,888B, BPFP=0.9492 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 10,628B, BPFP=0.2076 ⌛️ [2/4] FRONTEND: Frontend time: 2.493s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.239s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014342 layer.1.conv_state 0.00050939 1.47218215 layer.2.ssm_state 0.00000001 0.00001415 layer.2.conv_state 0.00012181 0.27306125 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00007190 0.25980631 layer.4.ssm_state 0.00000003 0.00001775 layer.4.conv_state 0.00024002 0.49764338 layer.4.output 0.00000295 0.00160288 ------------------------------------------------------------------------------------- TOTAL 0.00003005 0.07777099 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 110652 BPFP 0.7204 bits/point EBPFP 1.3716 equivalent bits/point MSE 0.077771 ---------------------- -------------------------------------------------------- Time: 4.742s Load: 0.009s, Pack+Encode: 2.493s, Decode+Unpack: 2.239s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0778 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample187-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst (33/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,400B, BPFP=0.8179 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,596B, BPFP=0.4636 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,888B, BPFP=0.9492 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,612B, BPFP=0.1682 ⌛️ [2/4] FRONTEND: Frontend time: 2.477s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.237s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014368 layer.1.conv_state 0.00050793 1.47327662 layer.2.ssm_state 0.00000001 0.00001414 layer.2.conv_state 0.00012495 0.27339956 layer.3.ssm_state 0.00000001 0.00001539 layer.3.conv_state 0.00006861 0.26043713 layer.4.ssm_state 0.00000002 0.00001787 layer.4.conv_state 0.00029770 0.49851036 layer.4.output 0.00000302 0.00161776 ------------------------------------------------------------------------------------- TOTAL 0.00003157 0.07785413 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 107524 BPFP 0.7000 bits/point EBPFP 1.3440 equivalent bits/point MSE 0.077854 ---------------------- -------------------------------------------------------- Time: 4.722s Load: 0.008s, Pack+Encode: 2.477s, Decode+Unpack: 2.237s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample19-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst (34/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,680B, BPFP=0.8960 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,172B, BPFP=0.4988 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,872B, BPFP=0.1928 ⌛️ [2/4] FRONTEND: Frontend time: 2.468s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.237s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000029 0.00014358 layer.1.conv_state 0.00050905 1.47396147 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00011183 0.27274242 layer.3.ssm_state 0.00000001 0.00001543 layer.3.conv_state 0.00006817 0.25907156 layer.4.ssm_state 0.00000005 0.00001746 layer.4.conv_state 0.00022612 0.48202175 layer.4.output 0.00000307 0.00161369 ------------------------------------------------------------------------------------- TOTAL 0.00002936 0.07737735 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 110632 BPFP 0.7203 bits/point EBPFP 1.3762 equivalent bits/point MSE 0.077377 ---------------------- -------------------------------------------------------- Time: 4.714s Load: 0.008s, Pack+Encode: 2.468s, Decode+Unpack: 2.237s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0774 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample196-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst (35/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,288B, BPFP=0.8110 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,292B, BPFP=0.5061 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,116B, BPFP=0.2440 ⌛️ [2/4] FRONTEND: Frontend time: 2.485s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014269 layer.1.conv_state 0.00049482 1.47542644 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00013819 0.27339572 layer.3.ssm_state 0.00000001 0.00001506 layer.3.conv_state 0.00007115 0.26108196 layer.4.ssm_state 0.00000002 0.00001762 layer.4.conv_state 0.00022532 0.48523510 layer.4.output 0.00000291 0.00156411 ------------------------------------------------------------------------------------- TOTAL 0.00002923 0.07631071 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 112644 BPFP 0.7213 bits/point EBPFP 1.3587 equivalent bits/point MSE 0.076311 ---------------------- -------------------------------------------------------- Time: 4.726s Load: 0.008s, Pack+Encode: 2.485s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0763 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample2-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst (36/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,260B, BPFP=0.8704 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,452B, BPFP=0.4548 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,888B, BPFP=0.9492 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,888B, BPFP=0.0918 ⌛️ [2/4] FRONTEND: Frontend time: 2.489s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.235s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014130 layer.1.conv_state 0.00050941 1.47769225 layer.2.ssm_state 0.00000001 0.00001415 layer.2.conv_state 0.00014139 0.27267337 layer.3.ssm_state 0.00000001 0.00001510 layer.3.conv_state 0.00007401 0.25915405 layer.4.ssm_state 0.00000002 0.00001751 layer.4.conv_state 0.00025094 0.48858967 layer.4.output 0.00000282 0.00150924 ------------------------------------------------------------------------------------- TOTAL 0.00003050 0.07661583 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 104492 BPFP 0.6713 bits/point EBPFP 1.3113 equivalent bits/point MSE 0.076616 ---------------------- -------------------------------------------------------- Time: 4.732s Load: 0.008s, Pack+Encode: 2.489s, Decode+Unpack: 2.235s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0766 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample21-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst (37/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,568B, BPFP=0.8892 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,736B, BPFP=0.4722 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,920B, BPFP=0.1487 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.238s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000947 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014313 layer.1.conv_state 0.00050400 1.47378957 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00013267 0.27103606 layer.3.ssm_state 0.00000001 0.00001510 layer.3.conv_state 0.00007191 0.25795862 layer.4.ssm_state 0.00000002 0.00001755 layer.4.conv_state 0.00022821 0.48721528 layer.4.output 0.00000291 0.00158025 ------------------------------------------------------------------------------------- TOTAL 0.00002951 0.07642691 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 108168 BPFP 0.6950 bits/point EBPFP 1.3390 equivalent bits/point MSE 0.076427 ---------------------- -------------------------------------------------------- Time: 4.714s Load: 0.007s, Pack+Encode: 2.469s, Decode+Unpack: 2.238s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0764 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample22-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst (38/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,784B, BPFP=0.8413 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,976B, BPFP=0.4868 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,860B, BPFP=0.1327 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.243s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014156 layer.1.conv_state 0.00051497 1.47252786 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00014786 0.27255732 layer.3.ssm_state 0.00000001 0.00001544 layer.3.conv_state 0.00006809 0.26121411 layer.4.ssm_state 0.00000002 0.00001760 layer.4.conv_state 0.00024536 0.49415204 layer.4.output 0.00000298 0.00158446 ------------------------------------------------------------------------------------- TOTAL 0.00003085 0.07745196 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 106544 BPFP 0.6913 bits/point EBPFP 1.3382 equivalent bits/point MSE 0.077452 ---------------------- -------------------------------------------------------- Time: 4.729s Load: 0.007s, Pack+Encode: 2.479s, Decode+Unpack: 2.243s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0775 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample220-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst (39/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 109, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,408B, BPFP=0.8794 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,076B, BPFP=0.4929 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,744B, BPFP=0.9141 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,924B, BPFP=0.9580 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,708B, BPFP=0.2456 ⌛️ [2/4] FRONTEND: Frontend time: 2.480s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 109, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014468 layer.1.conv_state 0.00052078 1.47479999 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00012449 0.27514470 layer.3.ssm_state 0.00000001 0.00001540 layer.3.conv_state 0.00006985 0.26206014 layer.4.ssm_state 0.00000002 0.00001739 layer.4.conv_state 0.00025944 0.49281299 layer.4.output 0.00000291 0.00156226 ------------------------------------------------------------------------------------- TOTAL 0.00003006 0.07559326 (elements=1,265,664) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1265664 Total Bytes 114120 BPFP 0.7213 bits/point EBPFP 1.3560 equivalent bits/point MSE 0.075593 ---------------------- -------------------------------------------------------- Time: 4.718s Load: 0.009s, Pack+Encode: 2.480s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0756 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample23-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst (40/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 96, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,428B, BPFP=0.8806 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,804B, BPFP=0.9287 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,176B, BPFP=0.4990 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,684B, BPFP=1.2014 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 11,036B, BPFP=0.2245 ⌛️ [2/4] FRONTEND: Frontend time: 2.471s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 96, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.230s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 96, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014314 layer.1.conv_state 0.00050897 1.47427320 layer.2.ssm_state 0.00000001 0.00001413 layer.2.conv_state 0.00013010 0.27474624 layer.3.ssm_state 0.00000001 0.00001561 layer.3.conv_state 0.00007001 0.26124826 layer.4.ssm_state 0.00000002 0.00001759 layer.4.conv_state 0.00022365 0.49558392 layer.4.output 0.00000328 0.00168695 ------------------------------------------------------------------------------------- TOTAL 0.00003024 0.07891289 (elements=1,212,416) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1212416 Total Bytes 111612 BPFP 0.7365 bits/point EBPFP 1.4001 equivalent bits/point MSE 0.078913 ---------------------- -------------------------------------------------------- Time: 4.709s Load: 0.008s, Pack+Encode: 2.471s, Decode+Unpack: 2.230s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0789 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample236-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst (41/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,852B, BPFP=0.8455 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,832B, BPFP=0.9355 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,780B, BPFP=0.4749 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,216B, BPFP=0.1202 ⌛️ [2/4] FRONTEND: Frontend time: 2.480s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.244s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014373 layer.1.conv_state 0.00050621 1.47265506 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00012991 0.27173370 layer.3.ssm_state 0.00000001 0.00001496 layer.3.conv_state 0.00007072 0.25764385 layer.4.ssm_state 0.00000001 0.00001743 layer.4.conv_state 0.00026898 0.48092294 layer.4.output 0.00000300 0.00157959 ------------------------------------------------------------------------------------- TOTAL 0.00003084 0.07698548 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 105820 BPFP 0.6866 bits/point EBPFP 1.3330 equivalent bits/point MSE 0.076985 ---------------------- -------------------------------------------------------- Time: 4.732s Load: 0.008s, Pack+Encode: 2.480s, Decode+Unpack: 2.244s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0770 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample24-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst (42/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 106, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,932B, BPFP=0.8503 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,424B, BPFP=0.4531 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,676B, BPFP=0.1230 ⌛️ [2/4] FRONTEND: Frontend time: 2.480s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014323 layer.1.conv_state 0.00051084 1.47441733 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00012173 0.27234930 layer.3.ssm_state 0.00000001 0.00001540 layer.3.conv_state 0.00007090 0.26004303 layer.4.ssm_state 0.00000002 0.00001730 layer.4.conv_state 0.00021893 0.48764369 layer.4.output 0.00000300 0.00154590 ------------------------------------------------------------------------------------- TOTAL 0.00002901 0.07604227 (elements=1,253,376) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1253376 Total Bytes 105928 BPFP 0.6761 bits/point EBPFP 1.3096 equivalent bits/point MSE 0.076042 ---------------------- -------------------------------------------------------- Time: 4.723s Load: 0.008s, Pack+Encode: 2.480s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0760 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample28-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst (43/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,312B, BPFP=0.8735 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,512B, BPFP=0.5195 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,920B, BPFP=0.9570 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,080B, BPFP=0.0973 ⌛️ [2/4] FRONTEND: Frontend time: 2.486s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.237s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014355 layer.1.conv_state 0.00051310 1.47652769 layer.2.ssm_state 0.00000001 0.00001403 layer.2.conv_state 0.00013079 0.27447814 layer.3.ssm_state 0.00000001 0.00001543 layer.3.conv_state 0.00006987 0.26214483 layer.4.ssm_state 0.00000001 0.00001749 layer.4.conv_state 0.00025978 0.48615310 layer.4.output 0.00000300 0.00157949 ------------------------------------------------------------------------------------- TOTAL 0.00003069 0.07716885 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 105816 BPFP 0.6843 bits/point EBPFP 1.3358 equivalent bits/point MSE 0.077169 ---------------------- -------------------------------------------------------- Time: 4.732s Load: 0.009s, Pack+Encode: 2.486s, Decode+Unpack: 2.237s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample29-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst (44/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 108, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,380B, BPFP=0.8777 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,692B, BPFP=0.4695 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,872B, BPFP=0.9453 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,680B, BPFP=0.1208 ⌛️ [2/4] FRONTEND: Frontend time: 2.489s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014260 layer.1.conv_state 0.00051348 1.47434330 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00012963 0.27362439 layer.3.ssm_state 0.00000001 0.00001554 layer.3.conv_state 0.00007448 0.26228580 layer.4.ssm_state 0.00000002 0.00001755 layer.4.conv_state 0.00025769 0.49825135 layer.4.output 0.00000283 0.00146417 ------------------------------------------------------------------------------------- TOTAL 0.00003014 0.07589482 (elements=1,261,568) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1261568 Total Bytes 106632 BPFP 0.6762 bits/point EBPFP 1.3100 equivalent bits/point MSE 0.075895 ---------------------- -------------------------------------------------------- Time: 4.726s Load: 0.008s, Pack+Encode: 2.489s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0759 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample30-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst (45/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 107, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,416B, BPFP=0.8799 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,836B, BPFP=0.4783 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,676B, BPFP=0.1584 ⌛️ [2/4] FRONTEND: Frontend time: 2.472s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.235s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000025 0.00014253 layer.1.conv_state 0.00050435 1.47051394 layer.2.ssm_state 0.00000001 0.00001416 layer.2.conv_state 0.00013424 0.27107036 layer.3.ssm_state 0.00000001 0.00001500 layer.3.conv_state 0.00007347 0.25812900 layer.4.ssm_state 0.00000005 0.00001744 layer.4.conv_state 0.00029085 0.47801590 layer.4.output 0.00000286 0.00150747 ------------------------------------------------------------------------------------- TOTAL 0.00003097 0.07535031 (elements=1,257,472) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1257472 Total Bytes 108872 BPFP 0.6926 bits/point EBPFP 1.3301 equivalent bits/point MSE 0.075350 ---------------------- -------------------------------------------------------- Time: 4.716s Load: 0.009s, Pack+Encode: 2.472s, Decode+Unpack: 2.235s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0754 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample31-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst (46/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,464B, BPFP=0.8828 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,812B, BPFP=0.4768 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,924B, BPFP=0.9580 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,672B, BPFP=0.0869 ⌛️ [2/4] FRONTEND: Frontend time: 2.471s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014357 layer.1.conv_state 0.00051859 1.47360182 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00013121 0.27323821 layer.3.ssm_state 0.00000001 0.00001558 layer.3.conv_state 0.00006790 0.26049435 layer.4.ssm_state 0.00000004 0.00001747 layer.4.conv_state 0.00024976 0.48773751 layer.4.output 0.00000299 0.00151775 ------------------------------------------------------------------------------------- TOTAL 0.00003026 0.07629312 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 104876 BPFP 0.6716 bits/point EBPFP 1.3133 equivalent bits/point MSE 0.076293 ---------------------- -------------------------------------------------------- Time: 4.713s Load: 0.009s, Pack+Encode: 2.471s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0763 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample32-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst (47/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 107, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,916B, BPFP=0.9104 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,796B, BPFP=0.9268 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,780B, BPFP=0.4749 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,816B, BPFP=0.1427 ⌛️ [2/4] FRONTEND: Frontend time: 2.477s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.227s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000947 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014197 layer.1.conv_state 0.00050815 1.47292566 layer.2.ssm_state 0.00000001 0.00001407 layer.2.conv_state 0.00013662 0.27079859 layer.3.ssm_state 0.00000001 0.00001540 layer.3.conv_state 0.00006925 0.25712094 layer.4.ssm_state 0.00000002 0.00001742 layer.4.conv_state 0.00024406 0.47633177 layer.4.output 0.00000278 0.00151173 ------------------------------------------------------------------------------------- TOTAL 0.00002977 0.07533738 (elements=1,257,472) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1257472 Total Bytes 108452 BPFP 0.6900 bits/point EBPFP 1.3302 equivalent bits/point MSE 0.075337 ---------------------- -------------------------------------------------------- Time: 4.712s Load: 0.007s, Pack+Encode: 2.477s, Decode+Unpack: 2.227s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0753 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample33-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst (48/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,292B, BPFP=0.8723 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,804B, BPFP=0.9287 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,168B, BPFP=0.4985 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,504B, BPFP=0.1245 ⌛️ [2/4] FRONTEND: Frontend time: 2.483s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.227s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014289 layer.1.conv_state 0.00050924 1.47228229 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00010465 0.27177799 layer.3.ssm_state 0.00000001 0.00001566 layer.3.conv_state 0.00006887 0.25865257 layer.4.ssm_state 0.00000002 0.00001755 layer.4.conv_state 0.00021814 0.49543121 layer.4.output 0.00000296 0.00157321 ------------------------------------------------------------------------------------- TOTAL 0.00002875 0.07713598 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 106884 BPFP 0.6913 bits/point EBPFP 1.3404 equivalent bits/point MSE 0.077136 ---------------------- -------------------------------------------------------- Time: 4.718s Load: 0.008s, Pack+Encode: 2.483s, Decode+Unpack: 2.227s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0771 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample36-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst (49/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 106, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,008B, BPFP=0.7939 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,796B, BPFP=0.9268 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,496B, BPFP=0.5186 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,552B, BPFP=0.1023 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.232s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014263 layer.1.conv_state 0.00051009 1.47747636 layer.2.ssm_state 0.00000001 0.00001392 layer.2.conv_state 0.00012190 0.27334943 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00011307 0.25899377 layer.4.ssm_state 0.00000002 0.00001712 layer.4.conv_state 0.00021713 0.47742856 layer.4.output 0.00000285 0.00150020 ------------------------------------------------------------------------------------- TOTAL 0.00003000 0.07583796 (elements=1,253,376) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1253376 Total Bytes 104948 BPFP 0.6699 bits/point EBPFP 1.3043 equivalent bits/point MSE 0.075838 ---------------------- -------------------------------------------------------- Time: 4.710s Load: 0.009s, Pack+Encode: 2.469s, Decode+Unpack: 2.232s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0758 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample37-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst (50/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,356B, BPFP=0.8152 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,436B, BPFP=0.5149 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,344B, BPFP=0.1004 ⌛️ [2/4] FRONTEND: Frontend time: 2.464s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.228s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014217 layer.1.conv_state 0.00050753 1.47783065 layer.2.ssm_state 0.00000001 0.00001401 layer.2.conv_state 0.00011558 0.27382648 layer.3.ssm_state 0.00000001 0.00001513 layer.3.conv_state 0.00006813 0.26022717 layer.4.ssm_state 0.00000002 0.00001710 layer.4.conv_state 0.00021580 0.47795105 layer.4.output 0.00000303 0.00152691 ------------------------------------------------------------------------------------- TOTAL 0.00002878 0.07640418 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 105068 BPFP 0.6750 bits/point EBPFP 1.3157 equivalent bits/point MSE 0.076404 ---------------------- -------------------------------------------------------- Time: 4.701s Load: 0.009s, Pack+Encode: 2.464s, Decode+Unpack: 2.228s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0764 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample38-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst (51/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,712B, BPFP=0.8369 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,828B, BPFP=0.9346 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,752B, BPFP=0.4731 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,616B, BPFP=1.1973 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,296B, BPFP=0.1194 ⌛️ [2/4] FRONTEND: Frontend time: 2.471s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014204 layer.1.conv_state 0.00050918 1.47291291 layer.2.ssm_state 0.00000001 0.00001402 layer.2.conv_state 0.00013447 0.27090123 layer.3.ssm_state 0.00000001 0.00001523 layer.3.conv_state 0.00006924 0.25899243 layer.4.ssm_state 0.00000002 0.00001756 layer.4.conv_state 0.00025518 0.48867384 layer.4.output 0.00000276 0.00154155 ------------------------------------------------------------------------------------- TOTAL 0.00003039 0.07669975 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 105732 BPFP 0.6815 bits/point EBPFP 1.3225 equivalent bits/point MSE 0.076700 ---------------------- -------------------------------------------------------- Time: 4.710s Load: 0.009s, Pack+Encode: 2.471s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0767 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample39-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst (52/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 110, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 110, 4096]) -> torch.Size([1, 1, 110, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,404B, BPFP=0.8792 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,576B, BPFP=0.4624 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,392B, BPFP=0.2378 ⌛️ [2/4] FRONTEND: Frontend time: 2.466s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 110, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.222s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 110, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014278 layer.1.conv_state 0.00050842 1.47217965 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00013835 0.27331504 layer.3.ssm_state 0.00000001 0.00001558 layer.3.conv_state 0.00007010 0.26175806 layer.4.ssm_state 0.00000002 0.00001768 layer.4.conv_state 0.00024386 0.49360162 layer.4.output 0.00000300 0.00156432 ------------------------------------------------------------------------------------- TOTAL 0.00002965 0.07525276 (elements=1,269,760) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1269760 Total Bytes 113280 BPFP 0.7137 bits/point EBPFP 1.3430 equivalent bits/point MSE 0.075253 ---------------------- -------------------------------------------------------- Time: 4.697s Load: 0.009s, Pack+Encode: 2.466s, Decode+Unpack: 2.222s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0753 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample4-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst (53/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 11,976B, BPFP=0.7310 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,300B, BPFP=0.5066 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 16,160B, BPFP=0.3188 ⌛️ [2/4] FRONTEND: Frontend time: 2.470s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.247s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014340 layer.1.conv_state 0.00049945 1.48106015 layer.2.ssm_state 0.00000001 0.00001397 layer.2.conv_state 0.00012969 0.27439851 layer.3.ssm_state 0.00000001 0.00001558 layer.3.conv_state 0.00007002 0.26115561 layer.4.ssm_state 0.00000003 0.00001753 layer.4.conv_state 0.00023573 0.49924630 layer.4.output 0.00000289 0.00167942 ------------------------------------------------------------------------------------- TOTAL 0.00002990 0.07840335 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 114352 BPFP 0.7470 bits/point EBPFP 1.3884 equivalent bits/point MSE 0.078403 ---------------------- -------------------------------------------------------- Time: 4.725s Load: 0.008s, Pack+Encode: 2.470s, Decode+Unpack: 2.247s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0784 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample40-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst (54/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,320B, BPFP=0.8740 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,080B, BPFP=0.4932 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,520B, BPFP=0.1631 ⌛️ [2/4] FRONTEND: Frontend time: 2.457s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.223s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014423 layer.1.conv_state 0.00051118 1.47475946 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00014206 0.27296177 layer.3.ssm_state 0.00000001 0.00001567 layer.3.conv_state 0.00007124 0.26044461 layer.4.ssm_state 0.00000002 0.00001774 layer.4.conv_state 0.00028525 0.49723291 layer.4.output 0.00000298 0.00161884 ------------------------------------------------------------------------------------- TOTAL 0.00003164 0.07734373 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 108848 BPFP 0.7040 bits/point EBPFP 1.3528 equivalent bits/point MSE 0.077344 ---------------------- -------------------------------------------------------- Time: 4.688s Load: 0.008s, Pack+Encode: 2.457s, Decode+Unpack: 2.223s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0773 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample41-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst (55/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.010s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,360B, BPFP=0.8765 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,636B, BPFP=0.4661 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,616B, BPFP=1.1973 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,600B, BPFP=0.1680 ⌛️ [2/4] FRONTEND: Frontend time: 2.475s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.235s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000025 0.00014275 layer.1.conv_state 0.00050676 1.46758056 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00012238 0.27117342 layer.3.ssm_state 0.00000001 0.00001547 layer.3.conv_state 0.00007028 0.25890809 layer.4.ssm_state 0.00000002 0.00001740 layer.4.conv_state 0.00023202 0.48207095 layer.4.output 0.00000314 0.00159786 ------------------------------------------------------------------------------------- TOTAL 0.00002981 0.07715694 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 108528 BPFP 0.7066 bits/point EBPFP 1.3571 equivalent bits/point MSE 0.077157 ---------------------- -------------------------------------------------------- Time: 4.720s Load: 0.010s, Pack+Encode: 2.475s, Decode+Unpack: 2.235s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample43-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst (56/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.010s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,788B, BPFP=0.8416 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,576B, BPFP=0.5234 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,620B, BPFP=1.1975 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,044B, BPFP=0.1418 ⌛️ [2/4] FRONTEND: Frontend time: 2.474s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014260 layer.1.conv_state 0.00051299 1.47350800 layer.2.ssm_state 0.00000001 0.00001401 layer.2.conv_state 0.00011946 0.27356440 layer.3.ssm_state 0.00000001 0.00001538 layer.3.conv_state 0.00007081 0.25916332 layer.4.ssm_state 0.00000002 0.00001751 layer.4.conv_state 0.00021939 0.49194941 layer.4.output 0.00000316 0.00166359 ------------------------------------------------------------------------------------- TOTAL 0.00002984 0.07843863 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 107348 BPFP 0.7059 bits/point EBPFP 1.3656 equivalent bits/point MSE 0.078439 ---------------------- -------------------------------------------------------- Time: 4.717s Load: 0.010s, Pack+Encode: 2.474s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0784 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample44-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst (57/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 107, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,180B, BPFP=0.8655 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,704B, BPFP=0.4702 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,744B, BPFP=0.9141 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 12,532B, BPFP=0.2288 ⌛️ [2/4] FRONTEND: Frontend time: 2.481s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.236s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 107, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000029 0.00014479 layer.1.conv_state 0.00050544 1.47597098 layer.2.ssm_state 0.00000001 0.00001412 layer.2.conv_state 0.00013757 0.27270523 layer.3.ssm_state 0.00000001 0.00001581 layer.3.conv_state 0.00007295 0.26112565 layer.4.ssm_state 0.00000001 0.00001789 layer.4.conv_state 0.00026139 0.50408578 layer.4.output 0.00000288 0.00156458 ------------------------------------------------------------------------------------- TOTAL 0.00003031 0.07631282 (elements=1,257,472) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1257472 Total Bytes 112316 BPFP 0.7146 bits/point EBPFP 1.3494 equivalent bits/point MSE 0.076313 ---------------------- -------------------------------------------------------- Time: 4.726s Load: 0.009s, Pack+Encode: 2.481s, Decode+Unpack: 2.236s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0763 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample45-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst (58/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 106, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 11,868B, BPFP=0.7244 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,804B, BPFP=0.9287 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,832B, BPFP=0.4780 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,568B, BPFP=1.1943 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,708B, BPFP=0.1420 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014332 layer.1.conv_state 0.00050804 1.47724128 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00011728 0.27318445 layer.3.ssm_state 0.00000001 0.00001500 layer.3.conv_state 0.00007019 0.26030621 layer.4.ssm_state 0.00000002 0.00001724 layer.4.conv_state 0.00023224 0.48319361 layer.4.output 0.00000302 0.00151951 ------------------------------------------------------------------------------------- TOTAL 0.00002915 0.07601928 (elements=1,253,376) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1253376 Total Bytes 105268 BPFP 0.6719 bits/point EBPFP 1.2946 equivalent bits/point MSE 0.076019 ---------------------- -------------------------------------------------------- Time: 4.719s Load: 0.009s, Pack+Encode: 2.479s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0760 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample46-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst (59/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,788B, BPFP=0.8416 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,796B, BPFP=0.9268 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,568B, BPFP=0.5229 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,744B, BPFP=0.9141 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,616B, BPFP=1.1973 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,180B, BPFP=0.1244 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014302 layer.1.conv_state 0.00051689 1.47416401 layer.2.ssm_state 0.00000001 0.00001403 layer.2.conv_state 0.00012124 0.27418613 layer.3.ssm_state 0.00000001 0.00001537 layer.3.conv_state 0.00007153 0.25977954 layer.4.ssm_state 0.00000002 0.00001753 layer.4.conv_state 0.00021058 0.49154568 layer.4.output 0.00000315 0.00164928 ------------------------------------------------------------------------------------- TOTAL 0.00002977 0.07847414 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 106464 BPFP 0.7001 bits/point EBPFP 1.3596 equivalent bits/point MSE 0.078474 ---------------------- -------------------------------------------------------- Time: 4.717s Load: 0.008s, Pack+Encode: 2.479s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0785 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample47-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst (60/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,292B, BPFP=0.8723 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,756B, BPFP=0.4734 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,612B, BPFP=1.1970 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,780B, BPFP=0.9229 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 10,968B, BPFP=0.2164 ⌛️ [2/4] FRONTEND: Frontend time: 2.467s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014235 layer.1.conv_state 0.00050257 1.47397375 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013543 0.27260083 layer.3.ssm_state 0.00000001 0.00001536 layer.3.conv_state 0.00007075 0.26032463 layer.4.ssm_state 0.00000002 0.00001776 layer.4.conv_state 0.00024175 0.49236053 layer.4.output 0.00000314 0.00163535 ------------------------------------------------------------------------------------- TOTAL 0.00003040 0.07794449 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 110968 BPFP 0.7249 bits/point EBPFP 1.3781 equivalent bits/point MSE 0.077944 ---------------------- -------------------------------------------------------- Time: 4.710s Load: 0.009s, Pack+Encode: 2.467s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample49-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst (61/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 15,112B, BPFP=0.9224 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,692B, BPFP=1.2019 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,364B, BPFP=0.4495 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,284B, BPFP=0.0820 ⌛️ [2/4] FRONTEND: Frontend time: 2.505s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.239s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000948 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014260 layer.1.conv_state 0.00051037 1.47208929 layer.2.ssm_state 0.00000001 0.00001417 layer.2.conv_state 0.00013143 0.27227512 layer.3.ssm_state 0.00000001 0.00001527 layer.3.conv_state 0.00007554 0.25975448 layer.4.ssm_state 0.00000002 0.00001749 layer.4.conv_state 0.00025897 0.48872408 layer.4.output 0.00000311 0.00156014 ------------------------------------------------------------------------------------- TOTAL 0.00003080 0.07699107 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 104684 BPFP 0.6770 bits/point EBPFP 1.3263 equivalent bits/point MSE 0.076991 ---------------------- -------------------------------------------------------- Time: 4.752s Load: 0.008s, Pack+Encode: 2.505s, Decode+Unpack: 2.239s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0770 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample50-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst (62/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 15,280B, BPFP=0.9326 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,796B, BPFP=0.4758 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,332B, BPFP=0.0992 ⌛️ [2/4] FRONTEND: Frontend time: 2.487s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.239s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014343 layer.1.conv_state 0.00051108 1.47743452 layer.2.ssm_state 0.00000001 0.00001414 layer.2.conv_state 0.00013025 0.27400202 layer.3.ssm_state 0.00000001 0.00001579 layer.3.conv_state 0.00006983 0.26047099 layer.4.ssm_state 0.00000002 0.00001794 layer.4.conv_state 0.00023909 0.50717103 layer.4.output 0.00000287 0.00155787 ------------------------------------------------------------------------------------- TOTAL 0.00002976 0.07693668 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 106340 BPFP 0.6810 bits/point EBPFP 1.3278 equivalent bits/point MSE 0.076937 ---------------------- -------------------------------------------------------- Time: 4.735s Load: 0.008s, Pack+Encode: 2.487s, Decode+Unpack: 2.239s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0769 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample51-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst (63/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,272B, BPFP=0.8711 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 9,260B, BPFP=0.5652 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,584B, BPFP=1.1953 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,036B, BPFP=0.0813 ⌛️ [2/4] FRONTEND: Frontend time: 2.485s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.227s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014252 layer.1.conv_state 0.00050795 1.47087073 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00011983 0.27230981 layer.3.ssm_state 0.00000001 0.00001539 layer.3.conv_state 0.00006952 0.25871733 layer.4.ssm_state 0.00000002 0.00001740 layer.4.conv_state 0.00020281 0.48091137 layer.4.output 0.00000316 0.00159608 ------------------------------------------------------------------------------------- TOTAL 0.00002923 0.07800241 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 105492 BPFP 0.6937 bits/point EBPFP 1.3609 equivalent bits/point MSE 0.078002 ---------------------- -------------------------------------------------------- Time: 4.721s Load: 0.009s, Pack+Encode: 2.485s, Decode+Unpack: 2.227s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0780 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample53-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst (64/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,896B, BPFP=0.8481 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,804B, BPFP=0.4763 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,884B, BPFP=0.9482 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,916B, BPFP=0.0914 ⌛️ [2/4] FRONTEND: Frontend time: 2.489s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014094 layer.1.conv_state 0.00050049 1.47851086 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013306 0.27137777 layer.3.ssm_state 0.00000001 0.00001559 layer.3.conv_state 0.00007056 0.25831482 layer.4.ssm_state 0.00000001 0.00001749 layer.4.conv_state 0.00023923 0.48678368 layer.4.output 0.00000281 0.00152990 ------------------------------------------------------------------------------------- TOTAL 0.00002958 0.07629480 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 104556 BPFP 0.6695 bits/point EBPFP 1.3076 equivalent bits/point MSE 0.076295 ---------------------- -------------------------------------------------------- Time: 4.730s Load: 0.007s, Pack+Encode: 2.489s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0763 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample54-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst (65/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,184B, BPFP=0.8657 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,296B, BPFP=0.5063 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,864B, BPFP=0.1583 ⌛️ [2/4] FRONTEND: Frontend time: 2.486s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014299 layer.1.conv_state 0.00050183 1.47708440 layer.2.ssm_state 0.00000001 0.00001417 layer.2.conv_state 0.00013844 0.27251312 layer.3.ssm_state 0.00000001 0.00001559 layer.3.conv_state 0.00006878 0.25945696 layer.4.ssm_state 0.00000002 0.00001775 layer.4.conv_state 0.00025869 0.49544215 layer.4.output 0.00000296 0.00165282 ------------------------------------------------------------------------------------- TOTAL 0.00003099 0.07860523 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 108260 BPFP 0.7119 bits/point EBPFP 1.3722 equivalent bits/point MSE 0.078605 ---------------------- -------------------------------------------------------- Time: 4.725s Load: 0.008s, Pack+Encode: 2.486s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0786 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample55-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst (66/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,564B, BPFP=0.8889 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,816B, BPFP=0.4771 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,644B, BPFP=1.1990 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,104B, BPFP=0.1146 ⌛️ [2/4] FRONTEND: Frontend time: 2.487s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.241s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014295 layer.1.conv_state 0.00050960 1.47130585 layer.2.ssm_state 0.00000001 0.00001420 layer.2.conv_state 0.00014477 0.27236101 layer.3.ssm_state 0.00000001 0.00001545 layer.3.conv_state 0.00007109 0.26028496 layer.4.ssm_state 0.00000003 0.00001769 layer.4.conv_state 0.00024741 0.49348545 layer.4.output 0.00000297 0.00154774 ------------------------------------------------------------------------------------- TOTAL 0.00003048 0.07661155 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 106364 BPFP 0.6834 bits/point EBPFP 1.3275 equivalent bits/point MSE 0.076612 ---------------------- -------------------------------------------------------- Time: 4.735s Load: 0.008s, Pack+Encode: 2.487s, Decode+Unpack: 2.241s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0766 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample56-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst (67/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 106, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 12,376B, BPFP=0.7554 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,176B, BPFP=0.4990 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,688B, BPFP=0.1601 ⌛️ [2/4] FRONTEND: Frontend time: 2.476s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 106, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014259 layer.1.conv_state 0.00051526 1.47429645 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013199 0.27365363 layer.3.ssm_state 0.00000001 0.00001516 layer.3.conv_state 0.00007059 0.26012099 layer.4.ssm_state 0.00000002 0.00001730 layer.4.conv_state 0.00023257 0.48468709 layer.4.output 0.00000278 0.00152140 ------------------------------------------------------------------------------------- TOTAL 0.00002965 0.07598936 (elements=1,253,376) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1253376 Total Bytes 107188 BPFP 0.6842 bits/point EBPFP 1.3129 equivalent bits/point MSE 0.075989 ---------------------- -------------------------------------------------------- Time: 4.718s Load: 0.009s, Pack+Encode: 2.476s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0760 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample57-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst (68/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,372B, BPFP=0.8772 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,112B, BPFP=0.4951 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,740B, BPFP=0.9131 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,916B, BPFP=0.1133 ⌛️ [2/4] FRONTEND: Frontend time: 2.486s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014329 layer.1.conv_state 0.00051277 1.47442019 layer.2.ssm_state 0.00000001 0.00001403 layer.2.conv_state 0.00013342 0.27275977 layer.3.ssm_state 0.00000001 0.00001561 layer.3.conv_state 0.00006746 0.25924838 layer.4.ssm_state 0.00000002 0.00001747 layer.4.conv_state 0.00024674 0.48931795 layer.4.output 0.00000297 0.00156909 ------------------------------------------------------------------------------------- TOTAL 0.00003035 0.07707110 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 106336 BPFP 0.6877 bits/point EBPFP 1.3372 equivalent bits/point MSE 0.077071 ---------------------- -------------------------------------------------------- Time: 4.729s Load: 0.008s, Pack+Encode: 2.486s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0771 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample58-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst (69/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,868B, BPFP=0.8464 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,688B, BPFP=0.4692 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,072B, BPFP=0.0820 ⌛️ [2/4] FRONTEND: Frontend time: 2.488s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.242s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000010 0.00014253 layer.1.conv_state 0.00050499 1.47361028 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013541 0.27259696 layer.3.ssm_state 0.00000001 0.00001539 layer.3.conv_state 0.00007246 0.25841448 layer.4.ssm_state 0.00000002 0.00001729 layer.4.conv_state 0.00022642 0.47937307 layer.4.output 0.00000327 0.00162711 ------------------------------------------------------------------------------------- TOTAL 0.00003032 0.07804446 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 103552 BPFP 0.6810 bits/point EBPFP 1.3352 equivalent bits/point MSE 0.078044 ---------------------- -------------------------------------------------------- Time: 4.738s Load: 0.008s, Pack+Encode: 2.488s, Decode+Unpack: 2.242s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0780 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample60-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst (70/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 11,192B, BPFP=0.6831 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,192B, BPFP=0.5000 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,552B, BPFP=0.1670 ⌛️ [2/4] FRONTEND: Frontend time: 2.484s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.244s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014293 layer.1.conv_state 0.00050447 1.47619557 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00013123 0.27449048 layer.3.ssm_state 0.00000001 0.00001507 layer.3.conv_state 0.00007682 0.25972670 layer.4.ssm_state 0.00000002 0.00001787 layer.4.conv_state 0.00025075 0.49677944 layer.4.output 0.00000309 0.00158387 ------------------------------------------------------------------------------------- TOTAL 0.00003063 0.07788454 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 105868 BPFP 0.6892 bits/point EBPFP 1.3228 equivalent bits/point MSE 0.077885 ---------------------- -------------------------------------------------------- Time: 4.736s Load: 0.009s, Pack+Encode: 2.484s, Decode+Unpack: 2.244s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample61-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst (71/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 11,876B, BPFP=0.7249 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,116B, BPFP=0.4954 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,784B, BPFP=0.9238 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,460B, BPFP=0.1457 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.241s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014287 layer.1.conv_state 0.00050765 1.47487462 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00013519 0.27463436 layer.3.ssm_state 0.00000001 0.00001507 layer.3.conv_state 0.00007639 0.25977966 layer.4.ssm_state 0.00000003 0.00001793 layer.4.conv_state 0.00024199 0.50011116 layer.4.output 0.00000307 0.00158255 ------------------------------------------------------------------------------------- TOTAL 0.00003057 0.07794297 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 105392 BPFP 0.6861 bits/point EBPFP 1.3237 equivalent bits/point MSE 0.077943 ---------------------- -------------------------------------------------------- Time: 4.728s Load: 0.008s, Pack+Encode: 2.479s, Decode+Unpack: 2.241s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample62-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst (72/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,284B, BPFP=0.8718 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,024B, BPFP=0.4897 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,852B, BPFP=0.1338 ⌛️ [2/4] FRONTEND: Frontend time: 2.481s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.236s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014185 layer.1.conv_state 0.00051429 1.47458386 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00012088 0.27229336 layer.3.ssm_state 0.00000001 0.00001555 layer.3.conv_state 0.00007035 0.26057190 layer.4.ssm_state 0.00000002 0.00001755 layer.4.conv_state 0.00025261 0.49083689 layer.4.output 0.00000308 0.00160772 ------------------------------------------------------------------------------------- TOTAL 0.00003049 0.07765489 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 107084 BPFP 0.6972 bits/point EBPFP 1.3497 equivalent bits/point MSE 0.077655 ---------------------- -------------------------------------------------------- Time: 4.726s Load: 0.008s, Pack+Encode: 2.481s, Decode+Unpack: 2.236s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0777 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample63-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst (73/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,268B, BPFP=0.8708 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,320B, BPFP=0.4468 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,912B, BPFP=0.9551 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 12,036B, BPFP=0.2351 ⌛️ [2/4] FRONTEND: Frontend time: 2.480s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.240s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014347 layer.1.conv_state 0.00051142 1.47435927 layer.2.ssm_state 0.00000001 0.00001405 layer.2.conv_state 0.00015094 0.27383426 layer.3.ssm_state 0.00000001 0.00001547 layer.3.conv_state 0.00006998 0.26136065 layer.4.ssm_state 0.00000002 0.00001778 layer.4.conv_state 0.00021745 0.49915931 layer.4.output 0.00000300 0.00158899 ------------------------------------------------------------------------------------- TOTAL 0.00003024 0.07792691 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 111592 BPFP 0.7265 bits/point EBPFP 1.3747 equivalent bits/point MSE 0.077927 ---------------------- -------------------------------------------------------- Time: 4.729s Load: 0.008s, Pack+Encode: 2.480s, Decode+Unpack: 2.240s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample65-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst (74/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,196B, BPFP=0.8665 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,688B, BPFP=0.4692 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,896B, BPFP=0.9512 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,420B, BPFP=0.1769 ⌛️ [2/4] FRONTEND: Frontend time: 2.474s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.240s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000026 0.00014300 layer.1.conv_state 0.00050859 1.47835934 layer.2.ssm_state 0.00000001 0.00001415 layer.2.conv_state 0.00015031 0.27389172 layer.3.ssm_state 0.00000001 0.00001564 layer.3.conv_state 0.00007130 0.26070893 layer.4.ssm_state 0.00000001 0.00001740 layer.4.conv_state 0.00023281 0.48352972 layer.4.output 0.00000304 0.00157074 ------------------------------------------------------------------------------------- TOTAL 0.00003026 0.07659448 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 109228 BPFP 0.7018 bits/point EBPFP 1.3430 equivalent bits/point MSE 0.076594 ---------------------- -------------------------------------------------------- Time: 4.723s Load: 0.009s, Pack+Encode: 2.474s, Decode+Unpack: 2.240s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0766 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample66-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst (75/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,400B, BPFP=0.8789 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,712B, BPFP=1.2031 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,172B, BPFP=0.4988 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,916B, BPFP=0.9561 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,776B, BPFP=0.1489 ⌛️ [2/4] FRONTEND: Frontend time: 2.484s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.235s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000014 0.00014430 layer.1.conv_state 0.00050679 1.47528481 layer.2.ssm_state 0.00000001 0.00001413 layer.2.conv_state 0.00013770 0.27328777 layer.3.ssm_state 0.00000001 0.00001524 layer.3.conv_state 0.00006920 0.26051366 layer.4.ssm_state 0.00000002 0.00001757 layer.4.conv_state 0.00024331 0.49825314 layer.4.output 0.00000290 0.00160169 ------------------------------------------------------------------------------------- TOTAL 0.00003022 0.07738930 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 108332 BPFP 0.7006 bits/point EBPFP 1.3509 equivalent bits/point MSE 0.077389 ---------------------- -------------------------------------------------------- Time: 4.727s Load: 0.008s, Pack+Encode: 2.484s, Decode+Unpack: 2.235s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0774 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample67-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst (76/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,232B, BPFP=0.8687 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,932B, BPFP=0.4841 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,884B, BPFP=0.9482 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,356B, BPFP=0.1409 ⌛️ [2/4] FRONTEND: Frontend time: 2.477s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.226s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014287 layer.1.conv_state 0.00050838 1.47569299 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00010671 0.27191162 layer.3.ssm_state 0.00000001 0.00001541 layer.3.conv_state 0.00007018 0.25849387 layer.4.ssm_state 0.00000002 0.00001747 layer.4.conv_state 0.00026429 0.49082482 layer.4.output 0.00000311 0.00159974 ------------------------------------------------------------------------------------- TOTAL 0.00003009 0.07711257 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 107448 BPFP 0.6949 bits/point EBPFP 1.3422 equivalent bits/point MSE 0.077113 ---------------------- -------------------------------------------------------- Time: 4.710s Load: 0.008s, Pack+Encode: 2.477s, Decode+Unpack: 2.226s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0771 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample68-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst (77/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,916B, BPFP=0.8494 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,936B, BPFP=0.4844 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,176B, BPFP=0.0972 ⌛️ [2/4] FRONTEND: Frontend time: 2.476s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.234s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014210 layer.1.conv_state 0.00051536 1.47549927 layer.2.ssm_state 0.00000001 0.00001405 layer.2.conv_state 0.00012850 0.27186891 layer.3.ssm_state 0.00000001 0.00001514 layer.3.conv_state 0.00007391 0.25928950 layer.4.ssm_state 0.00000003 0.00001778 layer.4.conv_state 0.00023144 0.49666166 layer.4.output 0.00000309 0.00152719 ------------------------------------------------------------------------------------- TOTAL 0.00002990 0.07675919 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 104968 BPFP 0.6744 bits/point EBPFP 1.3155 equivalent bits/point MSE 0.076759 ---------------------- -------------------------------------------------------- Time: 4.718s Load: 0.008s, Pack+Encode: 2.476s, Decode+Unpack: 2.234s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0768 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample7-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst (78/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,284B, BPFP=0.8718 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,824B, BPFP=0.4775 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,576B, BPFP=1.1948 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,984B, BPFP=0.1135 ⌛️ [2/4] FRONTEND: Frontend time: 2.489s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.236s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014368 layer.1.conv_state 0.00051718 1.47545767 layer.2.ssm_state 0.00000001 0.00001411 layer.2.conv_state 0.00013311 0.27269790 layer.3.ssm_state 0.00000001 0.00001514 layer.3.conv_state 0.00006873 0.25871187 layer.4.ssm_state 0.00000004 0.00001750 layer.4.conv_state 0.00022143 0.48851600 layer.4.output 0.00000284 0.00157244 ------------------------------------------------------------------------------------- TOTAL 0.00002967 0.07681347 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 106000 BPFP 0.6833 bits/point EBPFP 1.3280 equivalent bits/point MSE 0.076813 ---------------------- -------------------------------------------------------- Time: 4.733s Load: 0.008s, Pack+Encode: 2.489s, Decode+Unpack: 2.236s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0768 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample71-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst (79/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,732B, BPFP=0.8381 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,396B, BPFP=0.4514 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,536B, BPFP=0.1050 ⌛️ [2/4] FRONTEND: Frontend time: 2.487s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.242s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014264 layer.1.conv_state 0.00050321 1.46990502 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00014768 0.27380630 layer.3.ssm_state 0.00000001 0.00001554 layer.3.conv_state 0.00007730 0.26154354 layer.4.ssm_state 0.00000002 0.00001751 layer.4.conv_state 0.00022845 0.48941344 layer.4.output 0.00000298 0.00154935 ------------------------------------------------------------------------------------- TOTAL 0.00003014 0.07678667 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 104584 BPFP 0.6741 bits/point EBPFP 1.3126 equivalent bits/point MSE 0.076787 ---------------------- -------------------------------------------------------- Time: 4.738s Load: 0.009s, Pack+Encode: 2.487s, Decode+Unpack: 2.242s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0768 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample72-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst (80/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,368B, BPFP=0.8159 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,800B, BPFP=0.9277 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,976B, BPFP=0.4868 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,912B, BPFP=0.9551 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 12,680B, BPFP=0.2452 ⌛️ [2/4] FRONTEND: Frontend time: 2.495s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000011 0.00014380 layer.1.conv_state 0.00050786 1.47214735 layer.2.ssm_state 0.00000001 0.00001400 layer.2.conv_state 0.00013194 0.27260810 layer.3.ssm_state 0.00000001 0.00001551 layer.3.conv_state 0.00006897 0.25741938 layer.4.ssm_state 0.00000002 0.00001717 layer.4.conv_state 0.00028450 0.47699642 layer.4.output 0.00000302 0.00162478 ------------------------------------------------------------------------------------- TOTAL 0.00003131 0.07690009 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 111936 BPFP 0.7263 bits/point EBPFP 1.3704 equivalent bits/point MSE 0.076900 ---------------------- -------------------------------------------------------- Time: 4.734s Load: 0.007s, Pack+Encode: 2.495s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0769 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample74-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst (81/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 104, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,380B, BPFP=0.8777 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,896B, BPFP=0.4819 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,656B, BPFP=1.1997 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,928B, BPFP=0.9590 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 4,800B, BPFP=0.0901 ⌛️ [2/4] FRONTEND: Frontend time: 2.469s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 104, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014313 layer.1.conv_state 0.00051907 1.47624385 layer.2.ssm_state 0.00000001 0.00001408 layer.2.conv_state 0.00013541 0.27328900 layer.3.ssm_state 0.00000001 0.00001528 layer.3.conv_state 0.00006970 0.25975379 layer.4.ssm_state 0.00000001 0.00001730 layer.4.conv_state 0.00021955 0.48876140 layer.4.output 0.00000291 0.00155066 ------------------------------------------------------------------------------------- TOTAL 0.00002969 0.07662857 (elements=1,245,184) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1245184 Total Bytes 104996 BPFP 0.6746 bits/point EBPFP 1.3183 equivalent bits/point MSE 0.076629 ---------------------- -------------------------------------------------------- Time: 4.711s Load: 0.009s, Pack+Encode: 2.469s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0766 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample75-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst (82/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,824B, BPFP=0.8438 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,828B, BPFP=0.9346 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,364B, BPFP=0.4495 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,568B, BPFP=0.1463 ⌛️ [2/4] FRONTEND: Frontend time: 2.488s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.229s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014283 layer.1.conv_state 0.00050645 1.46971774 layer.2.ssm_state 0.00000001 0.00001405 layer.2.conv_state 0.00013674 0.27238637 layer.3.ssm_state 0.00000001 0.00001550 layer.3.conv_state 0.00006915 0.25988239 layer.4.ssm_state 0.00000001 0.00001755 layer.4.conv_state 0.00024526 0.48975474 layer.4.output 0.00000306 0.00159283 ------------------------------------------------------------------------------------- TOTAL 0.00003040 0.07722340 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 106708 BPFP 0.6924 bits/point EBPFP 1.3357 equivalent bits/point MSE 0.077223 ---------------------- -------------------------------------------------------- Time: 4.726s Load: 0.009s, Pack+Encode: 2.488s, Decode+Unpack: 2.229s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample78-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst (83/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,032B, BPFP=0.7954 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,816B, BPFP=0.4771 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,776B, BPFP=0.9219 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,116B, BPFP=0.1696 ⌛️ [2/4] FRONTEND: Frontend time: 2.482s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.232s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014219 layer.1.conv_state 0.00051340 1.47452497 layer.2.ssm_state 0.00000001 0.00001402 layer.2.conv_state 0.00013703 0.27298027 layer.3.ssm_state 0.00000001 0.00001538 layer.3.conv_state 0.00007042 0.26215267 layer.4.ssm_state 0.00000002 0.00001747 layer.4.conv_state 0.00024895 0.49050862 layer.4.output 0.00000292 0.00154178 ------------------------------------------------------------------------------------- TOTAL 0.00003030 0.07643485 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 107888 BPFP 0.6909 bits/point EBPFP 1.3234 equivalent bits/point MSE 0.076435 ---------------------- -------------------------------------------------------- Time: 4.723s Load: 0.009s, Pack+Encode: 2.482s, Decode+Unpack: 2.232s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0764 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample8-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst (84/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,236B, BPFP=0.8689 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,760B, BPFP=0.4736 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,992B, BPFP=0.1159 ⌛️ [2/4] FRONTEND: Frontend time: 2.479s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.239s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000010 0.00014161 layer.1.conv_state 0.00050642 1.47459519 layer.2.ssm_state 0.00000001 0.00001403 layer.2.conv_state 0.00011949 0.27247247 layer.3.ssm_state 0.00000001 0.00001529 layer.3.conv_state 0.00006817 0.25901693 layer.4.ssm_state 0.00000004 0.00001749 layer.4.conv_state 0.00024581 0.48485935 layer.4.output 0.00000308 0.00157435 ------------------------------------------------------------------------------------- TOTAL 0.00002991 0.07719585 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 105936 BPFP 0.6874 bits/point EBPFP 1.3359 equivalent bits/point MSE 0.077196 ---------------------- -------------------------------------------------------- Time: 4.724s Load: 0.007s, Pack+Encode: 2.479s, Decode+Unpack: 2.239s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0772 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample80-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst (85/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,896B, BPFP=0.8481 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,600B, BPFP=0.4639 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,920B, BPFP=0.9570 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,988B, BPFP=0.1970 ⌛️ [2/4] FRONTEND: Frontend time: 2.483s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.226s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000943 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014320 layer.1.conv_state 0.00050345 1.47363353 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00011822 0.27256536 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00007003 0.26118335 layer.4.ssm_state 0.00000003 0.00001766 layer.4.conv_state 0.00022051 0.49584177 layer.4.output 0.00000301 0.00164420 ------------------------------------------------------------------------------------- TOTAL 0.00002933 0.07805358 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 109464 BPFP 0.7150 bits/point EBPFP 1.3648 equivalent bits/point MSE 0.078054 ---------------------- -------------------------------------------------------- Time: 4.716s Load: 0.008s, Pack+Encode: 2.483s, Decode+Unpack: 2.226s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0781 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample81-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst (86/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,428B, BPFP=0.8806 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,768B, BPFP=0.4741 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,580B, BPFP=1.1951 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,888B, BPFP=0.9492 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,304B, BPFP=0.1399 ⌛️ [2/4] FRONTEND: Frontend time: 2.475s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.233s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014279 layer.1.conv_state 0.00050585 1.47054553 layer.2.ssm_state 0.00000001 0.00001413 layer.2.conv_state 0.00012980 0.27182943 layer.3.ssm_state 0.00000001 0.00001504 layer.3.conv_state 0.00006838 0.26044062 layer.4.ssm_state 0.00000001 0.00001752 layer.4.conv_state 0.00027100 0.48777050 layer.4.output 0.00000302 0.00158072 ------------------------------------------------------------------------------------- TOTAL 0.00003074 0.07693823 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 107388 BPFP 0.6945 bits/point EBPFP 1.3418 equivalent bits/point MSE 0.076938 ---------------------- -------------------------------------------------------- Time: 4.716s Load: 0.008s, Pack+Encode: 2.475s, Decode+Unpack: 2.233s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0769 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample83-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst (87/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 102, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,428B, BPFP=0.8806 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,196B, BPFP=0.5002 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,012B, BPFP=0.0960 ⌛️ [2/4] FRONTEND: Frontend time: 2.482s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.236s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 102, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000027 0.00014108 layer.1.conv_state 0.00051229 1.47643971 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00012754 0.27266398 layer.3.ssm_state 0.00000001 0.00001521 layer.3.conv_state 0.00007197 0.26014119 layer.4.ssm_state 0.00000004 0.00001757 layer.4.conv_state 0.00024522 0.48496470 layer.4.output 0.00000311 0.00155803 ------------------------------------------------------------------------------------- TOTAL 0.00003031 0.07702639 (elements=1,236,992) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1236992 Total Bytes 105580 BPFP 0.6828 bits/point EBPFP 1.3332 equivalent bits/point MSE 0.077026 ---------------------- -------------------------------------------------------- Time: 4.727s Load: 0.009s, Pack+Encode: 2.482s, Decode+Unpack: 2.236s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0770 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample84-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst (88/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 108, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,868B, BPFP=0.9075 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,492B, BPFP=0.4573 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,608B, BPFP=1.1968 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,648B, BPFP=1.1992 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,888B, BPFP=0.9492 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 5,452B, BPFP=0.0986 ⌛️ [2/4] FRONTEND: Frontend time: 2.481s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.232s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 108, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000946 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014284 layer.1.conv_state 0.00051405 1.47230911 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00011333 0.27182537 layer.3.ssm_state 0.00000001 0.00001560 layer.3.conv_state 0.00007471 0.25985163 layer.4.ssm_state 0.00000002 0.00001763 layer.4.conv_state 0.00029453 0.49567693 layer.4.output 0.00000297 0.00149967 ------------------------------------------------------------------------------------- TOTAL 0.00003074 0.07567765 (elements=1,261,568) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1261568 Total Bytes 105724 BPFP 0.6704 bits/point EBPFP 1.3063 equivalent bits/point MSE 0.075678 ---------------------- -------------------------------------------------------- Time: 4.721s Load: 0.008s, Pack+Encode: 2.481s, Decode+Unpack: 2.232s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0757 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample85-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst (89/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,280B, BPFP=0.8716 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,812B, BPFP=0.9307 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,580B, BPFP=0.4626 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,884B, BPFP=0.9482 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,184B, BPFP=0.1403 ⌛️ [2/4] FRONTEND: Frontend time: 2.474s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.237s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014192 layer.1.conv_state 0.00051465 1.47519052 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00011702 0.27241063 layer.3.ssm_state 0.00000001 0.00001554 layer.3.conv_state 0.00007124 0.26099116 layer.4.ssm_state 0.00000002 0.00001766 layer.4.conv_state 0.00024120 0.49121538 layer.4.output 0.00000314 0.00160705 ------------------------------------------------------------------------------------- TOTAL 0.00003014 0.07769527 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 106952 BPFP 0.6963 bits/point EBPFP 1.3458 equivalent bits/point MSE 0.077695 ---------------------- -------------------------------------------------------- Time: 4.719s Load: 0.009s, Pack+Encode: 2.474s, Decode+Unpack: 2.237s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0777 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample86-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst (90/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 101, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,832B, BPFP=0.8442 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,208B, BPFP=0.4399 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,604B, BPFP=1.1965 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,684B, BPFP=1.2014 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,284B, BPFP=0.1409 ⌛️ [2/4] FRONTEND: Frontend time: 2.471s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.235s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 101, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000028 0.00014288 layer.1.conv_state 0.00050888 1.47068787 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013734 0.27249235 layer.3.ssm_state 0.00000001 0.00001550 layer.3.conv_state 0.00006850 0.25986999 layer.4.ssm_state 0.00000001 0.00001758 layer.4.conv_state 0.00023156 0.49151725 layer.4.output 0.00000306 0.00158046 ------------------------------------------------------------------------------------- TOTAL 0.00003009 0.07729438 (elements=1,232,896) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1232896 Total Bytes 106280 BPFP 0.6896 bits/point EBPFP 1.3320 equivalent bits/point MSE 0.077294 ---------------------- -------------------------------------------------------- Time: 4.714s Load: 0.009s, Pack+Encode: 2.471s, Decode+Unpack: 2.235s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0773 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample87-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst (91/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 15,120B, BPFP=0.9229 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,688B, BPFP=1.2017 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,664B, BPFP=0.4678 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,296B, BPFP=0.1194 ⌛️ [2/4] FRONTEND: Frontend time: 2.505s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.261s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000948 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000010 0.00014197 layer.1.conv_state 0.00050284 1.46992338 layer.2.ssm_state 0.00000001 0.00001410 layer.2.conv_state 0.00012269 0.27150935 layer.3.ssm_state 0.00000001 0.00001565 layer.3.conv_state 0.00006872 0.25934955 layer.4.ssm_state 0.00000003 0.00001777 layer.4.conv_state 0.00023958 0.49553996 layer.4.output 0.00000284 0.00154876 ------------------------------------------------------------------------------------- TOTAL 0.00002950 0.07683012 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 107000 BPFP 0.6897 bits/point EBPFP 1.3389 equivalent bits/point MSE 0.076830 ---------------------- -------------------------------------------------------- Time: 4.775s Load: 0.008s, Pack+Encode: 2.505s, Decode+Unpack: 2.261s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0768 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample88-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst (92/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 105, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,684B, BPFP=0.8352 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,696B, BPFP=1.2021 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,824B, BPFP=0.9336 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,800B, BPFP=0.4761 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,780B, BPFP=0.9229 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,140B, BPFP=0.1514 ⌛️ [2/4] FRONTEND: Frontend time: 2.548s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.270s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 105, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014296 layer.1.conv_state 0.00051499 1.47528672 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00014898 0.27331510 layer.3.ssm_state 0.00000001 0.00001542 layer.3.conv_state 0.00006960 0.26216337 layer.4.ssm_state 0.00000003 0.00001752 layer.4.conv_state 0.00022645 0.49377954 layer.4.output 0.00000285 0.00153837 ------------------------------------------------------------------------------------- TOTAL 0.00003001 0.07654861 (elements=1,249,280) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1249280 Total Bytes 107564 BPFP 0.6888 bits/point EBPFP 1.3255 equivalent bits/point MSE 0.076549 ---------------------- -------------------------------------------------------- Time: 4.824s Load: 0.007s, Pack+Encode: 2.548s, Decode+Unpack: 2.270s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0765 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample9-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst (93/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,108B, BPFP=0.8611 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,800B, BPFP=0.9277 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,696B, BPFP=0.4697 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,616B, BPFP=1.1973 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,784B, BPFP=0.9238 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,892B, BPFP=0.9502 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,340B, BPFP=0.1881 ⌛️ [2/4] FRONTEND: Frontend time: 2.528s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.227s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014296 layer.1.conv_state 0.00051394 1.47719133 layer.2.ssm_state 0.00000001 0.00001402 layer.2.conv_state 0.00013358 0.27303579 layer.3.ssm_state 0.00000001 0.00001540 layer.3.conv_state 0.00006601 0.25966379 layer.4.ssm_state 0.00000002 0.00001780 layer.4.conv_state 0.00022760 0.49166012 layer.4.output 0.00000310 0.00164255 ------------------------------------------------------------------------------------- TOTAL 0.00003032 0.07852250 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 109084 BPFP 0.7174 bits/point EBPFP 1.3733 equivalent bits/point MSE 0.078522 ---------------------- -------------------------------------------------------- Time: 4.764s Load: 0.009s, Pack+Encode: 2.528s, Decode+Unpack: 2.227s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0785 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample91-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst (94/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.009s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 103, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,992B, BPFP=0.8540 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,828B, BPFP=0.9346 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,352B, BPFP=0.5098 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,588B, BPFP=1.1956 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,668B, BPFP=1.2004 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,532B, BPFP=0.1239 ⌛️ [2/4] FRONTEND: Frontend time: 2.530s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.232s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 103, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014221 layer.1.conv_state 0.00050709 1.46845543 layer.2.ssm_state 0.00000001 0.00001407 layer.2.conv_state 0.00014738 0.27112737 layer.3.ssm_state 0.00000001 0.00001537 layer.3.conv_state 0.00006942 0.26028752 layer.4.ssm_state 0.00000002 0.00001770 layer.4.conv_state 0.00022996 0.49342439 layer.4.output 0.00000303 0.00154470 ------------------------------------------------------------------------------------- TOTAL 0.00003009 0.07674878 (elements=1,241,088) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1241088 Total Bytes 106820 BPFP 0.6886 bits/point EBPFP 1.3350 equivalent bits/point MSE 0.076749 ---------------------- -------------------------------------------------------- Time: 4.771s Load: 0.009s, Pack+Encode: 2.530s, Decode+Unpack: 2.232s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0767 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample92-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst (95/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 100, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,420B, BPFP=0.8801 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,700B, BPFP=1.2024 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,832B, BPFP=0.9355 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,788B, BPFP=0.4753 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,600B, BPFP=1.1963 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,780B, BPFP=0.9229 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,680B, BPFP=1.2012 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 6,080B, BPFP=0.1187 ⌛️ [2/4] FRONTEND: Frontend time: 2.485s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.240s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 100, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000947 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000010 0.00014200 layer.1.conv_state 0.00052177 1.46675217 layer.2.ssm_state 0.00000001 0.00001417 layer.2.conv_state 0.00012015 0.27039769 layer.3.ssm_state 0.00000001 0.00001544 layer.3.conv_state 0.00007075 0.25878185 layer.4.ssm_state 0.00000003 0.00001784 layer.4.conv_state 0.00024699 0.49456167 layer.4.output 0.00000299 0.00158025 ------------------------------------------------------------------------------------- TOTAL 0.00003050 0.07743798 (elements=1,228,800) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1228800 Total Bytes 106260 BPFP 0.6918 bits/point EBPFP 1.3440 equivalent bits/point MSE 0.077438 ---------------------- -------------------------------------------------------- Time: 4.733s Load: 0.008s, Pack+Encode: 2.485s, Decode+Unpack: 2.240s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0774 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample93-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst (96/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,096B, BPFP=0.8604 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,808B, BPFP=0.9297 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,984B, BPFP=0.4873 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,676B, BPFP=1.2009 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,900B, BPFP=0.9521 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 7,692B, BPFP=0.1518 ⌛️ [2/4] FRONTEND: Frontend time: 2.493s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.236s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000944 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000012 0.00014343 layer.1.conv_state 0.00050618 1.47761965 layer.2.ssm_state 0.00000001 0.00001400 layer.2.conv_state 0.00013466 0.27205002 layer.3.ssm_state 0.00000001 0.00001544 layer.3.conv_state 0.00006821 0.25875014 layer.4.ssm_state 0.00000002 0.00001754 layer.4.conv_state 0.00020718 0.49469268 layer.4.output 0.00000289 0.00159011 ------------------------------------------------------------------------------------- TOTAL 0.00002940 0.07803269 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 107700 BPFP 0.7035 bits/point EBPFP 1.3568 equivalent bits/point MSE 0.078033 ---------------------- -------------------------------------------------------- Time: 4.736s Load: 0.008s, Pack+Encode: 2.493s, Decode+Unpack: 2.236s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0780 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample94-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst (97/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 14,100B, BPFP=0.8606 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,800B, BPFP=0.9277 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,976B, BPFP=0.4868 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,752B, BPFP=0.9160 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,764B, BPFP=0.9189 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,672B, BPFP=1.2007 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,904B, BPFP=0.9531 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 8,320B, BPFP=0.1641 ⌛️ [2/4] FRONTEND: Frontend time: 2.491s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.231s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000945 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014364 layer.1.conv_state 0.00050742 1.47794366 layer.2.ssm_state 0.00000001 0.00001404 layer.2.conv_state 0.00013289 0.27198002 layer.3.ssm_state 0.00000001 0.00001548 layer.3.conv_state 0.00006852 0.25871253 layer.4.ssm_state 0.00000003 0.00001758 layer.4.conv_state 0.00023682 0.49510682 layer.4.output 0.00000300 0.00159893 ------------------------------------------------------------------------------------- TOTAL 0.00003022 0.07805251 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 108312 BPFP 0.7075 bits/point EBPFP 1.3607 equivalent bits/point MSE 0.078053 ---------------------- -------------------------------------------------------- Time: 4.730s Load: 0.008s, Pack+Encode: 2.491s, Decode+Unpack: 2.231s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0781 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample95-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst (98/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 95, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,836B, BPFP=0.8445 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,708B, BPFP=1.2029 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,816B, BPFP=0.9316 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,308B, BPFP=0.4460 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,748B, BPFP=0.9150 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,584B, BPFP=1.1953 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,760B, BPFP=0.9180 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,660B, BPFP=1.2000 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 9,488B, BPFP=0.1951 ⌛️ [2/4] FRONTEND: Frontend time: 2.470s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.240s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 95, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000941 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014230 layer.1.conv_state 0.00050036 1.46986234 layer.2.ssm_state 0.00000001 0.00001409 layer.2.conv_state 0.00012267 0.27309439 layer.3.ssm_state 0.00000001 0.00001541 layer.3.conv_state 0.00007107 0.25939095 layer.4.ssm_state 0.00000002 0.00001762 layer.4.conv_state 0.00023296 0.49160898 layer.4.output 0.00000331 0.00170900 ------------------------------------------------------------------------------------- TOTAL 0.00003019 0.07885908 (elements=1,208,320) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1208320 Total Bytes 108544 BPFP 0.7186 bits/point EBPFP 1.3745 equivalent bits/point MSE 0.078859 ---------------------- -------------------------------------------------------- Time: 4.717s Load: 0.007s, Pack+Encode: 2.470s, Decode+Unpack: 2.240s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0789 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample96-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst (99/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.007s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 99, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 12,736B, BPFP=0.7773 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,684B, BPFP=1.2014 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,820B, BPFP=0.9326 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 8,224B, BPFP=0.5020 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,596B, BPFP=1.1960 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,772B, BPFP=0.9209 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,652B, BPFP=1.1995 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,912B, BPFP=0.9551 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,680B, BPFP=0.2699 ⌛️ [2/4] FRONTEND: Frontend time: 2.483s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.230s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 99, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000010 0.00014241 layer.1.conv_state 0.00050553 1.47499955 layer.2.ssm_state 0.00000001 0.00001406 layer.2.conv_state 0.00013994 0.27301818 layer.3.ssm_state 0.00000001 0.00001546 layer.3.conv_state 0.00007011 0.25898814 layer.4.ssm_state 0.00000002 0.00001764 layer.4.conv_state 0.00023247 0.49119931 layer.4.output 0.00000297 0.00161262 ------------------------------------------------------------------------------------- TOTAL 0.00003027 0.07790875 (elements=1,224,704) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1224704 Total Bytes 112560 BPFP 0.7353 bits/point EBPFP 1.3812 equivalent bits/point MSE 0.077909 ---------------------- -------------------------------------------------------- Time: 4.720s Load: 0.007s, Pack+Encode: 2.483s, Decode+Unpack: 2.230s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0779 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample97-layer4-item1.zst 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst (100/100) [1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst... Original data structure: root: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu ⌛️ [1/4] FRONTEND: Load time: 0.008s ------------------------------------------------------------ FalconMamba Features Summary ------------------------------------------------------------ Number of layers: 5 SSM state shape: (1, 8192, 16) Conv state shape: (1, 8192, 4) Output shape: (1, 97, 4096) Data type: torch.float32 ------------------------------------------------------------ [2/4] FRONTEND: Pack + Encode (strategy: individual)... IndividualPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state layer.0.ssm_state: 13,460B, BPFP=0.8215 Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state layer.0.conv_state: 3,728B, BPFP=0.9102 Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state layer.1.ssm_state: 19,704B, BPFP=1.2026 Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state layer.1.conv_state: 3,804B, BPFP=0.9287 Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state layer.2.ssm_state: 7,640B, BPFP=0.4663 Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state layer.2.conv_state: 3,768B, BPFP=0.9199 Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state layer.3.ssm_state: 19,592B, BPFP=1.1958 Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state layer.3.conv_state: 3,756B, BPFP=0.9170 Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state layer.4.ssm_state: 19,664B, BPFP=1.2002 Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state layer.4.conv_state: 3,908B, BPFP=0.9541 Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output layer.4.output: 13,756B, BPFP=0.2770 ⌛️ [2/4] FRONTEND: Frontend time: 2.476s (Pack+Encode) [3/4] BACKEND: Decode + Unpack... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) ⌛️ [3/4] BACKEND: Backend time: 2.230s [4/4] METRICS: Computing MSE Breakdown... Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output IndividualUnPacker: layer.0.ssm_state: torch.Size([1, 8192, 16]) layer.0.conv_state: torch.Size([1, 8192, 4]) layer.1.ssm_state: torch.Size([1, 8192, 16]) layer.1.conv_state: torch.Size([1, 8192, 4]) layer.2.ssm_state: torch.Size([1, 8192, 16]) layer.2.conv_state: torch.Size([1, 8192, 4]) layer.3.ssm_state: torch.Size([1, 8192, 16]) layer.3.conv_state: torch.Size([1, 8192, 4]) layer.4.ssm_state: torch.Size([1, 8192, 16]) layer.4.conv_state: torch.Size([1, 8192, 4]) layer.4.output: torch.Size([1, 97, 4096]) Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): Key Quant-MSE Total-MSE ------------------------------------------------------------------------------------- layer.0.ssm_state 0.00000001 0.00000942 layer.0.conv_state 0.00014612 0.39288223 layer.1.ssm_state 0.00000013 0.00014366 layer.1.conv_state 0.00050423 1.47800756 layer.2.ssm_state 0.00000001 0.00001400 layer.2.conv_state 0.00013920 0.27400956 layer.3.ssm_state 0.00000001 0.00001550 layer.3.conv_state 0.00007129 0.26114136 layer.4.ssm_state 0.00000002 0.00001764 layer.4.conv_state 0.00024943 0.50482774 layer.4.output 0.00000302 0.00173721 ------------------------------------------------------------------------------------- TOTAL 0.00003091 0.07899618 (elements=1,216,512) ---------------------- -------------------------------------------------------- SAMPLE-WISE STATISTICS ---------------------- -------------------------------------------------------- Handler falconmamba Strategy individual Architecture elic-featurecoding ---------------------- -------------------------------------------------------- Total Elements 1216512 Total Bytes 112780 BPFP 0.7417 bits/point EBPFP 1.3929 equivalent bits/point MSE 0.078996 ---------------------- -------------------------------------------------------- Time: 4.713s Load: 0.008s, Pack+Encode: 2.476s, Decode+Unpack: 2.230s ---------------------- -------------------------------------------------------- Restored Feature Format: [dict] with 3 keys key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu key['ssm_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu key['conv_state']: [list] with 5 items item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu 💾 Converting with 0.0790 MSE: from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst to output-fixed/falconmamba/lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample99-layer4-item1.zst ------------------------ ---------------------------- TOTAL PROCESSING SUMMARY ------------------------ ---------------------------- Total files 100 Avg BPFP 0.7000 bits/point Avg EBPFP 1.3462 equivalent bits/point Avg MSE 0.077244 Avg Time 4.728s ------------------------ ----------------------------