| { |
| "group_size": 64, |
| "mode": "affine", |
| "device": { |
| "device_name": "Apple M4 Max", |
| "max_recommended_working_set_size": 115448725504, |
| "memory_size": 137438953472, |
| "architecture": "applegpu_g16s", |
| "max_buffer_length": 86586540032, |
| "resource_limit": 499000 |
| }, |
| "results": [ |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09608591347932816, |
| "synthetic_matmul_relative_rmse": 0.09557931870222092, |
| "matmul_ms": 0.5911250016652048 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02344648726284504, |
| "synthetic_matmul_relative_rmse": 0.023693682625889778, |
| "matmul_ms": 0.4109000088647008 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007525453343987465, |
| "synthetic_matmul_relative_rmse": 0.008602947928011417, |
| "matmul_ms": 0.4843916976824403 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09441296756267548, |
| "synthetic_matmul_relative_rmse": 0.09449099004268646, |
| "matmul_ms": 0.4845582996495068 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02303435653448105, |
| "synthetic_matmul_relative_rmse": 0.02340426668524742, |
| "matmul_ms": 0.46681660460308194 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.0.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007468077354133129, |
| "synthetic_matmul_relative_rmse": 0.008637943305075169, |
| "matmul_ms": 0.36163330078125 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09535665810108185, |
| "synthetic_matmul_relative_rmse": 0.09527087211608887, |
| "matmul_ms": 0.57477499358356 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02322564087808132, |
| "synthetic_matmul_relative_rmse": 0.023668956011533737, |
| "matmul_ms": 0.4584625014103949 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.0074961441569030285, |
| "synthetic_matmul_relative_rmse": 0.008681540377438068, |
| "matmul_ms": 0.533016596455127 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09448622912168503, |
| "synthetic_matmul_relative_rmse": 0.09489456564188004, |
| "matmul_ms": 0.9572708979249 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02305353432893753, |
| "synthetic_matmul_relative_rmse": 0.02346830628812313, |
| "matmul_ms": 0.645641703158617 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.15.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007503869011998177, |
| "synthetic_matmul_relative_rmse": 0.00874588917940855, |
| "matmul_ms": 0.6554292049258947 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09847480803728104, |
| "synthetic_matmul_relative_rmse": 0.0985177680850029, |
| "matmul_ms": 0.5378333968110383 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.0240206029266119, |
| "synthetic_matmul_relative_rmse": 0.024259869009256363, |
| "matmul_ms": 0.5431666970252991 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.attn.to_q.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007802714128047228, |
| "synthetic_matmul_relative_rmse": 0.008978405967354774, |
| "matmul_ms": 0.5484999972395599 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09683173894882202, |
| "synthetic_matmul_relative_rmse": 0.09696819633245468, |
| "matmul_ms": 0.9709125035442412 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.023639416322112083, |
| "synthetic_matmul_relative_rmse": 0.0238750372081995, |
| "matmul_ms": 0.3942042007111013 |
| }, |
| { |
| "component": "transformer", |
| "tensor": "transformer_blocks.31.img_mlp.out.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007727540098130703, |
| "synthetic_matmul_relative_rmse": 0.008899149484932423, |
| "matmul_ms": 0.4157708026468754 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09275667369365692, |
| "synthetic_matmul_relative_rmse": 0.09330281615257263, |
| "matmul_ms": 0.4116749973036349 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.022620145231485367, |
| "synthetic_matmul_relative_rmse": 0.022967973724007607, |
| "matmul_ms": 0.3953375038690865 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.0073314751498401165, |
| "synthetic_matmul_relative_rmse": 0.008502925746142864, |
| "matmul_ms": 0.40935829747468233 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09217474609613419, |
| "synthetic_matmul_relative_rmse": 0.09356217831373215, |
| "matmul_ms": 0.28917910531163216 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.022505931556224823, |
| "synthetic_matmul_relative_rmse": 0.02307399921119213, |
| "matmul_ms": 0.2340042032301426 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.0.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007364622782915831, |
| "synthetic_matmul_relative_rmse": 0.008642347529530525, |
| "matmul_ms": 0.23023750400170684 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.0940268263220787, |
| "synthetic_matmul_relative_rmse": 0.09416663646697998, |
| "matmul_ms": 0.7600749959237874 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02295832335948944, |
| "synthetic_matmul_relative_rmse": 0.02339865453541279, |
| "matmul_ms": 0.3463750006631017 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007490983698517084, |
| "synthetic_matmul_relative_rmse": 0.00866878591477871, |
| "matmul_ms": 0.40897090220823884 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09371878206729889, |
| "synthetic_matmul_relative_rmse": 0.09475100785493851, |
| "matmul_ms": 0.22580409422516823 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.022860271856188774, |
| "synthetic_matmul_relative_rmse": 0.02319853939116001, |
| "matmul_ms": 0.2775333006866276 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.17.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007405295968055725, |
| "synthetic_matmul_relative_rmse": 0.008685958571732044, |
| "matmul_ms": 0.21041249856352806 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.09651479870080948, |
| "synthetic_matmul_relative_rmse": 0.09677258133888245, |
| "matmul_ms": 0.8071957970969379 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.02353922463953495, |
| "synthetic_matmul_relative_rmse": 0.02398456446826458, |
| "matmul_ms": 0.5642374977469444 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.mlp.down_proj.weight", |
| "source_shape": [ |
| 4096, |
| 12288 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.0075887893326580524, |
| "synthetic_matmul_relative_rmse": 0.008758915588259697, |
| "matmul_ms": 0.5837332922965288 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 4, |
| "weight_relative_rmse": 0.0928456261754036, |
| "synthetic_matmul_relative_rmse": 0.09324980527162552, |
| "matmul_ms": 0.344270805362612 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 6, |
| "weight_relative_rmse": 0.022607337683439255, |
| "synthetic_matmul_relative_rmse": 0.02301245927810669, |
| "matmul_ms": 0.24173329584300518 |
| }, |
| { |
| "component": "text_encoder", |
| "tensor": "model.language_model.layers.35.self_attn.q_proj.weight", |
| "source_shape": [ |
| 4096, |
| 4096 |
| ], |
| "sample_rows": 256, |
| "bits": 8, |
| "weight_relative_rmse": 0.007241594605147839, |
| "synthetic_matmul_relative_rmse": 0.008508691564202309, |
| "matmul_ms": 0.20297080045565963 |
| } |
| ], |
| "limitations": "Sampled rows and synthetic inputs; not calibrated activations, image quality or end-to-end speed." |
| } |
|
|
