{ "variants": { "q8f16": { "text": { "decoder": { "inv_freq_matches_theta_1e7": true, "float_max_abs_diff": 0.3125, "float_max_abs_diff_by_kind": { "model.layers.N.input_layernorm.weight": 0.046875, "model.layers.N.gdn.conv1d.weight": 0.06884765625, "model.layers.N.gdn.conv1d.weight_3d": 0.06884765625, "model.layers.N.gdn.A_neg_exp": 0.08991670608520508, "model.layers.N.gdn.dt_bias": 0.005859375, "model.layers.N.gdn.norm.weight": 0.01953125, "model.layers.N.post_attention_layernorm.weight": 0.0234375, "model.layers.N.attn.q_norm.layernorm.weight": 0.0703125, "model.layers.N.attn.k_norm.layernorm.weight": 0.064453125, "model.layers.N.final_norm_layernorm.weight": 0.3125 }, "q4_tensors": 0, "q4_identical": 0, "q4_worst_mismatch_fraction": 0.0 }, "embed": {}, "vocab": 248094, "decoder_model_merged_q8f16.onnx": { "graph_sha256": "2460943b101cb79960af673faecf5d1ad72d027ba861725fb6c6abd3a968ee4e", "data_bytes": 824095552 }, "embed_tokens_q8f16.onnx": { "graph_sha256": "605d1d2d6f65c2ae15de05f05d59624bf41e633e93c895159d233282d46ce759", "data_bytes": 277865280 } }, "chunked_prefill": { "tool": "scripts/models/bonsai/chunk_gdn_prefill.py", "source_sha256": "2460943b101cb79960af673faecf5d1ad72d027ba861725fb6c6abd3a968ee4e", "output_sha256": "c34d4ebadef268ff8c2124b33111376686d71b8111048179604d2c93dcf1601a", "output_bytes": 2879267, "gdn_layers_rewritten": 18, "chunk_tokens": 128, "context_tokens": 262144, "prefix_scan_steps": 12, "max_prompt_tokens": 524160, "logits": "last position only" } }, "q4f16": { "text": { "decoder": { "inv_freq_matches_theta_1e7": true, "float_max_abs_diff": 0.3125, "float_max_abs_diff_by_kind": { "model.layers.N.input_layernorm.weight": 0.046875, "model.layers.N.gdn.conv1d.weight": 0.06884765625, "model.layers.N.gdn.conv1d.weight_3d": 0.06884765625, "model.layers.N.gdn.A_neg_exp": 0.08991670608520508, "model.layers.N.gdn.dt_bias": 0.005859375, "model.layers.N.gdn.norm.weight": 0.01953125, "model.layers.N.post_attention_layernorm.weight": 0.0234375, "model.layers.N.attn.q_norm.layernorm.weight": 0.0703125, "model.layers.N.attn.k_norm.layernorm.weight": 0.064453125, "model.layers.N.final_norm_layernorm.weight": 0.3125 }, "q4_tensors": 0, "q4_identical": 0, "q4_worst_mismatch_fraction": 0.0 }, "embed": {}, "vocab": 248094, "decoder_model_merged_q4f16.onnx": { "graph_sha256": "f1b91931471cd931b34e89a8de276578b0f178a299a08f12731d0a5dd035bacb", "data_bytes": 436528480 }, "embed_tokens_q4f16.onnx": { "graph_sha256": "59eac222a8ffe689406113bba0ded116ed49382d5d81971256eea5091d0883a4", "data_bytes": 146871648 } }, "chunked_prefill": { "tool": "scripts/models/bonsai/chunk_gdn_prefill.py", "source_sha256": "f1b91931471cd931b34e89a8de276578b0f178a299a08f12731d0a5dd035bacb", "output_sha256": "49d001daf1555da07a07e9b97a1d4494ec4ead71120e4fb1ed6fb73f4882b096", "output_bytes": 2869812, "gdn_layers_rewritten": 18, "chunk_tokens": 128, "context_tokens": 262144, "prefix_scan_steps": 12, "max_prompt_tokens": 524160, "logits": "last position only" } } }, "vision": { "bucket_mismatches_cpu": [], "bucket_mismatches_cuda": [ [ 25, 1, 4 ], [ 25, 2, 8 ], [ 25, 3, 12 ], [ 25, 7, 28 ], [ 25, 13, 52 ], [ 25, 32, 128 ], [ 25, 56, 224 ], [ 25, 64, 256 ], [ 25, 70, 280 ], [ 25, 97, 388 ], [ 25, 128, 512 ], [ 49, 1, 4 ], [ 49, 2, 8 ], [ 49, 3, 12 ], [ 49, 7, 28 ], [ 49, 13, 52 ], [ 49, 32, 128 ], [ 49, 56, 224 ], [ 49, 64, 256 ], [ 49, 70, 280 ], [ 49, 97, 388 ], [ 49, 128, 512 ], [ 50, 1, 6 ], [ 50, 2, 12 ], [ 50, 3, 18 ], [ 50, 7, 42 ], [ 50, 13, 78 ], [ 50, 32, 192 ], [ 50, 56, 336 ], [ 50, 64, 384 ], [ 50, 70, 420 ], [ 50, 97, 582 ], [ 50, 128, 768 ], [ 55, 1, 4 ], [ 55, 2, 8 ], [ 55, 3, 12 ], [ 55, 7, 28 ], [ 55, 13, 52 ], [ 55, 32, 128 ], [ 55, 56, 224 ], [ 55, 64, 256 ], [ 55, 70, 280 ], [ 55, 97, 388 ], [ 55, 128, 512 ], [ 82, 1, 1 ], [ 82, 2, 2 ], [ 82, 3, 3 ], [ 82, 7, 7 ], [ 82, 13, 13 ], [ 82, 32, 32 ], [ 82, 56, 56 ], [ 82, 64, 64 ], [ 82, 70, 70 ], [ 82, 97, 97 ], [ 82, 128, 128 ], [ 94, 1, 1 ], [ 94, 2, 2 ], [ 94, 3, 3 ], [ 94, 7, 7 ], [ 94, 13, 13 ], [ 94, 32, 32 ], [ 94, 56, 56 ], [ 94, 64, 64 ], [ 94, 70, 70 ], [ 94, 97, 97 ], [ 94, 128, 128 ], [ 98, 1, 7 ], [ 98, 2, 14 ], [ 98, 3, 21 ], [ 98, 7, 49 ], [ 98, 13, 91 ], [ 98, 32, 224 ], [ 98, 56, 392 ], [ 98, 64, 448 ], [ 98, 70, 490 ], [ 98, 97, 679 ], [ 98, 128, 896 ], [ 100, 1, 6 ], [ 100, 2, 12 ], [ 100, 3, 18 ], [ 100, 7, 42 ], [ 100, 13, 78 ], [ 100, 32, 192 ], [ 100, 56, 336 ], [ 100, 64, 384 ], [ 100, 70, 420 ], [ 100, 97, 582 ], [ 100, 128, 768 ], [ 110, 1, 7 ], [ 110, 2, 14 ], [ 110, 3, 21 ], [ 110, 7, 49 ], [ 110, 13, 91 ], [ 110, 32, 224 ], [ 110, 56, 392 ], [ 110, 64, 448 ], [ 110, 70, 490 ], [ 110, 97, 679 ], [ 110, 128, 896 ], [ 115, 1, 4 ], [ 115, 2, 8 ], [ 115, 3, 12 ], [ 115, 7, 28 ], [ 115, 13, 52 ], [ 115, 32, 128 ], [ 115, 56, 224 ], [ 115, 64, 256 ], [ 115, 70, 280 ], [ 115, 97, 388 ], [ 115, 128, 512 ], [ 122, 1, 1 ], [ 122, 2, 2 ], [ 122, 3, 3 ], [ 122, 7, 7 ], [ 122, 13, 13 ], [ 122, 32, 32 ], [ 122, 56, 56 ], [ 122, 64, 64 ], [ 122, 70, 70 ], [ 122, 97, 97 ], [ 122, 128, 128 ] ], "export_seconds": 36.3, "ops": { "Shape": 3, "Squeeze": 2, "Cast": 8, "Div": 8, "Mul": 205, "Conv": 1, "Reshape": 129, "Concat": 8, "Transpose": 116, "Range": 2, "Floor": 2, "Clip": 2, "Unsqueeze": 3, "Add": 283, "Gather": 1, "LayerNormalization": 58, "MatMul": 224, "Softmax": 28, "Tanh": 28, "Gemm": 2, "ReduceMean": 1, "Erf": 1 }, "parity": [ { "image": "final_test-relocation_editor-desktop-completed.png", "crop_hw": [ 24, 40 ], "crops": 1, "torch_module_rel": 5.274350769468583e-06, "ort_fp32_rel": 4.5221663640404586e-06, "ort_fp16_rel": 0.0028502538334578276, "ort_fp16_cos": 0.9999967813491821, "ref_absmax": 5.001994609832764 }, { "image": "final_test-relocation_editor-desktop-completed.png", "crop_hw": [ 32, 32 ], "crops": 6, "torch_module_rel": 3.0137891826598207e-06, "ort_fp32_rel": 4.540931058727438e-06, "ort_fp16_rel": 0.002111530164256692, "ort_fp16_cos": 1.0000048875808716, "ref_absmax": 2.9292280673980713 }, { "image": "final_test-relocation_editor-desktop-not_completed.png", "crop_hw": [ 24, 40 ], "crops": 1, "torch_module_rel": 2.24648119910853e-05, "ort_fp32_rel": 2.3916370992083102e-05, "ort_fp16_rel": 0.013344613835215569, "ort_fp16_cos": 0.9999114871025085, "ref_absmax": 4.9408392906188965 }, { "image": "final_test-relocation_editor-desktop-not_completed.png", "crop_hw": [ 32, 32 ], "crops": 6, "torch_module_rel": 4.114119747100631e-06, "ort_fp32_rel": 6.51703703624662e-06, "ort_fp16_rel": 0.0026564407162368298, "ort_fp16_cos": 1.0000035762786865, "ref_absmax": 2.987318992614746 } ], "vision_encoder.onnx": { "graph_sha256": "53fa2d068d9627db06119168cf3096e3d6461005f09194d952162c517e6762a5" }, "vision_encoder_fp16.onnx": { "graph_sha256": "75127ae25e1313413dc829713f170b876f2f46d2a439369b87f0ff85a186696b" } } }