diff --git "a/evaluation/export.json" "b/evaluation/export.json" --- "a/evaluation/export.json" +++ "b/evaluation/export.json" @@ -1,6646 +1,164 @@ { - "run_dir": "art_3c5dbfc942580099ea75e8095e26ca25", - "source": "model", - "adapter_sha256": "60723ac470780ac7e8bfe7785c23665be2d5c5d6653cbd2fad653523214aea7d", - "source_sha256": "207483664dfef54c4c4438863260d3ba423bf3d6009b43fe84264b281e9fda23", - "checkpoint_sha256": { - "nvfp4-base": "602a1bc6a69c1a5c41253df785f335cea77eeeb854b985ab11c125bf1d08a547", - "nvfp4-merged": "9bc834f4dce29a48499b6614c11772d189710138b29c637a384f753b1326b4c0" + "source_sha256": { + "model-00001-of-00008.safetensors": "f8ae89948adc35f42b2d735abc460bf20df89fe27063009e7023af236421943e", + "model-00002-of-00008.safetensors": "09c16bdfe0f5fb160048a35f08ca998fa1e704efdcb4b3d504e46b2870c68e2d", + "model-00003-of-00008.safetensors": "978f4d0f66dce924a81ee2e57aaaa4d0431bf5a163eb3445f32414637846b74c", + "model-00004-of-00008.safetensors": "056ce5b367b2d955ce08ec5795d12ba6475640df7175c9f3eb7d5d35e18ddd86", + "model-00005-of-00008.safetensors": "dbc45d99c137991456115cd9b71e8fe69dcb7d5b923cee93835e34646af9da0d", + "model-00006-of-00008.safetensors": "bd559f8be05597b9915e91c8bd449d7c5194a2205c5a02e4ec750971a3f2814e", + "model-00007-of-00008.safetensors": "6c1ebd7492594617d7e0be46cce8b05d57f27930dac431bff1c23a3d496782a3", + "model-00008-of-00008.safetensors": "09dde4e45f4cc217f00222f3c57d8b8656d12a8f215e09d300a6f98d434fe80e" }, - "loops": 2, - "recipe": "mse", - "encoding": "max-abs global F32 scale; FP8 E4M3 block-16 scales; E2M1 nearest-even", - "block_scale_search": "all 126 positive finite E4M3 scales; minimum block squared error; maxabs wins ties", - "merge": "W + (alpha/rank) * BF16(B) @ BF16(A), accumulated F32, stored BF16", - "recipe_reference": "https://github.com/NVIDIA/Model-Optimizer/blob/72e48d515a12d79784b37ca1ac8c92cc34e85d67/modelopt/torch/quantization/qtensor/nvfp4_tensor.py", - "calibration_reference": "https://github.com/NVIDIA/Model-Optimizer/blob/72e48d515a12d79784b37ca1ac8c92cc34e85d67/modelopt/torch/quantization/calib/mse.py", - "script_sha256": "cbd6c833654b1fff319abf1d8f3cb491af5d96b78d541e593a283b684268da04", - "quantizer_sha256": "baf5ad4237ff4390c41696a32ba97cb3b82ec87fe0bb78ea8878909f0a1b3772", - "projections": [ - { - "name": "backbone.layers.0.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.56793212890625, - "squared_error": 27.054529190063477, - "changed_scale_blocks": 434964 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.64353942871094, - "squared_error": 27.120437622070312, - "changed_scale_blocks": 434924 - }, - "packed_byte_changes": 2094882, - "merged_bf16_changes": 10015065 - }, - { - "name": "backbone.layers.0.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.13469123840332, - "squared_error": 6.788881301879883, - "changed_scale_blocks": 107460 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.156957626342773, - "squared_error": 6.804623603820801, - "changed_scale_blocks": 107604 - }, - "packed_byte_changes": 549867, - "merged_bf16_changes": 2509077 - }, - { - "name": "backbone.layers.0.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.74720001220703, - "squared_error": 27.124553680419922, - "changed_scale_blocks": 438985 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.82270050048828, - "squared_error": 27.18092155456543, - "changed_scale_blocks": 438684 - }, - "packed_byte_changes": 2089604, - "merged_bf16_changes": 9982035 - }, - { - "name": "backbone.layers.0.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.586708068847656, - "squared_error": 27.041221618652344, - "changed_scale_blocks": 434725 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.679832458496094, - "squared_error": 27.106388092041016, - "changed_scale_blocks": 434825 - }, - "packed_byte_changes": 2120252, - "merged_bf16_changes": 10019313 - }, - { - "name": "backbone.layers.0.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.05076789855957, - "squared_error": 6.779272079467773, - "changed_scale_blocks": 106556 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.098959922790527, - "squared_error": 6.807531356811523, - "changed_scale_blocks": 106779 - }, - "packed_byte_changes": 561601, - "merged_bf16_changes": 2513056 - }, - { - "name": "backbone.layers.1.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.436214447021484, - "squared_error": 47.113182067871094, - "changed_scale_blocks": 753641 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.554100036621094, - "squared_error": 47.209930419921875, - "changed_scale_blocks": 753360 - }, - "packed_byte_changes": 3580658, - "merged_bf16_changes": 17481279 - }, - { - "name": "backbone.layers.1.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.39999389648438, - "squared_error": 47.31950378417969, - "changed_scale_blocks": 772854 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.61836242675781, - "squared_error": 47.44442367553711, - "changed_scale_blocks": 774693 - }, - "packed_byte_changes": 3924034, - "merged_bf16_changes": 17580372 - }, - { - "name": "backbone.layers.1.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.57077026367188, - "squared_error": 47.375186920166016, - "changed_scale_blocks": 770240 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.77439880371094, - "squared_error": 47.533302307128906, - "changed_scale_blocks": 769530 - }, - "packed_byte_changes": 4075007, - "merged_bf16_changes": 17595887 - }, - { - "name": "backbone.layers.10.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.840030670166016, - "squared_error": 27.052135467529297, - "changed_scale_blocks": 440732 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.97000503540039, - "squared_error": 27.144256591796875, - "changed_scale_blocks": 441152 - }, - "packed_byte_changes": 2338699, - "merged_bf16_changes": 10051882 - }, - { - "name": "backbone.layers.10.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.112932205200195, - "squared_error": 6.774394989013672, - "changed_scale_blocks": 108726 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.129868507385254, - "squared_error": 6.796084403991699, - "changed_scale_blocks": 108349 - }, - "packed_byte_changes": 566850, - "merged_bf16_changes": 2515352 - }, - { - "name": "backbone.layers.10.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.7869758605957, - "squared_error": 27.118873596191406, - "changed_scale_blocks": 437579 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.87635803222656, - "squared_error": 27.17495346069336, - "changed_scale_blocks": 437797 - }, - "packed_byte_changes": 2078707, - "merged_bf16_changes": 9985202 - }, - { - "name": "backbone.layers.10.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.634334564208984, - "squared_error": 27.07560157775879, - "changed_scale_blocks": 434490 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.77946853637695, - "squared_error": 27.177722930908203, - "changed_scale_blocks": 434807 - }, - "packed_byte_changes": 2332249, - "merged_bf16_changes": 10029721 - }, - { - "name": "backbone.layers.10.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.0637788772583, - "squared_error": 6.7910332679748535, - "changed_scale_blocks": 105881 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.096620559692383, - "squared_error": 6.8119401931762695, - "changed_scale_blocks": 105944 - }, - "packed_byte_changes": 542970, - "merged_bf16_changes": 2504080 - }, - { - "name": "backbone.layers.11.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.79780197143555, - "squared_error": 47.40611267089844, - "changed_scale_blocks": 753663 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.980857849121094, - "squared_error": 47.504966735839844, - "changed_scale_blocks": 755308 - }, - "packed_byte_changes": 3643222, - "merged_bf16_changes": 17490815 - }, - { - "name": "backbone.layers.11.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.2222671508789, - "squared_error": 47.36455154418945, - "changed_scale_blocks": 760207 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.36769104003906, - "squared_error": 47.48512268066406, - "changed_scale_blocks": 760547 - }, - "packed_byte_changes": 3884606, - "merged_bf16_changes": 17592729 - }, - { - "name": "backbone.layers.11.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.15992736816406, - "squared_error": 47.389957427978516, - "changed_scale_blocks": 760890 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.3367919921875, - "squared_error": 47.511634826660156, - "changed_scale_blocks": 760027 - }, - "packed_byte_changes": 3979195, - "merged_bf16_changes": 17623140 - }, - { - "name": "backbone.layers.12.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.739097595214844, - "squared_error": 27.045639038085938, - "changed_scale_blocks": 435098 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.885379791259766, - "squared_error": 27.141284942626953, - "changed_scale_blocks": 436176 - }, - "packed_byte_changes": 2348175, - "merged_bf16_changes": 10071142 - }, - { - "name": "backbone.layers.12.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.016687393188477, - "squared_error": 6.804931640625, - "changed_scale_blocks": 105958 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.055780410766602, - "squared_error": 6.8314313888549805, - "changed_scale_blocks": 105990 - }, - "packed_byte_changes": 566993, - "merged_bf16_changes": 2515191 - }, - { - "name": "backbone.layers.12.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.753135681152344, - "squared_error": 27.086475372314453, - "changed_scale_blocks": 439569 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.835044860839844, - "squared_error": 27.14173126220703, - "changed_scale_blocks": 439010 - }, - "packed_byte_changes": 2047995, - "merged_bf16_changes": 9987205 - }, - { - "name": "backbone.layers.12.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.62255859375, - "squared_error": 27.07930564880371, - "changed_scale_blocks": 432833 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.800968170166016, - "squared_error": 27.202857971191406, - "changed_scale_blocks": 433374 - }, - "packed_byte_changes": 2513294, - "merged_bf16_changes": 10093456 - }, - { - "name": "backbone.layers.12.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.050793647766113, - "squared_error": 6.8075995445251465, - "changed_scale_blocks": 105363 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.083625793457031, - "squared_error": 6.82252311706543, - "changed_scale_blocks": 105357 - }, - "packed_byte_changes": 534158, - "merged_bf16_changes": 2504889 - }, - { - "name": "backbone.layers.13.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.931339263916016, - "squared_error": 47.470767974853516, - "changed_scale_blocks": 754131 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.07542419433594, - "squared_error": 47.57240295410156, - "changed_scale_blocks": 753202 - }, - "packed_byte_changes": 3640000, - "merged_bf16_changes": 17481670 - }, - { - "name": "backbone.layers.13.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.21002960205078, - "squared_error": 47.406700134277344, - "changed_scale_blocks": 761706 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.37590026855469, - "squared_error": 47.584693908691406, - "changed_scale_blocks": 760881 - }, - "packed_byte_changes": 4130722, - "merged_bf16_changes": 17624393 - }, - { - "name": "backbone.layers.13.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.13106536865234, - "squared_error": 47.38280487060547, - "changed_scale_blocks": 763254 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.27314758300781, - "squared_error": 47.49352264404297, - "changed_scale_blocks": 763153 - }, - "packed_byte_changes": 3881613, - "merged_bf16_changes": 17596402 - }, - { - "name": "backbone.layers.14.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.6453742980957, - "squared_error": 27.116931915283203, - "changed_scale_blocks": 435785 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.92763900756836, - "squared_error": 27.319808959960938, - "changed_scale_blocks": 436224 - }, - "packed_byte_changes": 2671207, - "merged_bf16_changes": 10069655 - }, - { - "name": "backbone.layers.14.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.970661163330078, - "squared_error": 6.859411239624023, - "changed_scale_blocks": 103740 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.994955062866211, - "squared_error": 6.8796844482421875, - "changed_scale_blocks": 103933 - }, - "packed_byte_changes": 574002, - "merged_bf16_changes": 2520298 - }, - { - "name": "backbone.layers.14.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.50196075439453, - "squared_error": 27.13785171508789, - "changed_scale_blocks": 431675 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.536155700683594, - "squared_error": 27.17727279663086, - "changed_scale_blocks": 430965 - }, - "packed_byte_changes": 2001567, - "merged_bf16_changes": 9961474 - }, - { - "name": "backbone.layers.14.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.428714752197266, - "squared_error": 27.165325164794922, - "changed_scale_blocks": 428523 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.51548385620117, - "squared_error": 27.235729217529297, - "changed_scale_blocks": 428914 - }, - "packed_byte_changes": 2187677, - "merged_bf16_changes": 10042504 - }, - { - "name": "backbone.layers.14.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.993873596191406, - "squared_error": 6.8322954177856445, - "changed_scale_blocks": 103254 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.020578384399414, - "squared_error": 6.854691028594971, - "changed_scale_blocks": 103568 - }, - "packed_byte_changes": 527149, - "merged_bf16_changes": 2503899 - }, - { - "name": "backbone.layers.15.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.5236701965332, - "squared_error": 47.500614166259766, - "changed_scale_blocks": 747790 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.68345260620117, - "squared_error": 47.60614013671875, - "changed_scale_blocks": 748511 - }, - "packed_byte_changes": 3641626, - "merged_bf16_changes": 17461437 - }, - { - "name": "backbone.layers.15.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.90575408935547, - "squared_error": 47.40859603881836, - "changed_scale_blocks": 756530 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.16525268554688, - "squared_error": 47.59835433959961, - "changed_scale_blocks": 756420 - }, - "packed_byte_changes": 4189025, - "merged_bf16_changes": 17601186 - }, - { - "name": "backbone.layers.15.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.02972412109375, - "squared_error": 47.45419692993164, - "changed_scale_blocks": 755482 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.19737243652344, - "squared_error": 47.58234405517578, - "changed_scale_blocks": 756159 - }, - "packed_byte_changes": 3876033, - "merged_bf16_changes": 17604847 - }, - { - "name": "backbone.layers.16.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.41865539550781, - "squared_error": 27.162437438964844, - "changed_scale_blocks": 428650 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.66520309448242, - "squared_error": 27.315093994140625, - "changed_scale_blocks": 429291 - }, - "packed_byte_changes": 2454100, - "merged_bf16_changes": 10031280 - }, - { - "name": "backbone.layers.16.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.910487174987793, - "squared_error": 6.868203163146973, - "changed_scale_blocks": 102777 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.946792602539062, - "squared_error": 6.894364833831787, - "changed_scale_blocks": 102444 - }, - "packed_byte_changes": 594280, - "merged_bf16_changes": 2512190 - }, - { - "name": "backbone.layers.16.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.59331512451172, - "squared_error": 27.13104248046875, - "changed_scale_blocks": 431983 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.64288330078125, - "squared_error": 27.17746925354004, - "changed_scale_blocks": 431746 - }, - "packed_byte_changes": 1972409, - "merged_bf16_changes": 9940547 - }, - { - "name": "backbone.layers.16.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.08900451660156, - "squared_error": 27.276281356811523, - "changed_scale_blocks": 422251 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.21628952026367, - "squared_error": 27.372459411621094, - "changed_scale_blocks": 422818 - }, - "packed_byte_changes": 2394383, - "merged_bf16_changes": 10077587 - }, - { - "name": "backbone.layers.16.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.941841125488281, - "squared_error": 6.855693817138672, - "changed_scale_blocks": 101828 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.98011589050293, - "squared_error": 6.882214546203613, - "changed_scale_blocks": 101884 - }, - "packed_byte_changes": 523328, - "merged_bf16_changes": 2485311 - }, - { - "name": "backbone.layers.17.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.10968017578125, - "squared_error": 47.479793548583984, - "changed_scale_blocks": 756188 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.21755981445312, - "squared_error": 47.56600570678711, - "changed_scale_blocks": 755301 - }, - "packed_byte_changes": 3484525, - "merged_bf16_changes": 17422888 - }, - { - "name": "backbone.layers.17.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.13819122314453, - "squared_error": 47.41585922241211, - "changed_scale_blocks": 754445 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.34420776367188, - "squared_error": 47.572505950927734, - "changed_scale_blocks": 754618 - }, - "packed_byte_changes": 4112982, - "merged_bf16_changes": 17647221 - }, - { - "name": "backbone.layers.17.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.0208511352539, - "squared_error": 47.451873779296875, - "changed_scale_blocks": 752394 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.19557189941406, - "squared_error": 47.571468353271484, - "changed_scale_blocks": 753032 - }, - "packed_byte_changes": 3936300, - "merged_bf16_changes": 17611186 - }, - { - "name": "backbone.layers.18.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.612159729003906, - "squared_error": 27.066471099853516, - "changed_scale_blocks": 435469 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.77796173095703, - "squared_error": 27.184499740600586, - "changed_scale_blocks": 435309 - }, - "packed_byte_changes": 2445289, - "merged_bf16_changes": 10085436 - }, - { - "name": "backbone.layers.18.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.957234382629395, - "squared_error": 6.822916030883789, - "changed_scale_blocks": 103953 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.993890762329102, - "squared_error": 6.84952974319458, - "changed_scale_blocks": 104147 - }, - "packed_byte_changes": 604674, - "merged_bf16_changes": 2529096 - }, - { - "name": "backbone.layers.18.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.64729309082031, - "squared_error": 27.127403259277344, - "changed_scale_blocks": 432484 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.70751190185547, - "squared_error": 27.171558380126953, - "changed_scale_blocks": 432944 - }, - "packed_byte_changes": 2011529, - "merged_bf16_changes": 9957768 - }, - { - "name": "backbone.layers.18.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.14331817626953, - "squared_error": 27.210220336914062, - "changed_scale_blocks": 424311 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.30110168457031, - "squared_error": 27.325790405273438, - "changed_scale_blocks": 424792 - }, - "packed_byte_changes": 2411048, - "merged_bf16_changes": 10058525 - }, - { - "name": "backbone.layers.18.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.056896209716797, - "squared_error": 6.8292341232299805, - "changed_scale_blocks": 104704 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.084324836730957, - "squared_error": 6.8510332107543945, - "changed_scale_blocks": 104823 - }, - "packed_byte_changes": 537427, - "merged_bf16_changes": 2505607 - }, - { - "name": "backbone.layers.19.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.14530944824219, - "squared_error": 47.42831802368164, - "changed_scale_blocks": 758897 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.27040100097656, - "squared_error": 47.53883361816406, - "changed_scale_blocks": 759462 - }, - "packed_byte_changes": 3676034, - "merged_bf16_changes": 17485820 - }, - { - "name": "backbone.layers.19.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.01229858398438, - "squared_error": 47.4512939453125, - "changed_scale_blocks": 753055 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.27359008789062, - "squared_error": 47.61795425415039, - "changed_scale_blocks": 753049 - }, - "packed_byte_changes": 4207648, - "merged_bf16_changes": 17661652 - }, - { - "name": "backbone.layers.19.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.892913818359375, - "squared_error": 47.39109802246094, - "changed_scale_blocks": 756513 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.07403564453125, - "squared_error": 47.52857208251953, - "changed_scale_blocks": 756072 - }, - "packed_byte_changes": 3949469, - "merged_bf16_changes": 17610179 - }, - { - "name": "backbone.layers.2.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.77448272705078, - "squared_error": 27.030765533447266, - "changed_scale_blocks": 440284 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.905067443847656, - "squared_error": 27.118804931640625, - "changed_scale_blocks": 440886 - }, - "packed_byte_changes": 2331379, - "merged_bf16_changes": 10064228 - }, - { - "name": "backbone.layers.2.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.082761764526367, - "squared_error": 6.803312301635742, - "changed_scale_blocks": 106848 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.104716300964355, - "squared_error": 6.822260856628418, - "changed_scale_blocks": 106556 - }, - "packed_byte_changes": 594200, - "merged_bf16_changes": 2522908 - }, - { - "name": "backbone.layers.2.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.78772735595703, - "squared_error": 27.128934860229492, - "changed_scale_blocks": 440783 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.90254211425781, - "squared_error": 27.225204467773438, - "changed_scale_blocks": 440731 - }, - "packed_byte_changes": 2227355, - "merged_bf16_changes": 9997579 - }, - { - "name": "backbone.layers.2.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.55322265625, - "squared_error": 27.139564514160156, - "changed_scale_blocks": 433517 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.659088134765625, - "squared_error": 27.22863006591797, - "changed_scale_blocks": 432724 - }, - "packed_byte_changes": 2338701, - "merged_bf16_changes": 10075187 - }, - { - "name": "backbone.layers.2.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.123863220214844, - "squared_error": 6.799846649169922, - "changed_scale_blocks": 107745 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.162609100341797, - "squared_error": 6.829841136932373, - "changed_scale_blocks": 107777 - }, - "packed_byte_changes": 608338, - "merged_bf16_changes": 2515758 - }, - { - "name": "backbone.layers.20.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.47840118408203, - "squared_error": 27.09878921508789, - "changed_scale_blocks": 429512 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.6718864440918, - "squared_error": 27.24873924255371, - "changed_scale_blocks": 429636 - }, - "packed_byte_changes": 2457809, - "merged_bf16_changes": 10057703 - }, - { - "name": "backbone.layers.20.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.954449653625488, - "squared_error": 6.814853668212891, - "changed_scale_blocks": 103548 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.005058288574219, - "squared_error": 6.847256660461426, - "changed_scale_blocks": 103793 - }, - "packed_byte_changes": 608290, - "merged_bf16_changes": 2525083 - }, - { - "name": "backbone.layers.20.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.625831604003906, - "squared_error": 27.164897918701172, - "changed_scale_blocks": 430745 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.71808624267578, - "squared_error": 27.241649627685547, - "changed_scale_blocks": 431035 - }, - "packed_byte_changes": 2012607, - "merged_bf16_changes": 9910800 - }, - { - "name": "backbone.layers.20.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.12784957885742, - "squared_error": 27.204071044921875, - "changed_scale_blocks": 422835 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.26237487792969, - "squared_error": 27.314462661743164, - "changed_scale_blocks": 422951 - }, - "packed_byte_changes": 2395966, - "merged_bf16_changes": 10083280 - }, - { - "name": "backbone.layers.20.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.01357364654541, - "squared_error": 6.856293201446533, - "changed_scale_blocks": 103042 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.027425765991211, - "squared_error": 6.87813138961792, - "changed_scale_blocks": 103176 - }, - "packed_byte_changes": 554708, - "merged_bf16_changes": 2510377 - }, - { - "name": "backbone.layers.21.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.991355895996094, - "squared_error": 47.456565856933594, - "changed_scale_blocks": 755678 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.13379669189453, - "squared_error": 47.5694694519043, - "changed_scale_blocks": 755669 - }, - "packed_byte_changes": 3540329, - "merged_bf16_changes": 17437717 - }, - { - "name": "backbone.layers.21.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.78936004638672, - "squared_error": 47.48406982421875, - "changed_scale_blocks": 748908 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.03736114501953, - "squared_error": 47.664390563964844, - "changed_scale_blocks": 749809 - }, - "packed_byte_changes": 4102434, - "merged_bf16_changes": 17613365 - }, - { - "name": "backbone.layers.21.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.84050750732422, - "squared_error": 47.484683990478516, - "changed_scale_blocks": 745875 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.99045944213867, - "squared_error": 47.60504913330078, - "changed_scale_blocks": 745942 - }, - "packed_byte_changes": 3790296, - "merged_bf16_changes": 17578009 - }, - { - "name": "backbone.layers.22.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.60335922241211, - "squared_error": 27.121070861816406, - "changed_scale_blocks": 429419 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.78105926513672, - "squared_error": 27.24913787841797, - "changed_scale_blocks": 430059 - }, - "packed_byte_changes": 2522500, - "merged_bf16_changes": 10098643 - }, - { - "name": "backbone.layers.22.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.961618423461914, - "squared_error": 6.96592378616333, - "changed_scale_blocks": 101137 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.991783142089844, - "squared_error": 6.9862823486328125, - "changed_scale_blocks": 101266 - }, - "packed_byte_changes": 587340, - "merged_bf16_changes": 2525386 - }, - { - "name": "backbone.layers.22.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.67131423950195, - "squared_error": 27.19951057434082, - "changed_scale_blocks": 428904 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.71564483642578, - "squared_error": 27.24661636352539, - "changed_scale_blocks": 429431 - }, - "packed_byte_changes": 2020277, - "merged_bf16_changes": 9931977 - }, - { - "name": "backbone.layers.22.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.18889617919922, - "squared_error": 27.244613647460938, - "changed_scale_blocks": 422655 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.37016677856445, - "squared_error": 27.369813919067383, - "changed_scale_blocks": 423214 - }, - "packed_byte_changes": 2425671, - "merged_bf16_changes": 10074807 - }, - { - "name": "backbone.layers.22.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.027023315429688, - "squared_error": 6.850836753845215, - "changed_scale_blocks": 102786 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.053642272949219, - "squared_error": 6.876121520996094, - "changed_scale_blocks": 102791 - }, - "packed_byte_changes": 536057, - "merged_bf16_changes": 2486269 - }, - { - "name": "backbone.layers.23.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.04527282714844, - "squared_error": 47.428924560546875, - "changed_scale_blocks": 759079 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.15764617919922, - "squared_error": 47.50032424926758, - "changed_scale_blocks": 758779 - }, - "packed_byte_changes": 3314281, - "merged_bf16_changes": 17356289 - }, - { - "name": "backbone.layers.23.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.962127685546875, - "squared_error": 47.38285827636719, - "changed_scale_blocks": 757741 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.12702941894531, - "squared_error": 47.49626922607422, - "changed_scale_blocks": 757318 - }, - "packed_byte_changes": 3945552, - "merged_bf16_changes": 17620940 - }, - { - "name": "backbone.layers.23.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.10152435302734, - "squared_error": 47.38231658935547, - "changed_scale_blocks": 759941 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.22178649902344, - "squared_error": 47.51422119140625, - "changed_scale_blocks": 758787 - }, - "packed_byte_changes": 3959336, - "merged_bf16_changes": 17598337 - }, - { - "name": "backbone.layers.24.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.71173858642578, - "squared_error": 27.079927444458008, - "changed_scale_blocks": 436479 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.86252212524414, - "squared_error": 27.19440460205078, - "changed_scale_blocks": 436607 - }, - "packed_byte_changes": 2417346, - "merged_bf16_changes": 10077423 - }, - { - "name": "backbone.layers.24.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.003425598144531, - "squared_error": 6.809095859527588, - "changed_scale_blocks": 104732 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.023696899414062, - "squared_error": 6.826573371887207, - "changed_scale_blocks": 104645 - }, - "packed_byte_changes": 563470, - "merged_bf16_changes": 2513776 - }, - { - "name": "backbone.layers.24.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.8631591796875, - "squared_error": 27.085926055908203, - "changed_scale_blocks": 438623 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.911380767822266, - "squared_error": 27.13541603088379, - "changed_scale_blocks": 437779 - }, - "packed_byte_changes": 1941285, - "merged_bf16_changes": 9915708 - }, - { - "name": "backbone.layers.24.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.19871520996094, - "squared_error": 27.170562744140625, - "changed_scale_blocks": 426898 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.320030212402344, - "squared_error": 27.259170532226562, - "changed_scale_blocks": 426509 - }, - "packed_byte_changes": 2303969, - "merged_bf16_changes": 10046858 - }, - { - "name": "backbone.layers.24.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.069049835205078, - "squared_error": 6.801266670227051, - "changed_scale_blocks": 105872 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.090087890625, - "squared_error": 6.821489334106445, - "changed_scale_blocks": 106006 - }, - "packed_byte_changes": 542020, - "merged_bf16_changes": 2491835 - }, - { - "name": "backbone.layers.25.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.995872497558594, - "squared_error": 47.41883087158203, - "changed_scale_blocks": 759760 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.08768463134766, - "squared_error": 47.496490478515625, - "changed_scale_blocks": 759913 - }, - "packed_byte_changes": 3291463, - "merged_bf16_changes": 17366617 - }, - { - "name": "backbone.layers.25.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.99088668823242, - "squared_error": 47.379180908203125, - "changed_scale_blocks": 755542 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.2110824584961, - "squared_error": 47.53857421875, - "changed_scale_blocks": 755981 - }, - "packed_byte_changes": 4140314, - "merged_bf16_changes": 17621941 - }, - { - "name": "backbone.layers.25.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.12435913085938, - "squared_error": 47.393653869628906, - "changed_scale_blocks": 761647 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.3144760131836, - "squared_error": 47.540443420410156, - "changed_scale_blocks": 762056 - }, - "packed_byte_changes": 3946935, - "merged_bf16_changes": 17573477 - }, - { - "name": "backbone.layers.26.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.70680236816406, - "squared_error": 27.05986785888672, - "changed_scale_blocks": 435214 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.87453079223633, - "squared_error": 27.191089630126953, - "changed_scale_blocks": 434734 - }, - "packed_byte_changes": 2416969, - "merged_bf16_changes": 10062406 - }, - { - "name": "backbone.layers.26.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.064212799072266, - "squared_error": 6.787937164306641, - "changed_scale_blocks": 106485 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.096070289611816, - "squared_error": 6.816518783569336, - "changed_scale_blocks": 106373 - }, - "packed_byte_changes": 593066, - "merged_bf16_changes": 2521115 - }, - { - "name": "backbone.layers.26.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.77317428588867, - "squared_error": 27.033918380737305, - "changed_scale_blocks": 439935 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.79025650024414, - "squared_error": 27.08612632751465, - "changed_scale_blocks": 438673 - }, - "packed_byte_changes": 1840791, - "merged_bf16_changes": 9897911 - }, - { - "name": "backbone.layers.26.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.44761276245117, - "squared_error": 27.14809799194336, - "changed_scale_blocks": 431974 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.56886291503906, - "squared_error": 27.24574851989746, - "changed_scale_blocks": 431317 - }, - "packed_byte_changes": 2313103, - "merged_bf16_changes": 10055157 - }, - { - "name": "backbone.layers.26.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.025218963623047, - "squared_error": 6.813191890716553, - "changed_scale_blocks": 103952 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.03858470916748, - "squared_error": 6.823733329772949, - "changed_scale_blocks": 104014 - }, - "packed_byte_changes": 494170, - "merged_bf16_changes": 2494107 - }, - { - "name": "backbone.layers.27.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.96791076660156, - "squared_error": 47.416725158691406, - "changed_scale_blocks": 757445 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.11166381835938, - "squared_error": 47.518653869628906, - "changed_scale_blocks": 757622 - }, - "packed_byte_changes": 3442071, - "merged_bf16_changes": 17378538 - }, - { - "name": "backbone.layers.27.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.91527557373047, - "squared_error": 47.4300422668457, - "changed_scale_blocks": 755265 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.08857727050781, - "squared_error": 47.572486877441406, - "changed_scale_blocks": 754954 - }, - "packed_byte_changes": 4015404, - "merged_bf16_changes": 17603868 - }, - { - "name": "backbone.layers.27.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.86422348022461, - "squared_error": 47.41204071044922, - "changed_scale_blocks": 754380 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.04815673828125, - "squared_error": 47.539306640625, - "changed_scale_blocks": 755054 - }, - "packed_byte_changes": 3846109, - "merged_bf16_changes": 17576469 - }, - { - "name": "backbone.layers.28.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.563053131103516, - "squared_error": 27.081125259399414, - "changed_scale_blocks": 434744 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.74103546142578, - "squared_error": 27.196313858032227, - "changed_scale_blocks": 434275 - }, - "packed_byte_changes": 2433147, - "merged_bf16_changes": 10056915 - }, - { - "name": "backbone.layers.28.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.033262252807617, - "squared_error": 6.794902801513672, - "changed_scale_blocks": 105323 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.065530776977539, - "squared_error": 6.821264266967773, - "changed_scale_blocks": 105651 - }, - "packed_byte_changes": 570311, - "merged_bf16_changes": 2516693 - }, - { - "name": "backbone.layers.28.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.65633010864258, - "squared_error": 27.181106567382812, - "changed_scale_blocks": 431247 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.67401123046875, - "squared_error": 27.21933364868164, - "changed_scale_blocks": 430069 - }, - "packed_byte_changes": 1826576, - "merged_bf16_changes": 9880112 - }, - { - "name": "backbone.layers.28.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.483890533447266, - "squared_error": 27.10614013671875, - "changed_scale_blocks": 430950 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.61797332763672, - "squared_error": 27.217729568481445, - "changed_scale_blocks": 430991 - }, - "packed_byte_changes": 2408744, - "merged_bf16_changes": 10080943 - }, - { - "name": "backbone.layers.28.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.949706077575684, - "squared_error": 6.853264808654785, - "changed_scale_blocks": 101269 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.971803665161133, - "squared_error": 6.8700971603393555, - "changed_scale_blocks": 101399 - }, - "packed_byte_changes": 535711, - "merged_bf16_changes": 2503835 - }, - { - "name": "backbone.layers.29.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.74223709106445, - "squared_error": 47.54600143432617, - "changed_scale_blocks": 746459 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.83576583862305, - "squared_error": 47.60001754760742, - "changed_scale_blocks": 746737 - }, - "packed_byte_changes": 3183315, - "merged_bf16_changes": 17318801 - }, - { - "name": "backbone.layers.29.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.63849639892578, - "squared_error": 47.56550979614258, - "changed_scale_blocks": 740817 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.82256317138672, - "squared_error": 47.68706512451172, - "changed_scale_blocks": 741892 - }, - "packed_byte_changes": 3899179, - "merged_bf16_changes": 17579259 - }, - { - "name": "backbone.layers.29.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.643402099609375, - "squared_error": 47.510406494140625, - "changed_scale_blocks": 742078 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.82667541503906, - "squared_error": 47.64133071899414, - "changed_scale_blocks": 742512 - }, - "packed_byte_changes": 3813544, - "merged_bf16_changes": 17558461 - }, - { - "name": "backbone.layers.3.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.83732604980469, - "squared_error": 47.448753356933594, - "changed_scale_blocks": 755795 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.00119018554688, - "squared_error": 47.552642822265625, - "changed_scale_blocks": 755921 - }, - "packed_byte_changes": 3643842, - "merged_bf16_changes": 17506658 - }, - { - "name": "backbone.layers.3.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.35057067871094, - "squared_error": 47.28550338745117, - "changed_scale_blocks": 770157 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.5731201171875, - "squared_error": 47.44181442260742, - "changed_scale_blocks": 771042 - }, - "packed_byte_changes": 3971242, - "merged_bf16_changes": 17597405 - }, - { - "name": "backbone.layers.3.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.47018432617188, - "squared_error": 47.33487319946289, - "changed_scale_blocks": 769758 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.58030700683594, - "squared_error": 47.48486328125, - "changed_scale_blocks": 766895 - }, - "packed_byte_changes": 3964844, - "merged_bf16_changes": 17574709 - }, - { - "name": "backbone.layers.30.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.51323699951172, - "squared_error": 27.127275466918945, - "changed_scale_blocks": 427591 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.64283752441406, - "squared_error": 27.223670959472656, - "changed_scale_blocks": 427267 - }, - "packed_byte_changes": 2339495, - "merged_bf16_changes": 10037062 - }, - { - "name": "backbone.layers.30.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.59294319152832, - "squared_error": 6.813179016113281, - "changed_scale_blocks": 98540 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.613348007202148, - "squared_error": 6.8335771560668945, - "changed_scale_blocks": 98670 - }, - "packed_byte_changes": 569569, - "merged_bf16_changes": 2519679 - }, - { - "name": "backbone.layers.30.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.56007385253906, - "squared_error": 27.13707733154297, - "changed_scale_blocks": 430136 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.60805892944336, - "squared_error": 27.16585922241211, - "changed_scale_blocks": 430130 - }, - "packed_byte_changes": 1927079, - "merged_bf16_changes": 9913211 - }, - { - "name": "backbone.layers.30.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 35.97594451904297, - "squared_error": 27.300689697265625, - "changed_scale_blocks": 416647 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.07814025878906, - "squared_error": 27.370389938354492, - "changed_scale_blocks": 417117 - }, - "packed_byte_changes": 2173117, - "merged_bf16_changes": 10053468 - }, - { - "name": "backbone.layers.30.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.902606964111328, - "squared_error": 6.913654327392578, - "changed_scale_blocks": 98364 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.917821884155273, - "squared_error": 6.927826404571533, - "changed_scale_blocks": 98499 - }, - "packed_byte_changes": 503814, - "merged_bf16_changes": 2503255 - }, - { - "name": "backbone.layers.31.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.02537536621094, - "squared_error": 47.42814254760742, - "changed_scale_blocks": 756817 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.11172485351562, - "squared_error": 47.50090789794922, - "changed_scale_blocks": 756308 - }, - "packed_byte_changes": 3234298, - "merged_bf16_changes": 17319788 - }, - { - "name": "backbone.layers.31.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.75492858886719, - "squared_error": 47.4133186340332, - "changed_scale_blocks": 750102 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.90700912475586, - "squared_error": 47.52669906616211, - "changed_scale_blocks": 749370 - }, - "packed_byte_changes": 3846456, - "merged_bf16_changes": 17588540 - }, - { - "name": "backbone.layers.31.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.82606887817383, - "squared_error": 47.45951461791992, - "changed_scale_blocks": 749004 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.00785827636719, - "squared_error": 47.591407775878906, - "changed_scale_blocks": 748667 - }, - "packed_byte_changes": 3728580, - "merged_bf16_changes": 17498820 - }, - { - "name": "backbone.layers.32.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.56270217895508, - "squared_error": 27.070819854736328, - "changed_scale_blocks": 433692 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.71448516845703, - "squared_error": 27.173919677734375, - "changed_scale_blocks": 433840 - }, - "packed_byte_changes": 2316348, - "merged_bf16_changes": 10037074 - }, - { - "name": "backbone.layers.32.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.058204650878906, - "squared_error": 6.785831928253174, - "changed_scale_blocks": 105635 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.079750061035156, - "squared_error": 6.811295509338379, - "changed_scale_blocks": 105780 - }, - "packed_byte_changes": 571421, - "merged_bf16_changes": 2514487 - }, - { - "name": "backbone.layers.32.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.712059020996094, - "squared_error": 27.090225219726562, - "changed_scale_blocks": 437592 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.76802062988281, - "squared_error": 27.13926887512207, - "changed_scale_blocks": 437534 - }, - "packed_byte_changes": 1880066, - "merged_bf16_changes": 9894211 - }, - { - "name": "backbone.layers.32.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.32563781738281, - "squared_error": 27.165586471557617, - "changed_scale_blocks": 427966 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.41974639892578, - "squared_error": 27.226837158203125, - "changed_scale_blocks": 428257 - }, - "packed_byte_changes": 2042933, - "merged_bf16_changes": 9996383 - }, - { - "name": "backbone.layers.32.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.015520095825195, - "squared_error": 6.827552795410156, - "changed_scale_blocks": 103070 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.033891677856445, - "squared_error": 6.841629981994629, - "changed_scale_blocks": 103068 - }, - "packed_byte_changes": 478933, - "merged_bf16_changes": 2483435 - }, - { - "name": "backbone.layers.33.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.10975646972656, - "squared_error": 47.44975280761719, - "changed_scale_blocks": 755705 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.1598892211914, - "squared_error": 47.48780822753906, - "changed_scale_blocks": 755629 - }, - "packed_byte_changes": 2858804, - "merged_bf16_changes": 17185884 - }, - { - "name": "backbone.layers.33.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.85130310058594, - "squared_error": 47.42411804199219, - "changed_scale_blocks": 751215 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.04441833496094, - "squared_error": 47.54405212402344, - "changed_scale_blocks": 751241 - }, - "packed_byte_changes": 3696798, - "merged_bf16_changes": 17517962 - }, - { - "name": "backbone.layers.33.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.859832763671875, - "squared_error": 47.40791702270508, - "changed_scale_blocks": 753835 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.99737548828125, - "squared_error": 47.50971221923828, - "changed_scale_blocks": 753832 - }, - "packed_byte_changes": 3639756, - "merged_bf16_changes": 17530337 - }, - { - "name": "backbone.layers.34.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.5661735534668, - "squared_error": 27.06768035888672, - "changed_scale_blocks": 432058 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.7936897277832, - "squared_error": 27.23736000061035, - "changed_scale_blocks": 432721 - }, - "packed_byte_changes": 2576773, - "merged_bf16_changes": 10049735 - }, - { - "name": "backbone.layers.34.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.05767822265625, - "squared_error": 6.800314903259277, - "changed_scale_blocks": 105992 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.086889266967773, - "squared_error": 6.8227105140686035, - "changed_scale_blocks": 106000 - }, - "packed_byte_changes": 578479, - "merged_bf16_changes": 2518212 - }, - { - "name": "backbone.layers.34.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.793052673339844, - "squared_error": 27.11785888671875, - "changed_scale_blocks": 437543 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.85498809814453, - "squared_error": 27.15141487121582, - "changed_scale_blocks": 437780 - }, - "packed_byte_changes": 1701647, - "merged_bf16_changes": 9844595 - }, - { - "name": "backbone.layers.34.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.35059356689453, - "squared_error": 27.141254425048828, - "changed_scale_blocks": 428426 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.482601165771484, - "squared_error": 27.238216400146484, - "changed_scale_blocks": 428356 - }, - "packed_byte_changes": 2244082, - "merged_bf16_changes": 10032459 - }, - { - "name": "backbone.layers.34.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.978755950927734, - "squared_error": 6.843965530395508, - "changed_scale_blocks": 102008 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.007018089294434, - "squared_error": 6.859130859375, - "changed_scale_blocks": 101947 - }, - "packed_byte_changes": 487224, - "merged_bf16_changes": 2486747 - }, - { - "name": "backbone.layers.35.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.95600128173828, - "squared_error": 47.55970001220703, - "changed_scale_blocks": 756276 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.02386474609375, - "squared_error": 47.60801696777344, - "changed_scale_blocks": 756221 - }, - "packed_byte_changes": 2939597, - "merged_bf16_changes": 17187898 - }, - { - "name": "backbone.layers.35.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.83902359008789, - "squared_error": 47.50906753540039, - "changed_scale_blocks": 751543 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.999961853027344, - "squared_error": 47.619083404541016, - "changed_scale_blocks": 751205 - }, - "packed_byte_changes": 3797855, - "merged_bf16_changes": 17560626 - }, - { - "name": "backbone.layers.35.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.84233474731445, - "squared_error": 47.481380462646484, - "changed_scale_blocks": 752694 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.01268005371094, - "squared_error": 47.59815216064453, - "changed_scale_blocks": 752885 - }, - "packed_byte_changes": 3761396, - "merged_bf16_changes": 17553432 - }, - { - "name": "backbone.layers.36.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.61490249633789, - "squared_error": 27.072446823120117, - "changed_scale_blocks": 431674 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.73583984375, - "squared_error": 27.183006286621094, - "changed_scale_blocks": 431256 - }, - "packed_byte_changes": 2375187, - "merged_bf16_changes": 10049624 - }, - { - "name": "backbone.layers.36.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.06563663482666, - "squared_error": 6.810665607452393, - "changed_scale_blocks": 105575 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.090685844421387, - "squared_error": 6.832886695861816, - "changed_scale_blocks": 105333 - }, - "packed_byte_changes": 609919, - "merged_bf16_changes": 2524290 - }, - { - "name": "backbone.layers.36.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.61090087890625, - "squared_error": 27.135452270507812, - "changed_scale_blocks": 432993 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.68874740600586, - "squared_error": 27.174692153930664, - "changed_scale_blocks": 432824 - }, - "packed_byte_changes": 1837291, - "merged_bf16_changes": 9869804 - }, - { - "name": "backbone.layers.36.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.21887969970703, - "squared_error": 27.166831970214844, - "changed_scale_blocks": 424102 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.33198547363281, - "squared_error": 27.262371063232422, - "changed_scale_blocks": 423764 - }, - "packed_byte_changes": 2320416, - "merged_bf16_changes": 10032986 - }, - { - "name": "backbone.layers.36.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.918578147888184, - "squared_error": 6.890514373779297, - "changed_scale_blocks": 98647 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.939403533935547, - "squared_error": 6.902276039123535, - "changed_scale_blocks": 98753 - }, - "packed_byte_changes": 477623, - "merged_bf16_changes": 2489801 - }, - { - "name": "backbone.layers.37.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.2445068359375, - "squared_error": 47.46710968017578, - "changed_scale_blocks": 763559 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.29397583007812, - "squared_error": 47.51866149902344, - "changed_scale_blocks": 763695 - }, - "packed_byte_changes": 2758863, - "merged_bf16_changes": 17081121 - }, - { - "name": "backbone.layers.37.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.01371765136719, - "squared_error": 47.41359329223633, - "changed_scale_blocks": 756819 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.1575927734375, - "squared_error": 47.528076171875, - "changed_scale_blocks": 756533 - }, - "packed_byte_changes": 3814136, - "merged_bf16_changes": 17563604 - }, - { - "name": "backbone.layers.37.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.09864044189453, - "squared_error": 47.45244216918945, - "changed_scale_blocks": 758723 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.2398910522461, - "squared_error": 47.542259216308594, - "changed_scale_blocks": 759195 - }, - "packed_byte_changes": 3573079, - "merged_bf16_changes": 17486969 - }, - { - "name": "backbone.layers.38.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.66352844238281, - "squared_error": 27.079914093017578, - "changed_scale_blocks": 433292 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.75710678100586, - "squared_error": 27.15015983581543, - "changed_scale_blocks": 433571 - }, - "packed_byte_changes": 2204369, - "merged_bf16_changes": 10040425 - }, - { - "name": "backbone.layers.38.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.043045997619629, - "squared_error": 6.8354010581970215, - "changed_scale_blocks": 104886 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.050671577453613, - "squared_error": 6.8495073318481445, - "changed_scale_blocks": 104582 - }, - "packed_byte_changes": 503356, - "merged_bf16_changes": 2495617 - }, - { - "name": "backbone.layers.38.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.694908142089844, - "squared_error": 27.105579376220703, - "changed_scale_blocks": 434649 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.729736328125, - "squared_error": 27.137271881103516, - "changed_scale_blocks": 433862 - }, - "packed_byte_changes": 1630344, - "merged_bf16_changes": 9816398 - }, - { - "name": "backbone.layers.38.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.32619857788086, - "squared_error": 27.196975708007812, - "changed_scale_blocks": 424640 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.42517852783203, - "squared_error": 27.25893211364746, - "changed_scale_blocks": 424757 - }, - "packed_byte_changes": 2068089, - "merged_bf16_changes": 9975926 - }, - { - "name": "backbone.layers.38.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.96296215057373, - "squared_error": 6.851778507232666, - "changed_scale_blocks": 101386 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.974523544311523, - "squared_error": 6.864426612854004, - "changed_scale_blocks": 101384 - }, - "packed_byte_changes": 486031, - "merged_bf16_changes": 2492511 - }, - { - "name": "backbone.layers.39.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.36746215820312, - "squared_error": 47.45623016357422, - "changed_scale_blocks": 762019 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.44290161132812, - "squared_error": 47.51451110839844, - "changed_scale_blocks": 762146 - }, - "packed_byte_changes": 3040294, - "merged_bf16_changes": 17185085 - }, - { - "name": "backbone.layers.39.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.06636810302734, - "squared_error": 47.37819290161133, - "changed_scale_blocks": 756132 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.20793151855469, - "squared_error": 47.498695373535156, - "changed_scale_blocks": 755626 - }, - "packed_byte_changes": 3710952, - "merged_bf16_changes": 17528409 - }, - { - "name": "backbone.layers.39.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.16004943847656, - "squared_error": 47.43452453613281, - "changed_scale_blocks": 756163 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.31826782226562, - "squared_error": 47.57573699951172, - "changed_scale_blocks": 755645 - }, - "packed_byte_changes": 3808279, - "merged_bf16_changes": 17517212 - }, - { - "name": "backbone.layers.4.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.80106735229492, - "squared_error": 27.06269073486328, - "changed_scale_blocks": 437905 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.95250701904297, - "squared_error": 27.149215698242188, - "changed_scale_blocks": 438315 - }, - "packed_byte_changes": 2336294, - "merged_bf16_changes": 10060588 - }, - { - "name": "backbone.layers.4.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.088785171508789, - "squared_error": 6.793153285980225, - "changed_scale_blocks": 107713 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.125112533569336, - "squared_error": 6.818280220031738, - "changed_scale_blocks": 107628 - }, - "packed_byte_changes": 581893, - "merged_bf16_changes": 2518974 - }, - { - "name": "backbone.layers.4.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.78165817260742, - "squared_error": 27.13074493408203, - "changed_scale_blocks": 439219 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.87346649169922, - "squared_error": 27.223793029785156, - "changed_scale_blocks": 439254 - }, - "packed_byte_changes": 2255832, - "merged_bf16_changes": 10022138 - }, - { - "name": "backbone.layers.4.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.4749755859375, - "squared_error": 27.155635833740234, - "changed_scale_blocks": 434009 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.58317565917969, - "squared_error": 27.24495506286621, - "changed_scale_blocks": 433340 - }, - "packed_byte_changes": 2317572, - "merged_bf16_changes": 10076228 - }, - { - "name": "backbone.layers.4.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.142023086547852, - "squared_error": 6.782023906707764, - "changed_scale_blocks": 107606 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.15105152130127, - "squared_error": 6.798258304595947, - "changed_scale_blocks": 106973 - }, - "packed_byte_changes": 567620, - "merged_bf16_changes": 2505902 - }, - { - "name": "backbone.layers.40.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.696128845214844, - "squared_error": 27.078092575073242, - "changed_scale_blocks": 437304 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.82170486450195, - "squared_error": 27.168479919433594, - "changed_scale_blocks": 437416 - }, - "packed_byte_changes": 2343511, - "merged_bf16_changes": 10047826 - }, - { - "name": "backbone.layers.40.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.072616577148438, - "squared_error": 6.799140930175781, - "changed_scale_blocks": 105849 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.107688903808594, - "squared_error": 6.8239545822143555, - "changed_scale_blocks": 105924 - }, - "packed_byte_changes": 578208, - "merged_bf16_changes": 2515851 - }, - { - "name": "backbone.layers.40.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.79023361206055, - "squared_error": 27.10472869873047, - "changed_scale_blocks": 440129 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.8354377746582, - "squared_error": 27.139507293701172, - "changed_scale_blocks": 439656 - }, - "packed_byte_changes": 1795741, - "merged_bf16_changes": 9887790 - }, - { - "name": "backbone.layers.40.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.55284118652344, - "squared_error": 27.13604736328125, - "changed_scale_blocks": 434078 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.69804000854492, - "squared_error": 27.244781494140625, - "changed_scale_blocks": 434119 - }, - "packed_byte_changes": 2274205, - "merged_bf16_changes": 10031209 - }, - { - "name": "backbone.layers.40.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.914997100830078, - "squared_error": 6.862346649169922, - "changed_scale_blocks": 100933 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.943513870239258, - "squared_error": 6.8771467208862305, - "changed_scale_blocks": 100988 - }, - "packed_byte_changes": 498180, - "merged_bf16_changes": 2485003 - }, - { - "name": "backbone.layers.41.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.63038635253906, - "squared_error": 47.373016357421875, - "changed_scale_blocks": 773884 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.74267578125, - "squared_error": 47.439537048339844, - "changed_scale_blocks": 774511 - }, - "packed_byte_changes": 3010065, - "merged_bf16_changes": 17157501 - }, - { - "name": "backbone.layers.41.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.26085662841797, - "squared_error": 47.34027099609375, - "changed_scale_blocks": 761635 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.49362182617188, - "squared_error": 47.452449798583984, - "changed_scale_blocks": 763985 - }, - "packed_byte_changes": 3712369, - "merged_bf16_changes": 17539869 - }, - { - "name": "backbone.layers.41.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.29632568359375, - "squared_error": 47.3956298828125, - "changed_scale_blocks": 764927 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.40562438964844, - "squared_error": 47.4774284362793, - "changed_scale_blocks": 764864 - }, - "packed_byte_changes": 3390403, - "merged_bf16_changes": 17431677 - }, - { - "name": "backbone.layers.42.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.702842712402344, - "squared_error": 27.025035858154297, - "changed_scale_blocks": 437484 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.85614013671875, - "squared_error": 27.165054321289062, - "changed_scale_blocks": 436926 - }, - "packed_byte_changes": 2479062, - "merged_bf16_changes": 10056915 - }, - { - "name": "backbone.layers.42.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.143117904663086, - "squared_error": 6.7690887451171875, - "changed_scale_blocks": 107783 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.207053184509277, - "squared_error": 6.8169074058532715, - "changed_scale_blocks": 107844 - }, - "packed_byte_changes": 639608, - "merged_bf16_changes": 2518332 - }, - { - "name": "backbone.layers.42.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.904361724853516, - "squared_error": 27.106510162353516, - "changed_scale_blocks": 440415 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.9525260925293, - "squared_error": 27.1458797454834, - "changed_scale_blocks": 440302 - }, - "packed_byte_changes": 1863144, - "merged_bf16_changes": 9903172 - }, - { - "name": "backbone.layers.42.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.79168701171875, - "squared_error": 27.112302780151367, - "changed_scale_blocks": 439306 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.918067932128906, - "squared_error": 27.220001220703125, - "changed_scale_blocks": 438585 - }, - "packed_byte_changes": 2320129, - "merged_bf16_changes": 10020449 - }, - { - "name": "backbone.layers.42.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.082250595092773, - "squared_error": 6.803267478942871, - "changed_scale_blocks": 105899 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.097976684570312, - "squared_error": 6.816074848175049, - "changed_scale_blocks": 105830 - }, - "packed_byte_changes": 496679, - "merged_bf16_changes": 2494130 - }, - { - "name": "backbone.layers.43.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.59672546386719, - "squared_error": 47.360694885253906, - "changed_scale_blocks": 770015 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.68659973144531, - "squared_error": 47.43125534057617, - "changed_scale_blocks": 770677 - }, - "packed_byte_changes": 3016253, - "merged_bf16_changes": 17150225 - }, - { - "name": "backbone.layers.43.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.34446716308594, - "squared_error": 47.3411979675293, - "changed_scale_blocks": 765701 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.49988555908203, - "squared_error": 47.46274185180664, - "changed_scale_blocks": 766065 - }, - "packed_byte_changes": 3794695, - "merged_bf16_changes": 17549260 - }, - { - "name": "backbone.layers.43.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.28530883789062, - "squared_error": 47.36842346191406, - "changed_scale_blocks": 760333 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.43400573730469, - "squared_error": 47.46430587768555, - "changed_scale_blocks": 760198 - }, - "packed_byte_changes": 3447321, - "merged_bf16_changes": 17466975 - }, - { - "name": "backbone.layers.44.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.792442321777344, - "squared_error": 27.06794548034668, - "changed_scale_blocks": 439241 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.88636779785156, - "squared_error": 27.126380920410156, - "changed_scale_blocks": 439588 - }, - "packed_byte_changes": 2097299, - "merged_bf16_changes": 9996924 - }, - { - "name": "backbone.layers.44.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.14141845703125, - "squared_error": 6.779427528381348, - "changed_scale_blocks": 108109 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.15426254272461, - "squared_error": 6.797239780426025, - "changed_scale_blocks": 108169 - }, - "packed_byte_changes": 525965, - "merged_bf16_changes": 2497405 - }, - { - "name": "backbone.layers.44.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.85068893432617, - "squared_error": 27.077049255371094, - "changed_scale_blocks": 437475 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.89631652832031, - "squared_error": 27.097110748291016, - "changed_scale_blocks": 437713 - }, - "packed_byte_changes": 1653681, - "merged_bf16_changes": 9826812 - }, - { - "name": "backbone.layers.44.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.70995330810547, - "squared_error": 27.085586547851562, - "changed_scale_blocks": 439452 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.83536148071289, - "squared_error": 27.191450119018555, - "changed_scale_blocks": 438930 - }, - "packed_byte_changes": 2297051, - "merged_bf16_changes": 10016770 - }, - { - "name": "backbone.layers.44.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.081869125366211, - "squared_error": 6.803440570831299, - "changed_scale_blocks": 105373 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.10010051727295, - "squared_error": 6.814023017883301, - "changed_scale_blocks": 105433 - }, - "packed_byte_changes": 458791, - "merged_bf16_changes": 2480752 - }, - { - "name": "backbone.layers.45.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.57368469238281, - "squared_error": 47.40751647949219, - "changed_scale_blocks": 768782 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.64749145507812, - "squared_error": 47.467811584472656, - "changed_scale_blocks": 768264 - }, - "packed_byte_changes": 2804601, - "merged_bf16_changes": 17061600 - }, - { - "name": "backbone.layers.45.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.14527130126953, - "squared_error": 47.38916015625, - "changed_scale_blocks": 761650 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.28897094726562, - "squared_error": 47.49253845214844, - "changed_scale_blocks": 760929 - }, - "packed_byte_changes": 3768896, - "merged_bf16_changes": 17556773 - }, - { - "name": "backbone.layers.45.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.21233367919922, - "squared_error": 47.391910552978516, - "changed_scale_blocks": 762818 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.3114242553711, - "squared_error": 47.464962005615234, - "changed_scale_blocks": 762650 - }, - "packed_byte_changes": 3393666, - "merged_bf16_changes": 17451419 - }, - { - "name": "backbone.layers.46.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.793312072753906, - "squared_error": 27.069232940673828, - "changed_scale_blocks": 440375 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.878849029541016, - "squared_error": 27.125858306884766, - "changed_scale_blocks": 440095 - }, - "packed_byte_changes": 2014592, - "merged_bf16_changes": 9973815 - }, - { - "name": "backbone.layers.46.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.025177955627441, - "squared_error": 6.835558891296387, - "changed_scale_blocks": 104692 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.03026294708252, - "squared_error": 6.843270301818848, - "changed_scale_blocks": 104583 - }, - "packed_byte_changes": 388234, - "merged_bf16_changes": 2449508 - }, - { - "name": "backbone.layers.46.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.667667388916016, - "squared_error": 27.155208587646484, - "changed_scale_blocks": 434431 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.695281982421875, - "squared_error": 27.1744384765625, - "changed_scale_blocks": 434796 - }, - "packed_byte_changes": 1517986, - "merged_bf16_changes": 9751125 - }, - { - "name": "backbone.layers.46.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.498870849609375, - "squared_error": 27.21131706237793, - "changed_scale_blocks": 431444 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.540443420410156, - "squared_error": 27.255252838134766, - "changed_scale_blocks": 430871 - }, - "packed_byte_changes": 1710876, - "merged_bf16_changes": 9850795 - }, - { - "name": "backbone.layers.46.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.954573631286621, - "squared_error": 6.937809467315674, - "changed_scale_blocks": 100764 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.972150802612305, - "squared_error": 6.94871711730957, - "changed_scale_blocks": 100915 - }, - "packed_byte_changes": 440387, - "merged_bf16_changes": 2472055 - }, - { - "name": "backbone.layers.47.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.63693237304688, - "squared_error": 47.37879180908203, - "changed_scale_blocks": 776407 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.70797729492188, - "squared_error": 47.424896240234375, - "changed_scale_blocks": 776371 - }, - "packed_byte_changes": 2629068, - "merged_bf16_changes": 17044389 - }, - { - "name": "backbone.layers.47.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.40223693847656, - "squared_error": 47.347557067871094, - "changed_scale_blocks": 765998 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.577392578125, - "squared_error": 47.45609664916992, - "changed_scale_blocks": 766660 - }, - "packed_byte_changes": 3794300, - "merged_bf16_changes": 17556737 - }, - { - "name": "backbone.layers.47.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.42218017578125, - "squared_error": 47.3624267578125, - "changed_scale_blocks": 763066 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.53707885742188, - "squared_error": 47.44232177734375, - "changed_scale_blocks": 763087 - }, - "packed_byte_changes": 3383642, - "merged_bf16_changes": 17439867 - }, - { - "name": "backbone.layers.48.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.90557098388672, - "squared_error": 27.034095764160156, - "changed_scale_blocks": 441881 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 37.01151657104492, - "squared_error": 27.123878479003906, - "changed_scale_blocks": 441559 - }, - "packed_byte_changes": 2197017, - "merged_bf16_changes": 10007081 - }, - { - "name": "backbone.layers.48.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.115346908569336, - "squared_error": 6.7946014404296875, - "changed_scale_blocks": 107339 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.12849235534668, - "squared_error": 6.802264213562012, - "changed_scale_blocks": 107510 - }, - "packed_byte_changes": 414903, - "merged_bf16_changes": 2461216 - }, - { - "name": "backbone.layers.48.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.76905059814453, - "squared_error": 27.150436401367188, - "changed_scale_blocks": 437482 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.83323287963867, - "squared_error": 27.185867309570312, - "changed_scale_blocks": 438275 - }, - "packed_byte_changes": 1803334, - "merged_bf16_changes": 9875520 - }, - { - "name": "backbone.layers.48.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.64421844482422, - "squared_error": 27.11577606201172, - "changed_scale_blocks": 437289 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.70909881591797, - "squared_error": 27.16490364074707, - "changed_scale_blocks": 437508 - }, - "packed_byte_changes": 1815346, - "merged_bf16_changes": 9902357 - }, - { - "name": "backbone.layers.48.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.09391975402832, - "squared_error": 6.809078216552734, - "changed_scale_blocks": 106544 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.10519790649414, - "squared_error": 6.82123327255249, - "changed_scale_blocks": 106389 - }, - "packed_byte_changes": 484058, - "merged_bf16_changes": 2482739 - }, - { - "name": "backbone.layers.49.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.41964721679688, - "squared_error": 47.345645904541016, - "changed_scale_blocks": 773183 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.44599151611328, - "squared_error": 47.366355895996094, - "changed_scale_blocks": 771935 - }, - "packed_byte_changes": 2383739, - "merged_bf16_changes": 16886622 - }, - { - "name": "backbone.layers.49.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.36711120605469, - "squared_error": 47.40269470214844, - "changed_scale_blocks": 762912 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.54751586914062, - "squared_error": 47.52117156982422, - "changed_scale_blocks": 762860 - }, - "packed_byte_changes": 3902504, - "merged_bf16_changes": 17559776 - }, - { - "name": "backbone.layers.49.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.3221435546875, - "squared_error": 47.3704719543457, - "changed_scale_blocks": 767790 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.45811462402344, - "squared_error": 47.45301055908203, - "changed_scale_blocks": 768558 - }, - "packed_byte_changes": 3438146, - "merged_bf16_changes": 17458112 - }, - { - "name": "backbone.layers.5.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.39225769042969, - "squared_error": 47.38941955566406, - "changed_scale_blocks": 765078 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.47270202636719, - "squared_error": 47.466609954833984, - "changed_scale_blocks": 764791 - }, - "packed_byte_changes": 3503660, - "merged_bf16_changes": 17448693 - }, - { - "name": "backbone.layers.5.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.29411315917969, - "squared_error": 47.353553771972656, - "changed_scale_blocks": 762521 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.49703979492188, - "squared_error": 47.514488220214844, - "changed_scale_blocks": 762847 - }, - "packed_byte_changes": 3998741, - "merged_bf16_changes": 17599034 - }, - { - "name": "backbone.layers.5.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.35105895996094, - "squared_error": 47.41653823852539, - "changed_scale_blocks": 760784 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.68495178222656, - "squared_error": 47.64424133300781, - "changed_scale_blocks": 761341 - }, - "packed_byte_changes": 4336221, - "merged_bf16_changes": 17634178 - }, - { - "name": "backbone.layers.50.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.758140563964844, - "squared_error": 27.079242706298828, - "changed_scale_blocks": 437042 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.944908142089844, - "squared_error": 27.2027645111084, - "changed_scale_blocks": 437268 - }, - "packed_byte_changes": 2503898, - "merged_bf16_changes": 10079540 - }, - { - "name": "backbone.layers.50.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.186426162719727, - "squared_error": 6.784610748291016, - "changed_scale_blocks": 108703 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.221246719360352, - "squared_error": 6.816903114318848, - "changed_scale_blocks": 108745 - }, - "packed_byte_changes": 593780, - "merged_bf16_changes": 2510516 - }, - { - "name": "backbone.layers.50.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.84041976928711, - "squared_error": 27.08840560913086, - "changed_scale_blocks": 438121 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.89609909057617, - "squared_error": 27.13016700744629, - "changed_scale_blocks": 438466 - }, - "packed_byte_changes": 1680068, - "merged_bf16_changes": 9828753 - }, - { - "name": "backbone.layers.50.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.65130615234375, - "squared_error": 27.104459762573242, - "changed_scale_blocks": 437778 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.72466278076172, - "squared_error": 27.157421112060547, - "changed_scale_blocks": 438014 - }, - "packed_byte_changes": 2055984, - "merged_bf16_changes": 9991311 - }, - { - "name": "backbone.layers.50.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.090298652648926, - "squared_error": 6.788950443267822, - "changed_scale_blocks": 106103 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.120576858520508, - "squared_error": 6.805753707885742, - "changed_scale_blocks": 106250 - }, - "packed_byte_changes": 538771, - "merged_bf16_changes": 2502182 - }, - { - "name": "backbone.layers.51.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.23524475097656, - "squared_error": 47.50982666015625, - "changed_scale_blocks": 760064 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.28205871582031, - "squared_error": 47.53447341918945, - "changed_scale_blocks": 760267 - }, - "packed_byte_changes": 2362196, - "merged_bf16_changes": 16857270 - }, - { - "name": "backbone.layers.51.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.0574951171875, - "squared_error": 47.48149490356445, - "changed_scale_blocks": 753882 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.15447998046875, - "squared_error": 47.55788040161133, - "changed_scale_blocks": 752983 - }, - "packed_byte_changes": 3560338, - "merged_bf16_changes": 17498771 - }, - { - "name": "backbone.layers.51.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.07025146484375, - "squared_error": 47.421905517578125, - "changed_scale_blocks": 755311 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.1597671508789, - "squared_error": 47.48921585083008, - "changed_scale_blocks": 755159 - }, - "packed_byte_changes": 3156393, - "merged_bf16_changes": 17366427 - }, - { - "name": "backbone.layers.52.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.68498229980469, - "squared_error": 27.096630096435547, - "changed_scale_blocks": 436657 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.873878479003906, - "squared_error": 27.22201919555664, - "changed_scale_blocks": 436857 - }, - "packed_byte_changes": 2447332, - "merged_bf16_changes": 10062141 - }, - { - "name": "backbone.layers.52.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.127819061279297, - "squared_error": 6.783406734466553, - "changed_scale_blocks": 108081 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.16491413116455, - "squared_error": 6.802809238433838, - "changed_scale_blocks": 108002 - }, - "packed_byte_changes": 548860, - "merged_bf16_changes": 2503079 - }, - { - "name": "backbone.layers.52.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.722373962402344, - "squared_error": 27.148597717285156, - "changed_scale_blocks": 435876 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.75126647949219, - "squared_error": 27.19025230407715, - "changed_scale_blocks": 434874 - }, - "packed_byte_changes": 1868432, - "merged_bf16_changes": 9887328 - }, - { - "name": "backbone.layers.52.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.634056091308594, - "squared_error": 27.108367919921875, - "changed_scale_blocks": 436306 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.737388610839844, - "squared_error": 27.188602447509766, - "changed_scale_blocks": 436079 - }, - "packed_byte_changes": 2200047, - "merged_bf16_changes": 10006779 - }, - { - "name": "backbone.layers.52.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.03420352935791, - "squared_error": 6.8282976150512695, - "changed_scale_blocks": 104281 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.051586151123047, - "squared_error": 6.845522880554199, - "changed_scale_blocks": 104353 - }, - "packed_byte_changes": 473106, - "merged_bf16_changes": 2485829 - }, - { - "name": "backbone.layers.53.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.974884033203125, - "squared_error": 47.60799789428711, - "changed_scale_blocks": 749351 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.02677917480469, - "squared_error": 47.64311218261719, - "changed_scale_blocks": 748489 - }, - "packed_byte_changes": 2472075, - "merged_bf16_changes": 16962169 - }, - { - "name": "backbone.layers.53.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.73616027832031, - "squared_error": 47.52339553833008, - "changed_scale_blocks": 751214 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.82872772216797, - "squared_error": 47.58670425415039, - "changed_scale_blocks": 751039 - }, - "packed_byte_changes": 3408366, - "merged_bf16_changes": 17456385 - }, - { - "name": "backbone.layers.53.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.06885528564453, - "squared_error": 47.444419860839844, - "changed_scale_blocks": 759853 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.15724182128906, - "squared_error": 47.51889419555664, - "changed_scale_blocks": 759429 - }, - "packed_byte_changes": 3153843, - "merged_bf16_changes": 17364509 - }, - { - "name": "backbone.layers.54.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.77338409423828, - "squared_error": 27.0487060546875, - "changed_scale_blocks": 438169 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.85128402709961, - "squared_error": 27.1108455657959, - "changed_scale_blocks": 438386 - }, - "packed_byte_changes": 1918596, - "merged_bf16_changes": 9942395 - }, - { - "name": "backbone.layers.54.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.903303146362305, - "squared_error": 6.866729736328125, - "changed_scale_blocks": 102124 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.911737442016602, - "squared_error": 6.872180938720703, - "changed_scale_blocks": 102082 - }, - "packed_byte_changes": 338321, - "merged_bf16_changes": 2408628 - }, - { - "name": "backbone.layers.54.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.834659576416016, - "squared_error": 27.135875701904297, - "changed_scale_blocks": 438527 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.863853454589844, - "squared_error": 27.159587860107422, - "changed_scale_blocks": 438602 - }, - "packed_byte_changes": 1516399, - "merged_bf16_changes": 9738203 - }, - { - "name": "backbone.layers.54.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.37459182739258, - "squared_error": 27.238096237182617, - "changed_scale_blocks": 429520 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.44487762451172, - "squared_error": 27.284507751464844, - "changed_scale_blocks": 429541 - }, - "packed_byte_changes": 1917807, - "merged_bf16_changes": 9927477 - }, - { - "name": "backbone.layers.54.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.77547550201416, - "squared_error": 6.9699506759643555, - "changed_scale_blocks": 95355 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.784700393676758, - "squared_error": 6.977235794067383, - "changed_scale_blocks": 95217 - }, - "packed_byte_changes": 388935, - "merged_bf16_changes": 2458238 - }, - { - "name": "backbone.layers.55.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.35445404052734, - "squared_error": 47.359901428222656, - "changed_scale_blocks": 769695 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.40531921386719, - "squared_error": 47.408878326416016, - "changed_scale_blocks": 768736 - }, - "packed_byte_changes": 2850556, - "merged_bf16_changes": 17143353 - }, - { - "name": "backbone.layers.55.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.37393188476562, - "squared_error": 47.35872268676758, - "changed_scale_blocks": 763235 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.61998748779297, - "squared_error": 47.48493194580078, - "changed_scale_blocks": 764609 - }, - "packed_byte_changes": 3994854, - "merged_bf16_changes": 17565223 - }, - { - "name": "backbone.layers.55.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.34895324707031, - "squared_error": 47.36958312988281, - "changed_scale_blocks": 769341 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.5047607421875, - "squared_error": 47.49357604980469, - "changed_scale_blocks": 768322 - }, - "packed_byte_changes": 3643438, - "merged_bf16_changes": 17481835 - }, - { - "name": "backbone.layers.56.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.89326858520508, - "squared_error": 27.035385131835938, - "changed_scale_blocks": 442730 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.964141845703125, - "squared_error": 27.103363037109375, - "changed_scale_blocks": 442256 - }, - "packed_byte_changes": 1901019, - "merged_bf16_changes": 9773961 - }, - { - "name": "backbone.layers.56.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.085417747497559, - "squared_error": 6.80237340927124, - "changed_scale_blocks": 107080 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.207721710205078, - "squared_error": 6.891099452972412, - "changed_scale_blocks": 107465 - }, - "packed_byte_changes": 699565, - "merged_bf16_changes": 2485118 - }, - { - "name": "backbone.layers.56.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.81955337524414, - "squared_error": 27.123279571533203, - "changed_scale_blocks": 437339 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.847713470458984, - "squared_error": 27.140914916992188, - "changed_scale_blocks": 437479 - }, - "packed_byte_changes": 1127141, - "merged_bf16_changes": 9348539 - }, - { - "name": "backbone.layers.56.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.567893981933594, - "squared_error": 27.125709533691406, - "changed_scale_blocks": 436547 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.69207763671875, - "squared_error": 27.210403442382812, - "changed_scale_blocks": 437029 - }, - "packed_byte_changes": 1863706, - "merged_bf16_changes": 9740385 - }, - { - "name": "backbone.layers.56.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.069998741149902, - "squared_error": 6.810283184051514, - "changed_scale_blocks": 105973 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.088482856750488, - "squared_error": 6.826172351837158, - "changed_scale_blocks": 105859 - }, - "packed_byte_changes": 477392, - "merged_bf16_changes": 2461086 - }, - { - "name": "backbone.layers.57.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.57718658447266, - "squared_error": 47.36814880371094, - "changed_scale_blocks": 775834 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.60700988769531, - "squared_error": 47.3777961730957, - "changed_scale_blocks": 775935 - }, - "packed_byte_changes": 1701135, - "merged_bf16_changes": 15868353 - }, - { - "name": "backbone.layers.57.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.46617126464844, - "squared_error": 47.33961486816406, - "changed_scale_blocks": 772993 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.61878967285156, - "squared_error": 47.432159423828125, - "changed_scale_blocks": 773116 - }, - "packed_byte_changes": 3329010, - "merged_bf16_changes": 17330539 - }, - { - "name": "backbone.layers.57.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.60365295410156, - "squared_error": 47.39603042602539, - "changed_scale_blocks": 771219 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.66046905517578, - "squared_error": 47.42669677734375, - "changed_scale_blocks": 771307 - }, - "packed_byte_changes": 2482693, - "merged_bf16_changes": 16946986 - }, - { - "name": "backbone.layers.58.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.93636703491211, - "squared_error": 27.02027130126953, - "changed_scale_blocks": 442826 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.98753356933594, - "squared_error": 27.058082580566406, - "changed_scale_blocks": 442688 - }, - "packed_byte_changes": 1712083, - "merged_bf16_changes": 9813678 - }, - { - "name": "backbone.layers.58.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.088777542114258, - "squared_error": 6.7967753410339355, - "changed_scale_blocks": 107652 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.145649909973145, - "squared_error": 6.834855079650879, - "changed_scale_blocks": 107953 - }, - "packed_byte_changes": 595803, - "merged_bf16_changes": 2505013 - }, - { - "name": "backbone.layers.58.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.86149597167969, - "squared_error": 27.07076644897461, - "changed_scale_blocks": 441166 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.91656494140625, - "squared_error": 27.104923248291016, - "changed_scale_blocks": 440726 - }, - "packed_byte_changes": 1564734, - "merged_bf16_changes": 9566836 - }, - { - "name": "backbone.layers.58.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.628273010253906, - "squared_error": 27.146404266357422, - "changed_scale_blocks": 438679 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.791297912597656, - "squared_error": 27.268714904785156, - "changed_scale_blocks": 438402 - }, - "packed_byte_changes": 2096061, - "merged_bf16_changes": 9818378 - }, - { - "name": "backbone.layers.58.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.140776634216309, - "squared_error": 6.800271034240723, - "changed_scale_blocks": 106960 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.149040222167969, - "squared_error": 6.804337024688721, - "changed_scale_blocks": 107042 - }, - "packed_byte_changes": 326695, - "merged_bf16_changes": 2403429 - }, - { - "name": "backbone.layers.59.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.66641235351562, - "squared_error": 47.37487030029297, - "changed_scale_blocks": 773642 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.69369506835938, - "squared_error": 47.383907318115234, - "changed_scale_blocks": 774133 - }, - "packed_byte_changes": 1506066, - "merged_bf16_changes": 15603219 - }, - { - "name": "backbone.layers.59.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.57527923583984, - "squared_error": 47.330299377441406, - "changed_scale_blocks": 771329 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.72793579101562, - "squared_error": 47.44818878173828, - "changed_scale_blocks": 770631 - }, - "packed_byte_changes": 3421195, - "merged_bf16_changes": 17308315 - }, - { - "name": "backbone.layers.59.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.47987365722656, - "squared_error": 47.34129333496094, - "changed_scale_blocks": 766001 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.53377532958984, - "squared_error": 47.36557388305664, - "changed_scale_blocks": 766291 - }, - "packed_byte_changes": 2217419, - "merged_bf16_changes": 16768238 - }, - { - "name": "backbone.layers.6.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.78057098388672, - "squared_error": 27.054584503173828, - "changed_scale_blocks": 436453 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.879024505615234, - "squared_error": 27.146188735961914, - "changed_scale_blocks": 436496 - }, - "packed_byte_changes": 2205317, - "merged_bf16_changes": 10033822 - }, - { - "name": "backbone.layers.6.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.038930892944336, - "squared_error": 6.8026251792907715, - "changed_scale_blocks": 106118 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.065866470336914, - "squared_error": 6.819555282592773, - "changed_scale_blocks": 106589 - }, - "packed_byte_changes": 567885, - "merged_bf16_changes": 2512487 - }, - { - "name": "backbone.layers.6.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.57312774658203, - "squared_error": 27.112180709838867, - "changed_scale_blocks": 435357 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.56135559082031, - "squared_error": 27.203670501708984, - "changed_scale_blocks": 432427 - }, - "packed_byte_changes": 2262686, - "merged_bf16_changes": 10013075 - }, - { - "name": "backbone.layers.6.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.2861213684082, - "squared_error": 27.18181800842285, - "changed_scale_blocks": 428142 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.3906364440918, - "squared_error": 27.274734497070312, - "changed_scale_blocks": 427638 - }, - "packed_byte_changes": 2358280, - "merged_bf16_changes": 10052090 - }, - { - "name": "backbone.layers.6.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.037820816040039, - "squared_error": 6.796923637390137, - "changed_scale_blocks": 105511 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.064432144165039, - "squared_error": 6.8204569816589355, - "changed_scale_blocks": 105715 - }, - "packed_byte_changes": 535878, - "merged_bf16_changes": 2502428 - }, - { - "name": "backbone.layers.60.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.857357025146484, - "squared_error": 27.031274795532227, - "changed_scale_blocks": 442358 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.90869903564453, - "squared_error": 27.061687469482422, - "changed_scale_blocks": 442298 - }, - "packed_byte_changes": 1605686, - "merged_bf16_changes": 9780531 - }, - { - "name": "backbone.layers.60.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.018882751464844, - "squared_error": 6.82746696472168, - "changed_scale_blocks": 105559 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.025696754455566, - "squared_error": 6.841034889221191, - "changed_scale_blocks": 105458 - }, - "packed_byte_changes": 385840, - "merged_bf16_changes": 2423385 - }, - { - "name": "backbone.layers.60.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.99945068359375, - "squared_error": 27.085704803466797, - "changed_scale_blocks": 444673 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 37.033348083496094, - "squared_error": 27.113183975219727, - "changed_scale_blocks": 444705 - }, - "packed_byte_changes": 1377885, - "merged_bf16_changes": 9422336 - }, - { - "name": "backbone.layers.60.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.434783935546875, - "squared_error": 27.2128963470459, - "changed_scale_blocks": 431464 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.46234130859375, - "squared_error": 27.241310119628906, - "changed_scale_blocks": 431506 - }, - "packed_byte_changes": 1466262, - "merged_bf16_changes": 9698941 - }, - { - "name": "backbone.layers.60.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.029885292053223, - "squared_error": 6.846529960632324, - "changed_scale_blocks": 104670 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.041875839233398, - "squared_error": 6.852506637573242, - "changed_scale_blocks": 104639 - }, - "packed_byte_changes": 333100, - "merged_bf16_changes": 2405466 - }, - { - "name": "backbone.layers.61.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.68057250976562, - "squared_error": 47.383399963378906, - "changed_scale_blocks": 778051 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.6886978149414, - "squared_error": 47.38640594482422, - "changed_scale_blocks": 777621 - }, - "packed_byte_changes": 1277110, - "merged_bf16_changes": 15259093 - }, - { - "name": "backbone.layers.61.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.5614013671875, - "squared_error": 47.39662170410156, - "changed_scale_blocks": 771134 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.68569946289062, - "squared_error": 47.4802131652832, - "changed_scale_blocks": 771017 - }, - "packed_byte_changes": 3038725, - "merged_bf16_changes": 17175659 - }, - { - "name": "backbone.layers.61.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.60296630859375, - "squared_error": 47.42181396484375, - "changed_scale_blocks": 769675 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.6435546875, - "squared_error": 47.455909729003906, - "changed_scale_blocks": 769787 - }, - "packed_byte_changes": 2144195, - "merged_bf16_changes": 16654907 - }, - { - "name": "backbone.layers.62.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.84571838378906, - "squared_error": 27.03003692626953, - "changed_scale_blocks": 438205 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.86888885498047, - "squared_error": 27.054195404052734, - "changed_scale_blocks": 438140 - }, - "packed_byte_changes": 1412726, - "merged_bf16_changes": 9662181 - }, - { - "name": "backbone.layers.62.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.88232421875, - "squared_error": 6.855838775634766, - "changed_scale_blocks": 101880 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.9035005569458, - "squared_error": 6.873172760009766, - "changed_scale_blocks": 101728 - }, - "packed_byte_changes": 493048, - "merged_bf16_changes": 2464571 - }, - { - "name": "backbone.layers.62.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.93605041503906, - "squared_error": 27.083555221557617, - "changed_scale_blocks": 444148 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.93968200683594, - "squared_error": 27.09088134765625, - "changed_scale_blocks": 444042 - }, - "packed_byte_changes": 892463, - "merged_bf16_changes": 8989087 - }, - { - "name": "backbone.layers.62.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.239437103271484, - "squared_error": 27.265365600585938, - "changed_scale_blocks": 426623 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.282798767089844, - "squared_error": 27.301668167114258, - "changed_scale_blocks": 426089 - }, - "packed_byte_changes": 1672797, - "merged_bf16_changes": 9809067 - }, - { - "name": "backbone.layers.62.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.936456680297852, - "squared_error": 6.855668067932129, - "changed_scale_blocks": 101608 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 8.93877124786377, - "squared_error": 6.859035015106201, - "changed_scale_blocks": 101522 - }, - "packed_byte_changes": 268577, - "merged_bf16_changes": 2350186 - }, - { - "name": "backbone.layers.63.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.66661071777344, - "squared_error": 47.324851989746094, - "changed_scale_blocks": 777477 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.65339660644531, - "squared_error": 47.33160400390625, - "changed_scale_blocks": 777207 - }, - "packed_byte_changes": 1080302, - "merged_bf16_changes": 14626285 - }, - { - "name": "backbone.layers.63.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.74222564697266, - "squared_error": 47.34233856201172, - "changed_scale_blocks": 772366 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.81884765625, - "squared_error": 47.379581451416016, - "changed_scale_blocks": 772370 - }, - "packed_byte_changes": 2700684, - "merged_bf16_changes": 17026751 - }, - { - "name": "backbone.layers.63.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.71851348876953, - "squared_error": 47.361732482910156, - "changed_scale_blocks": 773616 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.75166320800781, - "squared_error": 47.382904052734375, - "changed_scale_blocks": 772941 - }, - "packed_byte_changes": 1879522, - "merged_bf16_changes": 16333213 - }, - { - "name": "backbone.layers.64.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.89783477783203, - "squared_error": 27.012939453125, - "changed_scale_blocks": 441107 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 37.01726531982422, - "squared_error": 27.09160614013672, - "changed_scale_blocks": 442087 - }, - "packed_byte_changes": 1957001, - "merged_bf16_changes": 9877931 - }, - { - "name": "backbone.layers.64.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.225997924804688, - "squared_error": 6.757835865020752, - "changed_scale_blocks": 110591 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.259174346923828, - "squared_error": 6.777194976806641, - "changed_scale_blocks": 110636 - }, - "packed_byte_changes": 524443, - "merged_bf16_changes": 2486117 - }, - { - "name": "backbone.layers.64.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.85458755493164, - "squared_error": 27.075895309448242, - "changed_scale_blocks": 439259 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.862632751464844, - "squared_error": 27.077322006225586, - "changed_scale_blocks": 439648 - }, - "packed_byte_changes": 789125, - "merged_bf16_changes": 8859145 - }, - { - "name": "backbone.layers.64.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.888145446777344, - "squared_error": 27.025264739990234, - "changed_scale_blocks": 443567 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.92539978027344, - "squared_error": 27.04962921142578, - "changed_scale_blocks": 443136 - }, - "packed_byte_changes": 1358395, - "merged_bf16_changes": 9603299 - }, - { - "name": "backbone.layers.64.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.22160530090332, - "squared_error": 6.760058403015137, - "changed_scale_blocks": 110416 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.240095138549805, - "squared_error": 6.780520439147949, - "changed_scale_blocks": 110186 - }, - "packed_byte_changes": 475732, - "merged_bf16_changes": 2463389 - }, - { - "name": "backbone.layers.65.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.75521850585938, - "squared_error": 47.4068603515625, - "changed_scale_blocks": 775460 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.76716613769531, - "squared_error": 47.416168212890625, - "changed_scale_blocks": 774950 - }, - "packed_byte_changes": 976856, - "merged_bf16_changes": 14192097 - }, - { - "name": "backbone.layers.65.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.71739196777344, - "squared_error": 47.3410758972168, - "changed_scale_blocks": 779835 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.7284927368164, - "squared_error": 47.353641510009766, - "changed_scale_blocks": 779090 - }, - "packed_byte_changes": 2192081, - "merged_bf16_changes": 16712752 - }, - { - "name": "backbone.layers.65.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.7364501953125, - "squared_error": 47.357749938964844, - "changed_scale_blocks": 774043 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.78224182128906, - "squared_error": 47.383445739746094, - "changed_scale_blocks": 773977 - }, - "packed_byte_changes": 1810229, - "merged_bf16_changes": 16255582 - }, - { - "name": "backbone.layers.66.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.895652770996094, - "squared_error": 27.023509979248047, - "changed_scale_blocks": 444629 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.93517303466797, - "squared_error": 27.05935287475586, - "changed_scale_blocks": 444520 - }, - "packed_byte_changes": 1592677, - "merged_bf16_changes": 9739382 - }, - { - "name": "backbone.layers.66.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.129745483398438, - "squared_error": 6.786417007446289, - "changed_scale_blocks": 108657 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.147880554199219, - "squared_error": 6.79976749420166, - "changed_scale_blocks": 108736 - }, - "packed_byte_changes": 459392, - "merged_bf16_changes": 2444359 - }, - { - "name": "backbone.layers.66.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.92998504638672, - "squared_error": 27.096656799316406, - "changed_scale_blocks": 440447 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.92471694946289, - "squared_error": 27.095623016357422, - "changed_scale_blocks": 440655 - }, - "packed_byte_changes": 647913, - "merged_bf16_changes": 8528876 - }, - { - "name": "backbone.layers.66.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.717308044433594, - "squared_error": 27.115650177001953, - "changed_scale_blocks": 436373 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.76991271972656, - "squared_error": 27.14719009399414, - "changed_scale_blocks": 435864 - }, - "packed_byte_changes": 1499361, - "merged_bf16_changes": 9709448 - }, - { - "name": "backbone.layers.66.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.17016315460205, - "squared_error": 6.782525062561035, - "changed_scale_blocks": 108612 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.172323226928711, - "squared_error": 6.785078048706055, - "changed_scale_blocks": 108642 - }, - "packed_byte_changes": 278687, - "merged_bf16_changes": 2346889 - }, - { - "name": "backbone.layers.67.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.6478500366211, - "squared_error": 47.39223861694336, - "changed_scale_blocks": 769833 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.66036987304688, - "squared_error": 47.40013885498047, - "changed_scale_blocks": 770059 - }, - "packed_byte_changes": 855022, - "merged_bf16_changes": 13695555 - }, - { - "name": "backbone.layers.67.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.65272521972656, - "squared_error": 47.333534240722656, - "changed_scale_blocks": 778389 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.67533874511719, - "squared_error": 47.35733413696289, - "changed_scale_blocks": 778060 - }, - "packed_byte_changes": 2012805, - "merged_bf16_changes": 16513009 - }, - { - "name": "backbone.layers.67.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.59417724609375, - "squared_error": 47.31169128417969, - "changed_scale_blocks": 777690 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.62214660644531, - "squared_error": 47.334068298339844, - "changed_scale_blocks": 777624 - }, - "packed_byte_changes": 1757108, - "merged_bf16_changes": 16188854 - }, - { - "name": "backbone.layers.68.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.86163330078125, - "squared_error": 27.03525161743164, - "changed_scale_blocks": 443890 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.892616271972656, - "squared_error": 27.058422088623047, - "changed_scale_blocks": 443714 - }, - "packed_byte_changes": 1321431, - "merged_bf16_changes": 9551101 - }, - { - "name": "backbone.layers.68.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.190608978271484, - "squared_error": 6.7662177085876465, - "changed_scale_blocks": 109851 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.210792541503906, - "squared_error": 6.778853416442871, - "changed_scale_blocks": 109724 - }, - "packed_byte_changes": 414456, - "merged_bf16_changes": 2425417 - }, - { - "name": "backbone.layers.68.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.79380798339844, - "squared_error": 27.11514663696289, - "changed_scale_blocks": 438471 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.80305099487305, - "squared_error": 27.115154266357422, - "changed_scale_blocks": 438572 - }, - "packed_byte_changes": 512903, - "merged_bf16_changes": 7989061 - }, - { - "name": "backbone.layers.68.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.88016891479492, - "squared_error": 27.09829330444336, - "changed_scale_blocks": 438973 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.89105987548828, - "squared_error": 27.115707397460938, - "changed_scale_blocks": 438969 - }, - "packed_byte_changes": 1311837, - "merged_bf16_changes": 9556506 - }, - { - "name": "backbone.layers.68.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.189908981323242, - "squared_error": 6.787233352661133, - "changed_scale_blocks": 108700 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.198338508605957, - "squared_error": 6.788807392120361, - "changed_scale_blocks": 108728 - }, - "packed_byte_changes": 226199, - "merged_bf16_changes": 2283354 - }, - { - "name": "backbone.layers.69.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.64434051513672, - "squared_error": 47.36891555786133, - "changed_scale_blocks": 770453 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.64288330078125, - "squared_error": 47.366695404052734, - "changed_scale_blocks": 770273 - }, - "packed_byte_changes": 515414, - "merged_bf16_changes": 11317731 - }, - { - "name": "backbone.layers.69.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.62332153320312, - "squared_error": 47.32560729980469, - "changed_scale_blocks": 777313 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.65489196777344, - "squared_error": 47.34916687011719, - "changed_scale_blocks": 777187 - }, - "packed_byte_changes": 1896721, - "merged_bf16_changes": 16296997 - }, - { - "name": "backbone.layers.69.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.65557861328125, - "squared_error": 47.30464172363281, - "changed_scale_blocks": 772629 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.67073059082031, - "squared_error": 47.318824768066406, - "changed_scale_blocks": 772460 - }, - "packed_byte_changes": 1637493, - "merged_bf16_changes": 15889873 - }, - { - "name": "backbone.layers.7.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.18142700195312, - "squared_error": 47.442543029785156, - "changed_scale_blocks": 761849 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.31427001953125, - "squared_error": 47.5344352722168, - "changed_scale_blocks": 761439 - }, - "packed_byte_changes": 3488680, - "merged_bf16_changes": 17458775 - }, - { - "name": "backbone.layers.7.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.24542236328125, - "squared_error": 47.366477966308594, - "changed_scale_blocks": 769479 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.39982604980469, - "squared_error": 47.49114227294922, - "changed_scale_blocks": 769602 - }, - "packed_byte_changes": 3932731, - "merged_bf16_changes": 17602167 - }, - { - "name": "backbone.layers.7.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.10650634765625, - "squared_error": 47.36436462402344, - "changed_scale_blocks": 761558 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.29986572265625, - "squared_error": 47.48699951171875, - "changed_scale_blocks": 762067 - }, - "packed_byte_changes": 3890307, - "merged_bf16_changes": 17599326 - }, - { - "name": "backbone.layers.70.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.88714599609375, - "squared_error": 27.035587310791016, - "changed_scale_blocks": 444177 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.899078369140625, - "squared_error": 27.051048278808594, - "changed_scale_blocks": 443781 - }, - "packed_byte_changes": 1198847, - "merged_bf16_changes": 9290100 - }, - { - "name": "backbone.layers.70.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.223716735839844, - "squared_error": 6.767487049102783, - "changed_scale_blocks": 109745 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.23514175415039, - "squared_error": 6.786311149597168, - "changed_scale_blocks": 109720 - }, - "packed_byte_changes": 507423, - "merged_bf16_changes": 2461079 - }, - { - "name": "backbone.layers.70.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.684844970703125, - "squared_error": 27.123226165771484, - "changed_scale_blocks": 436644 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.682308197021484, - "squared_error": 27.123428344726562, - "changed_scale_blocks": 436681 - }, - "packed_byte_changes": 354314, - "merged_bf16_changes": 7091616 - }, - { - "name": "backbone.layers.70.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.949729919433594, - "squared_error": 27.006649017333984, - "changed_scale_blocks": 442087 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 37.113651275634766, - "squared_error": 27.13768768310547, - "changed_scale_blocks": 441965 - }, - "packed_byte_changes": 2314184, - "merged_bf16_changes": 9726058 - }, - { - "name": "backbone.layers.70.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.17043685913086, - "squared_error": 6.7710347175598145, - "changed_scale_blocks": 109446 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.171856880187988, - "squared_error": 6.773630142211914, - "changed_scale_blocks": 109290 - }, - "packed_byte_changes": 246792, - "merged_bf16_changes": 2287855 - }, - { - "name": "backbone.layers.71.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.42701721191406, - "squared_error": 47.46422576904297, - "changed_scale_blocks": 767823 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.42674255371094, - "squared_error": 47.46379852294922, - "changed_scale_blocks": 767776 - }, - "packed_byte_changes": 131488, - "merged_bf16_changes": 5640653 - }, - { - "name": "backbone.layers.71.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.53834533691406, - "squared_error": 47.34141540527344, - "changed_scale_blocks": 776844 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.55842590332031, - "squared_error": 47.35210418701172, - "changed_scale_blocks": 776427 - }, - "packed_byte_changes": 1534318, - "merged_bf16_changes": 15575892 - }, - { - "name": "backbone.layers.71.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.63694763183594, - "squared_error": 47.33809280395508, - "changed_scale_blocks": 777483 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.64369201660156, - "squared_error": 47.33777618408203, - "changed_scale_blocks": 777355 - }, - "packed_byte_changes": 1192139, - "merged_bf16_changes": 14703530 - }, - { - "name": "backbone.layers.8.mixer.g_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.78464889526367, - "squared_error": 27.02900505065918, - "changed_scale_blocks": 440727 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.90562438964844, - "squared_error": 27.117496490478516, - "changed_scale_blocks": 441034 - }, - "packed_byte_changes": 2342141, - "merged_bf16_changes": 10064298 - }, - { - "name": "backbone.layers.8.mixer.k_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.128093719482422, - "squared_error": 6.783987045288086, - "changed_scale_blocks": 108205 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.159802436828613, - "squared_error": 6.802311897277832, - "changed_scale_blocks": 108663 - }, - "packed_byte_changes": 570316, - "merged_bf16_changes": 2514270 - }, - { - "name": "backbone.layers.8.mixer.o_proj.weight", - "shape": [ - 2560, - 4096 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.659515380859375, - "squared_error": 27.08603286743164, - "changed_scale_blocks": 435105 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.72933578491211, - "squared_error": 27.172008514404297, - "changed_scale_blocks": 434036 - }, - "packed_byte_changes": 2269267, - "merged_bf16_changes": 10010651 - }, - { - "name": "backbone.layers.8.mixer.q_proj.weight", - "shape": [ - 4096, - 2560 - ], - "nvfp4-base": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.6406364440918, - "squared_error": 27.099891662597656, - "changed_scale_blocks": 436705 - }, - "nvfp4-merged": { - "elements": 10485760, - "blocks": 655360, - "maxabs_squared_error": 36.770233154296875, - "squared_error": 27.1936092376709, - "changed_scale_blocks": 436792 - }, - "packed_byte_changes": 2289915, - "merged_bf16_changes": 10038432 - }, - { - "name": "backbone.layers.8.mixer.v_proj.weight", - "shape": [ - 1024, - 2560 - ], - "nvfp4-base": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.105602264404297, - "squared_error": 6.774622917175293, - "changed_scale_blocks": 106932 - }, - "nvfp4-merged": { - "elements": 2621440, - "blocks": 163840, - "maxabs_squared_error": 9.131813049316406, - "squared_error": 6.798840522766113, - "changed_scale_blocks": 106801 - }, - "packed_byte_changes": 580913, - "merged_bf16_changes": 2504276 - }, - { - "name": "backbone.layers.9.mixer.down_proj.weight", - "shape": [ - 2560, - 7168 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 63.9753532409668, - "squared_error": 47.43384552001953, - "changed_scale_blocks": 761512 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.10872650146484, - "squared_error": 47.51343536376953, - "changed_scale_blocks": 762177 - }, - "packed_byte_changes": 3425456, - "merged_bf16_changes": 17438778 - }, - { - "name": "backbone.layers.9.mixer.gate_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.30510711669922, - "squared_error": 47.324317932128906, - "changed_scale_blocks": 769008 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.4754409790039, - "squared_error": 47.46645736694336, - "changed_scale_blocks": 769126 - }, - "packed_byte_changes": 3955980, - "merged_bf16_changes": 17622964 - }, - { - "name": "backbone.layers.9.mixer.up_proj.weight", - "shape": [ - 7168, - 2560 - ], - "nvfp4-base": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.286376953125, - "squared_error": 47.36985778808594, - "changed_scale_blocks": 767617 - }, - "nvfp4-merged": { - "elements": 18350080, - "blocks": 1146880, - "maxabs_squared_error": 64.46890258789062, - "squared_error": 47.50370788574219, - "changed_scale_blocks": 767988 - }, - "packed_byte_changes": 3777949, - "merged_bf16_changes": 17567517 + "producer": { + "name": "modelopt", + "version": "0.0.1.dev1+gfcc59e3a9" + }, + "quantization": { + "quant_cfg": [ + { + "quantizer_name": "*", + "enable": false + }, + { + "quantizer_name": "*weight_quantizer", + "cfg": { + "num_bits": [ + 2, + 1 + ], + "effective_bits": 4.5, + "block_sizes": { + "-1": 16, + "type": "static", + "scale_bits": [ + 4, + 3 + ] + } + } + }, + { + "quantizer_name": "*gdn_state_quantizer", + "enable": false + }, + { + "quantizer_name": "*gdn_w_quantizer", + "enable": false + }, + { + "quantizer_name": "*block_sparse_moe.gate*", + "enable": false + }, + { + "quantizer_name": "*linear_attn.conv1d*", + "enable": false + }, + { + "quantizer_name": "*self_attention.conv1d*", + "enable": false + }, + { + "quantizer_name": "*linear_attn.in_proj_a*", + "enable": false + }, + { + "quantizer_name": "*linear_attn.in_proj_b*", + "enable": false + }, + { + "quantizer_name": "*lm_head*", + "enable": false + }, + { + "quantizer_name": "*mixer.conv1d*", + "enable": false + }, + { + "quantizer_name": "*mlp.gate.*", + "enable": false + }, + { + "quantizer_name": "*mlp.shared_expert_gate.*", + "enable": false + }, + { + "quantizer_name": "*output_layer*", + "enable": false + }, + { + "quantizer_name": "*proj_out.*", + "enable": false + }, + { + "quantizer_name": "*router*", + "enable": false + }, + { + "quantizer_name": "*mtp.*", + "enable": false + }, + { + "quantizer_name": "output.*", + "enable": false + }, + { + "quantizer_name": "*embed_vision*", + "enable": false + }, + { + "quantizer_name": "*vision_tower*", + "enable": false + }, + { + "quantizer_name": "*visual*", + "enable": false + }, + { + "quantizer_name": "*vision_model*", + "enable": false + }, + { + "quantizer_name": "*multi_modal_projector*", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.BatchNorm1d", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.BatchNorm2d", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.BatchNorm3d", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.LeakyReLU", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.Embedding", + "enable": false + }, + { + "quantizer_name": "*", + "parent_class": "nn.LayerNorm", + "enable": false + } + ], + "algorithm": { + "method": "local_hessian" } - ], - "elapsed_s": 52.77437930973247 + }, + "loops": 2, + "script_sha256": "5b4755c6f19f54539e6fe63ccb3172e833cb0602373610bc9966408bc7dececd", + "torch": "2.10.0+cu128", + "weights_sha256": "62bc20a6fbafaf157154650dfb4e9ec32a19d67e8e1f60ddcc3318fd37d6ccb4", + "seconds": 523.3408162756823 }