| { |
| "algorithm": "activation_calibrated_block_gptq_mxfp8_cuda", |
| "scope": "exact 32x32 Hessian and GPTQ error propagation within each MXFP8 group", |
| "source_model": "/home/ubuntu/artifacts/LFM2.5-Encoder-230M-Code-BF16", |
| "source_revision": null, |
| "corpus_sha256": "ee9edaf80a6854c18053b96521090a51bdb76642abeb98618d7aed36e70b6de9", |
| "calibration": { |
| "samples": 256, |
| "sequence_length": 512, |
| "tokens": 131072, |
| "tokens_sha256": "b70b7dbcca44bd4febb6db72971ad810eb47f3bf841e468dc22b38a6ba341526", |
| "damp": 0.01, |
| "scale_offsets": [ |
| 0, |
| -1, |
| -2, |
| -3, |
| -4, |
| 1, |
| 2 |
| ] |
| }, |
| "weighted_error_rtn": 294328111.73957825, |
| "weighted_error_calibrated": 23714384.461992264, |
| "elapsed_seconds": 11.880840373000012, |
| "cuda": { |
| "device": "NVIDIA RTX A6000", |
| "torch": "2.13.0+cu130", |
| "peak_memory_bytes": 561071104 |
| }, |
| "layers": [ |
| { |
| "path": "layers.0.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 1, |
| "layer_count": 82, |
| "elapsed_seconds": 0.48996823300001324, |
| "weighted_error_rtn": 179342.94677734375, |
| "weighted_error_gptq_base_scale": 152730.77734375, |
| "weighted_error_calibrated": 17813.387084960938, |
| "improvement_vs_rtn": 0.9006741697677331 |
| }, |
| { |
| "path": "layers.0.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 2, |
| "layer_count": 82, |
| "elapsed_seconds": 0.10845255200001702, |
| "weighted_error_rtn": 2049.188217163086, |
| "weighted_error_gptq_base_scale": 1946.7259368896484, |
| "weighted_error_calibrated": 426.3872947692871, |
| "improvement_vs_rtn": 0.7919238012408731 |
| }, |
| { |
| "path": "layers.0.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 3, |
| "layer_count": 82, |
| "elapsed_seconds": 0.26959894100002657, |
| "weighted_error_rtn": 45438.20104980469, |
| "weighted_error_gptq_base_scale": 36424.11474609375, |
| "weighted_error_calibrated": 3932.596405029297, |
| "improvement_vs_rtn": 0.9134517583405473 |
| }, |
| { |
| "path": "layers.0.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 4, |
| "layer_count": 82, |
| "elapsed_seconds": 0.27062287600000445, |
| "weighted_error_rtn": 29514.763916015625, |
| "weighted_error_gptq_base_scale": 27104.729248046875, |
| "weighted_error_calibrated": 3485.0870971679688, |
| "improvement_vs_rtn": 0.8819205497599507 |
| }, |
| { |
| "path": "layers.0.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 5, |
| "layer_count": 82, |
| "elapsed_seconds": 0.1147948999999926, |
| "weighted_error_rtn": 25550.91748046875, |
| "weighted_error_gptq_base_scale": 24562.5712890625, |
| "weighted_error_calibrated": 1835.4121398925781, |
| "improvement_vs_rtn": 0.9281664879041789 |
| }, |
| { |
| "path": "layers.1.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 6, |
| "layer_count": 82, |
| "elapsed_seconds": 0.241891972000019, |
| "weighted_error_rtn": 177933.29736328125, |
| "weighted_error_gptq_base_scale": 157157.2763671875, |
| "weighted_error_calibrated": 17781.86407470703, |
| "improvement_vs_rtn": 0.9000644379764271 |
| }, |
| { |
| "path": "layers.1.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 7, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08138655700003028, |
| "weighted_error_rtn": 1921.2550048828125, |
| "weighted_error_gptq_base_scale": 1856.0776977539062, |
| "weighted_error_calibrated": 258.04229164123535, |
| "improvement_vs_rtn": 0.8656907641175021 |
| }, |
| { |
| "path": "layers.1.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 8, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2044535949999613, |
| "weighted_error_rtn": 42958.769287109375, |
| "weighted_error_gptq_base_scale": 37517.923583984375, |
| "weighted_error_calibrated": 4333.152923583984, |
| "improvement_vs_rtn": 0.8991322843858977 |
| }, |
| { |
| "path": "layers.1.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 9, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20486433500002477, |
| "weighted_error_rtn": 33531.23254394531, |
| "weighted_error_gptq_base_scale": 30663.897216796875, |
| "weighted_error_calibrated": 3872.6761169433594, |
| "improvement_vs_rtn": 0.8845054051661264 |
| }, |
| { |
| "path": "layers.1.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 10, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08416208999994979, |
| "weighted_error_rtn": 6168.6309814453125, |
| "weighted_error_gptq_base_scale": 5937.9517822265625, |
| "weighted_error_calibrated": 814.3446960449219, |
| "improvement_vs_rtn": 0.8679861547084924 |
| }, |
| { |
| "path": "layers.2.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 11, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08355473500000699, |
| "weighted_error_rtn": 4103.093719482422, |
| "weighted_error_gptq_base_scale": 3328.605499267578, |
| "weighted_error_calibrated": 350.45429611206055, |
| "improvement_vs_rtn": 0.9145877915369021 |
| }, |
| { |
| "path": "layers.2.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 12, |
| "layer_count": 82, |
| "elapsed_seconds": 0.04143708899999865, |
| "weighted_error_rtn": 2893.2667846679688, |
| "weighted_error_gptq_base_scale": 2331.9813232421875, |
| "weighted_error_calibrated": 200.38968658447266, |
| "improvement_vs_rtn": 0.9307392986895021 |
| }, |
| { |
| "path": "layers.2.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 13, |
| "layer_count": 82, |
| "elapsed_seconds": 0.05350548999996363, |
| "weighted_error_rtn": 726.1098022460938, |
| "weighted_error_gptq_base_scale": 656.0045013427734, |
| "weighted_error_calibrated": 94.92555236816406, |
| "improvement_vs_rtn": 0.8692683226771922 |
| }, |
| { |
| "path": "layers.2.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 14, |
| "layer_count": 82, |
| "elapsed_seconds": 0.11616594700001315, |
| "weighted_error_rtn": 8108.913879394531, |
| "weighted_error_gptq_base_scale": 4213.835510253906, |
| "weighted_error_calibrated": 600.181037902832, |
| "improvement_vs_rtn": 0.9259850274858702 |
| }, |
| { |
| "path": "layers.2.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 15, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2501158540000006, |
| "weighted_error_rtn": 75591.42602539062, |
| "weighted_error_gptq_base_scale": 52641.105224609375, |
| "weighted_error_calibrated": 6263.699859619141, |
| "improvement_vs_rtn": 0.9171374296138399 |
| }, |
| { |
| "path": "layers.2.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 16, |
| "layer_count": 82, |
| "elapsed_seconds": 0.23029257299998562, |
| "weighted_error_rtn": 44216.74206542969, |
| "weighted_error_gptq_base_scale": 35437.559326171875, |
| "weighted_error_calibrated": 4459.20361328125, |
| "improvement_vs_rtn": 0.8991512398927368 |
| }, |
| { |
| "path": "layers.2.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 17, |
| "layer_count": 82, |
| "elapsed_seconds": 0.0937899209999955, |
| "weighted_error_rtn": 11485.158081054688, |
| "weighted_error_gptq_base_scale": 10267.645263671875, |
| "weighted_error_calibrated": 1259.6156005859375, |
| "improvement_vs_rtn": 0.8903266640566547 |
| }, |
| { |
| "path": "layers.3.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 18, |
| "layer_count": 82, |
| "elapsed_seconds": 0.259505192000006, |
| "weighted_error_rtn": 284194.7138671875, |
| "weighted_error_gptq_base_scale": 203158.6611328125, |
| "weighted_error_calibrated": 23717.09112548828, |
| "improvement_vs_rtn": 0.9165463326085228 |
| }, |
| { |
| "path": "layers.3.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 19, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08513244100004158, |
| "weighted_error_rtn": 7492.77734375, |
| "weighted_error_gptq_base_scale": 6945.483703613281, |
| "weighted_error_calibrated": 893.3844680786133, |
| "improvement_vs_rtn": 0.8807672472979838 |
| }, |
| { |
| "path": "layers.3.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 20, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21195685700001832, |
| "weighted_error_rtn": 114002.86328125, |
| "weighted_error_gptq_base_scale": 81732.78125, |
| "weighted_error_calibrated": 9485.263854980469, |
| "improvement_vs_rtn": 0.9167980208393546 |
| }, |
| { |
| "path": "layers.3.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 21, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21021360700001424, |
| "weighted_error_rtn": 63289.369384765625, |
| "weighted_error_gptq_base_scale": 51932.40625, |
| "weighted_error_calibrated": 6605.5665283203125, |
| "improvement_vs_rtn": 0.8956291302546879 |
| }, |
| { |
| "path": "layers.3.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 22, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08588793199999145, |
| "weighted_error_rtn": 25236.03369140625, |
| "weighted_error_gptq_base_scale": 23294.947021484375, |
| "weighted_error_calibrated": 2842.044158935547, |
| "improvement_vs_rtn": 0.88738150401569 |
| }, |
| { |
| "path": "layers.4.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 23, |
| "layer_count": 82, |
| "elapsed_seconds": 0.0951457720000235, |
| "weighted_error_rtn": 25046.245849609375, |
| "weighted_error_gptq_base_scale": 17007.4404296875, |
| "weighted_error_calibrated": 1375.3707885742188, |
| "improvement_vs_rtn": 0.9450867488551914 |
| }, |
| { |
| "path": "layers.4.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 24, |
| "layer_count": 82, |
| "elapsed_seconds": 0.052635054000006676, |
| "weighted_error_rtn": 18130.5302734375, |
| "weighted_error_gptq_base_scale": 12619.007080078125, |
| "weighted_error_calibrated": 908.9774627685547, |
| "improvement_vs_rtn": 0.9498648164692529 |
| }, |
| { |
| "path": "layers.4.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 25, |
| "layer_count": 82, |
| "elapsed_seconds": 0.043256893999966906, |
| "weighted_error_rtn": 3642.7069702148438, |
| "weighted_error_gptq_base_scale": 3117.4861450195312, |
| "weighted_error_calibrated": 443.7787094116211, |
| "improvement_vs_rtn": 0.8781733713306488 |
| }, |
| { |
| "path": "layers.4.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 26, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08624938899998824, |
| "weighted_error_rtn": 50970.6474609375, |
| "weighted_error_gptq_base_scale": 26589.30419921875, |
| "weighted_error_calibrated": 3778.4235229492188, |
| "improvement_vs_rtn": 0.925870599822282 |
| }, |
| { |
| "path": "layers.4.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 27, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21452962099999695, |
| "weighted_error_rtn": 126925.16748046875, |
| "weighted_error_gptq_base_scale": 90475.85546875, |
| "weighted_error_calibrated": 11116.107788085938, |
| "improvement_vs_rtn": 0.9124199872354197 |
| }, |
| { |
| "path": "layers.4.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 28, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21508126899999525, |
| "weighted_error_rtn": 72193.33203125, |
| "weighted_error_gptq_base_scale": 59255.047607421875, |
| "weighted_error_calibrated": 7850.59228515625, |
| "improvement_vs_rtn": 0.8912559918725181 |
| }, |
| { |
| "path": "layers.4.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 29, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08801339999996571, |
| "weighted_error_rtn": 25644.060546875, |
| "weighted_error_gptq_base_scale": 23068.67578125, |
| "weighted_error_calibrated": 2873.8348388671875, |
| "improvement_vs_rtn": 0.8879337055996229 |
| }, |
| { |
| "path": "layers.5.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 30, |
| "layer_count": 82, |
| "elapsed_seconds": 0.25715951600000153, |
| "weighted_error_rtn": 416175.9013671875, |
| "weighted_error_gptq_base_scale": 316003.4775390625, |
| "weighted_error_calibrated": 35648.6591796875, |
| "improvement_vs_rtn": 0.9143423272165029 |
| }, |
| { |
| "path": "layers.5.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 31, |
| "layer_count": 82, |
| "elapsed_seconds": 0.09009318300002178, |
| "weighted_error_rtn": 18902.208129882812, |
| "weighted_error_gptq_base_scale": 17356.250366210938, |
| "weighted_error_calibrated": 2259.593994140625, |
| "improvement_vs_rtn": 0.880458728492763 |
| }, |
| { |
| "path": "layers.5.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 32, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2144410559999983, |
| "weighted_error_rtn": 148010.7802734375, |
| "weighted_error_gptq_base_scale": 106211.1689453125, |
| "weighted_error_calibrated": 12685.857543945312, |
| "improvement_vs_rtn": 0.9142909893420651 |
| }, |
| { |
| "path": "layers.5.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 33, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21448284600000989, |
| "weighted_error_rtn": 92784.2265625, |
| "weighted_error_gptq_base_scale": 76535.36669921875, |
| "weighted_error_calibrated": 9873.634582519531, |
| "improvement_vs_rtn": 0.8935849880058159 |
| }, |
| { |
| "path": "layers.5.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 34, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08794359600000234, |
| "weighted_error_rtn": 127423.55688476562, |
| "weighted_error_gptq_base_scale": 123329.5361328125, |
| "weighted_error_calibrated": 4034.936737060547, |
| "improvement_vs_rtn": 0.9683344521554244 |
| }, |
| { |
| "path": "layers.6.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 35, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08651574099997106, |
| "weighted_error_rtn": 49559.43212890625, |
| "weighted_error_gptq_base_scale": 38113.135498046875, |
| "weighted_error_calibrated": 4374.100891113281, |
| "improvement_vs_rtn": 0.9117402943654388 |
| }, |
| { |
| "path": "layers.6.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 36, |
| "layer_count": 82, |
| "elapsed_seconds": 0.04434033500001533, |
| "weighted_error_rtn": 33573.0693359375, |
| "weighted_error_gptq_base_scale": 25960.28369140625, |
| "weighted_error_calibrated": 2564.0931396484375, |
| "improvement_vs_rtn": 0.9236264902088126 |
| }, |
| { |
| "path": "layers.6.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 37, |
| "layer_count": 82, |
| "elapsed_seconds": 0.04377803100004485, |
| "weighted_error_rtn": 9334.384643554688, |
| "weighted_error_gptq_base_scale": 8107.2325439453125, |
| "weighted_error_calibrated": 1098.9798278808594, |
| "improvement_vs_rtn": 0.8822654229660768 |
| }, |
| { |
| "path": "layers.6.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 38, |
| "layer_count": 82, |
| "elapsed_seconds": 0.0874465470000132, |
| "weighted_error_rtn": 115330.2783203125, |
| "weighted_error_gptq_base_scale": 69973.4384765625, |
| "weighted_error_calibrated": 9910.2294921875, |
| "improvement_vs_rtn": 0.9140708785539966 |
| }, |
| { |
| "path": "layers.6.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 39, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21564778899994508, |
| "weighted_error_rtn": 153980.93359375, |
| "weighted_error_gptq_base_scale": 114128.791015625, |
| "weighted_error_calibrated": 14548.955932617188, |
| "improvement_vs_rtn": 0.9055145621405186 |
| }, |
| { |
| "path": "layers.6.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 40, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20937601299999642, |
| "weighted_error_rtn": 114420.4677734375, |
| "weighted_error_gptq_base_scale": 93202.98828125, |
| "weighted_error_calibrated": 12073.403198242188, |
| "improvement_vs_rtn": 0.8944821373904136 |
| }, |
| { |
| "path": "layers.6.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 41, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08570223000003807, |
| "weighted_error_rtn": 56730.39404296875, |
| "weighted_error_gptq_base_scale": 51821.8134765625, |
| "weighted_error_calibrated": 6358.197692871094, |
| "improvement_vs_rtn": 0.8879225536833877 |
| }, |
| { |
| "path": "layers.7.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 42, |
| "layer_count": 82, |
| "elapsed_seconds": 0.24807579299999816, |
| "weighted_error_rtn": 549649.673828125, |
| "weighted_error_gptq_base_scale": 439786.166015625, |
| "weighted_error_calibrated": 52085.45422363281, |
| "improvement_vs_rtn": 0.9052388153696606 |
| }, |
| { |
| "path": "layers.7.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 43, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08272821799999974, |
| "weighted_error_rtn": 37919.469970703125, |
| "weighted_error_gptq_base_scale": 35703.880615234375, |
| "weighted_error_calibrated": 4494.651123046875, |
| "improvement_vs_rtn": 0.8814685140240759 |
| }, |
| { |
| "path": "layers.7.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 44, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20718469299998787, |
| "weighted_error_rtn": 277325.125, |
| "weighted_error_gptq_base_scale": 204097.71484375, |
| "weighted_error_calibrated": 24464.388427734375, |
| "improvement_vs_rtn": 0.9117844500106711 |
| }, |
| { |
| "path": "layers.7.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 45, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20585456700001714, |
| "weighted_error_rtn": 197980.34765625, |
| "weighted_error_gptq_base_scale": 168000.03662109375, |
| "weighted_error_calibrated": 20700.371826171875, |
| "improvement_vs_rtn": 0.8954422897462854 |
| }, |
| { |
| "path": "layers.7.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 46, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08562121599999273, |
| "weighted_error_rtn": 636492.015625, |
| "weighted_error_gptq_base_scale": 607885.734375, |
| "weighted_error_calibrated": 42527.3427734375, |
| "improvement_vs_rtn": 0.9331847977202385 |
| }, |
| { |
| "path": "layers.8.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 47, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08255767800000058, |
| "weighted_error_rtn": 188737.013671875, |
| "weighted_error_gptq_base_scale": 153218.427734375, |
| "weighted_error_calibrated": 16722.84228515625, |
| "improvement_vs_rtn": 0.9113960639738138 |
| }, |
| { |
| "path": "layers.8.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 48, |
| "layer_count": 82, |
| "elapsed_seconds": 0.04179186099997878, |
| "weighted_error_rtn": 178395.83203125, |
| "weighted_error_gptq_base_scale": 146007.4140625, |
| "weighted_error_calibrated": 12102.09521484375, |
| "improvement_vs_rtn": 0.9321615585014128 |
| }, |
| { |
| "path": "layers.8.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 49, |
| "layer_count": 82, |
| "elapsed_seconds": 0.041977677000033964, |
| "weighted_error_rtn": 41494.6484375, |
| "weighted_error_gptq_base_scale": 37166.8251953125, |
| "weighted_error_calibrated": 5127.5679931640625, |
| "improvement_vs_rtn": 0.8764282097512575 |
| }, |
| { |
| "path": "layers.8.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 50, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08648361899997781, |
| "weighted_error_rtn": 319536.54296875, |
| "weighted_error_gptq_base_scale": 215609.54296875, |
| "weighted_error_calibrated": 31307.155517578125, |
| "improvement_vs_rtn": 0.9020232389487924 |
| }, |
| { |
| "path": "layers.8.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 51, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20859776999998303, |
| "weighted_error_rtn": 293820.0244140625, |
| "weighted_error_gptq_base_scale": 227860.4853515625, |
| "weighted_error_calibrated": 28461.096313476562, |
| "improvement_vs_rtn": 0.9031342524382611 |
| }, |
| { |
| "path": "layers.8.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 52, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20707119700000476, |
| "weighted_error_rtn": 248586.99609375, |
| "weighted_error_gptq_base_scale": 207142.9453125, |
| "weighted_error_calibrated": 27051.83837890625, |
| "improvement_vs_rtn": 0.8911775804688346 |
| }, |
| { |
| "path": "layers.8.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 53, |
| "layer_count": 82, |
| "elapsed_seconds": 0.0863417979999781, |
| "weighted_error_rtn": 236962.1171875, |
| "weighted_error_gptq_base_scale": 217055.416015625, |
| "weighted_error_calibrated": 27855.535400390625, |
| "improvement_vs_rtn": 0.8824473053709699 |
| }, |
| { |
| "path": "layers.9.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 54, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2504155689999834, |
| "weighted_error_rtn": 852138.2734375, |
| "weighted_error_gptq_base_scale": 704399.27734375, |
| "weighted_error_calibrated": 87968.05053710938, |
| "improvement_vs_rtn": 0.8967678682213758 |
| }, |
| { |
| "path": "layers.9.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 55, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08314246999998431, |
| "weighted_error_rtn": 158893.34375, |
| "weighted_error_gptq_base_scale": 151830.630859375, |
| "weighted_error_calibrated": 19551.391845703125, |
| "improvement_vs_rtn": 0.8769527320385117 |
| }, |
| { |
| "path": "layers.9.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 56, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20971068800002968, |
| "weighted_error_rtn": 381702.6318359375, |
| "weighted_error_gptq_base_scale": 307972.9072265625, |
| "weighted_error_calibrated": 41283.94470214844, |
| "improvement_vs_rtn": 0.8918426511664898 |
| }, |
| { |
| "path": "layers.9.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 57, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2108086679999701, |
| "weighted_error_rtn": 350850.1015625, |
| "weighted_error_gptq_base_scale": 302119.310546875, |
| "weighted_error_calibrated": 39981.46862792969, |
| "improvement_vs_rtn": 0.8860440158065411 |
| }, |
| { |
| "path": "layers.9.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 58, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08557464299997264, |
| "weighted_error_rtn": 70525482.68359375, |
| "weighted_error_gptq_base_scale": 69916479.68359375, |
| "weighted_error_calibrated": 82143.8291015625, |
| "improvement_vs_rtn": 0.9988352603062627 |
| }, |
| { |
| "path": "layers.10.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 59, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08308798000001616, |
| "weighted_error_rtn": 615073.17578125, |
| "weighted_error_gptq_base_scale": 488613.55859375, |
| "weighted_error_calibrated": 55698.18408203125, |
| "improvement_vs_rtn": 0.9094446217536883 |
| }, |
| { |
| "path": "layers.10.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 60, |
| "layer_count": 82, |
| "elapsed_seconds": 0.041676313000039045, |
| "weighted_error_rtn": 352273.375, |
| "weighted_error_gptq_base_scale": 297185.3515625, |
| "weighted_error_calibrated": 31452.2939453125, |
| "improvement_vs_rtn": 0.9107162329673297 |
| }, |
| { |
| "path": "layers.10.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 61, |
| "layer_count": 82, |
| "elapsed_seconds": 0.041405505000000176, |
| "weighted_error_rtn": 211297.01171875, |
| "weighted_error_gptq_base_scale": 193459.70703125, |
| "weighted_error_calibrated": 20308.2861328125, |
| "improvement_vs_rtn": 0.9038874900898071 |
| }, |
| { |
| "path": "layers.10.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 62, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08455199000002267, |
| "weighted_error_rtn": 1261526.890625, |
| "weighted_error_gptq_base_scale": 939701.265625, |
| "weighted_error_calibrated": 138940.46875, |
| "improvement_vs_rtn": 0.8898632523947512 |
| }, |
| { |
| "path": "layers.10.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 63, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2061380930000496, |
| "weighted_error_rtn": 544789.158203125, |
| "weighted_error_gptq_base_scale": 443061.16796875, |
| "weighted_error_calibrated": 61692.224853515625, |
| "improvement_vs_rtn": 0.8867594482661976 |
| }, |
| { |
| "path": "layers.10.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 64, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20676381999999194, |
| "weighted_error_rtn": 674400.87109375, |
| "weighted_error_gptq_base_scale": 575180.0859375, |
| "weighted_error_calibrated": 74212.90502929688, |
| "improvement_vs_rtn": 0.8899572817737651 |
| }, |
| { |
| "path": "layers.10.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 65, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08747031199999356, |
| "weighted_error_rtn": 1743346.09375, |
| "weighted_error_gptq_base_scale": 1583750.765625, |
| "weighted_error_calibrated": 198069.6953125, |
| "improvement_vs_rtn": 0.8863853276050053 |
| }, |
| { |
| "path": "layers.11.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 66, |
| "layer_count": 82, |
| "elapsed_seconds": 0.25331018600002153, |
| "weighted_error_rtn": 1242598.046875, |
| "weighted_error_gptq_base_scale": 1065857.421875, |
| "weighted_error_calibrated": 142212.6416015625, |
| "improvement_vs_rtn": 0.8855521767805672 |
| }, |
| { |
| "path": "layers.11.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 67, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08594013200001882, |
| "weighted_error_rtn": 1484520.796875, |
| "weighted_error_gptq_base_scale": 1416829.47265625, |
| "weighted_error_calibrated": 95212.8876953125, |
| "improvement_vs_rtn": 0.9358628805364391 |
| }, |
| { |
| "path": "layers.11.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 68, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21086014799999475, |
| "weighted_error_rtn": 529678.21875, |
| "weighted_error_gptq_base_scale": 453391.23046875, |
| "weighted_error_calibrated": 64637.472412109375, |
| "improvement_vs_rtn": 0.8779684153057137 |
| }, |
| { |
| "path": "layers.11.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 69, |
| "layer_count": 82, |
| "elapsed_seconds": 0.21230689199995822, |
| "weighted_error_rtn": 789014.59375, |
| "weighted_error_gptq_base_scale": 685786.77734375, |
| "weighted_error_calibrated": 88777.3935546875, |
| "improvement_vs_rtn": 0.8874832046733768 |
| }, |
| { |
| "path": "layers.11.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 70, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08750873299999284, |
| "weighted_error_rtn": 1818762.28125, |
| "weighted_error_gptq_base_scale": 1717772.203125, |
| "weighted_error_calibrated": 486465.73828125, |
| "improvement_vs_rtn": 0.7325292352407311 |
| }, |
| { |
| "path": "layers.12.self_attn.q_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 71, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08506403199999113, |
| "weighted_error_rtn": 2717019.140625, |
| "weighted_error_gptq_base_scale": 2243394.125, |
| "weighted_error_calibrated": 245132.185546875, |
| "improvement_vs_rtn": 0.909778999388835 |
| }, |
| { |
| "path": "layers.12.self_attn.k_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 72, |
| "layer_count": 82, |
| "elapsed_seconds": 0.042338916999995035, |
| "weighted_error_rtn": 1960639.15625, |
| "weighted_error_gptq_base_scale": 1581443.84375, |
| "weighted_error_calibrated": 137055.4609375, |
| "improvement_vs_rtn": 0.9300965399468825 |
| }, |
| { |
| "path": "layers.12.self_attn.v_proj", |
| "shape": [ |
| 512, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 73, |
| "layer_count": 82, |
| "elapsed_seconds": 0.04232235599999967, |
| "weighted_error_rtn": 1804754.4375, |
| "weighted_error_gptq_base_scale": 1683808.34375, |
| "weighted_error_calibrated": 166360.37109375, |
| "improvement_vs_rtn": 0.9078210488712263 |
| }, |
| { |
| "path": "layers.12.self_attn.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 74, |
| "layer_count": 82, |
| "elapsed_seconds": 0.0835323680000215, |
| "weighted_error_rtn": 8453758.0, |
| "weighted_error_gptq_base_scale": 6436060.0, |
| "weighted_error_calibrated": 920533.2734375, |
| "improvement_vs_rtn": 0.8911095783156437 |
| }, |
| { |
| "path": "layers.12.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 75, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2090240500000391, |
| "weighted_error_rtn": 660609.6484375, |
| "weighted_error_gptq_base_scale": 539012.5703125, |
| "weighted_error_calibrated": 78259.03173828125, |
| "improvement_vs_rtn": 0.8815351366372219 |
| }, |
| { |
| "path": "layers.12.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 76, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20943459300002587, |
| "weighted_error_rtn": 1045918.828125, |
| "weighted_error_gptq_base_scale": 873864.1953125, |
| "weighted_error_calibrated": 116525.4677734375, |
| "improvement_vs_rtn": 0.888590333551667 |
| }, |
| { |
| "path": "layers.12.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 77, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08651274799996145, |
| "weighted_error_rtn": 12907661.625, |
| "weighted_error_gptq_base_scale": 10184180.8125, |
| "weighted_error_calibrated": 1007256.0703125, |
| "improvement_vs_rtn": 0.921964481284386 |
| }, |
| { |
| "path": "layers.13.conv.in_proj", |
| "shape": [ |
| 3072, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 78, |
| "layer_count": 82, |
| "elapsed_seconds": 0.24930786300001273, |
| "weighted_error_rtn": 1415559.75, |
| "weighted_error_gptq_base_scale": 1175251.4296875, |
| "weighted_error_calibrated": 146300.390625, |
| "improvement_vs_rtn": 0.8966483819386641 |
| }, |
| { |
| "path": "layers.13.conv.out_proj", |
| "shape": [ |
| 1024, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 79, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08337048300001015, |
| "weighted_error_rtn": 5583583.671875, |
| "weighted_error_gptq_base_scale": 4498203.296875, |
| "weighted_error_calibrated": 2135965.73046875, |
| "improvement_vs_rtn": 0.6174561256728728 |
| }, |
| { |
| "path": "layers.13.feed_forward.w1", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 80, |
| "layer_count": 82, |
| "elapsed_seconds": 0.2074719589999745, |
| "weighted_error_rtn": 675282.33984375, |
| "weighted_error_gptq_base_scale": 563844.666015625, |
| "weighted_error_calibrated": 79871.166015625, |
| "improvement_vs_rtn": 0.8817218201884178 |
| }, |
| { |
| "path": "layers.13.feed_forward.w3", |
| "shape": [ |
| 2560, |
| 1024 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 81, |
| "layer_count": 82, |
| "elapsed_seconds": 0.20756134200001952, |
| "weighted_error_rtn": 1037956.85546875, |
| "weighted_error_gptq_base_scale": 874257.96875, |
| "weighted_error_calibrated": 115698.4833984375, |
| "improvement_vs_rtn": 0.8885324733982444 |
| }, |
| { |
| "path": "layers.13.feed_forward.w2", |
| "shape": [ |
| 1024, |
| 2560 |
| ], |
| "calibration_tokens": 131072, |
| "layer_index": 82, |
| "layer_count": 82, |
| "elapsed_seconds": 0.08574635999997327, |
| "weighted_error_rtn": 166477592.9375, |
| "weighted_error_gptq_base_scale": 153208565.5, |
| "weighted_error_calibrated": 16288815.1171875, |
| "improvement_vs_rtn": 0.9021561110431076 |
| } |
| ], |
| "improvement_vs_rtn": 0.9194287479988498 |
| } |
|
|
