{ "description": "One complete six-layer Baberu decoder graph shared by prefill and token steps; weights, dimensions, vocabulary, and generation logic are unchanged", "fp32": { "graph": { "bytes": 133619464, "sha256": "494eb3552d9d6b16034618f73d1b1ae92d14777e8f62f56f0691065dc9d985cd", "opsets": { "ai.onnx": 17 }, "operators": { "Add": 57, "Cast": 76, "Concat": 74, "Constant": 541, "ConstantOfShape": 13, "Div": 38, "Equal": 13, "Expand": 13, "Gather": 77, "LessOrEqual": 1, "MatMul": 56, "Mul": 149, "Neg": 12, "Range": 2, "ReduceMean": 25, "Reshape": 49, "Shape": 88, "Sigmoid": 6, "Slice": 31, "Softmax": 6, "Sqrt": 25, "Sub": 1, "Tanh": 1, "Transpose": 30, "Unsqueeze": 195, "Where": 14 }, "inputs": { "vision_embeds": [ 1, "vision_len", 512 ], "token_one_hot": [ 1, 1, 14630 ], "position_ids": [ 1, "input_len" ], "past_k0": [ 1, 2, "past_len", 64 ], "past_k1": [ 1, 2, "past_len", 64 ], "past_k2": [ 1, 2, "past_len", 64 ], "past_k3": [ 1, 2, "past_len", 64 ], "past_k4": [ 1, 2, "past_len", 64 ], "past_k5": [ 1, 2, "past_len", 64 ], "past_v0": [ 1, 2, "past_len", 64 ], "past_v1": [ 1, 2, "past_len", 64 ], "past_v2": [ 1, 2, "past_len", 64 ], "past_v3": [ 1, 2, "past_len", 64 ], "past_v4": [ 1, 2, "past_len", 64 ], "past_v5": [ 1, 2, "past_len", 64 ] }, "outputs": { "logits": [ "Mullogits_dim_0", "Mullogits_dim_1", 14630 ], "present_k0": [ 1, 2, "total_len", 64 ], "present_k1": [ 1, 2, "total_len", 64 ], "present_k2": [ 1, 2, "total_len", 64 ], "present_k3": [ 1, 2, "total_len", 64 ], "present_k4": [ 1, 2, "total_len", 64 ], "present_k5": [ 1, 2, "total_len", 64 ], "present_v0": [ 1, 2, "total_len", 64 ], "present_v1": [ 1, 2, "total_len", 64 ], "present_v2": [ 1, 2, "total_len", 64 ], "present_v3": [ 1, 2, "total_len", 64 ], "present_v4": [ 1, 2, "total_len", 64 ], "present_v5": [ 1, 2, "total_len", 64 ] } }, "parity": { "prefill_zero_past": { "max_abs_by_output": [ 1.33514404296875e-05, 2.682209014892578e-07, 4.76837158203125e-07, 9.834766387939453e-07, 1.5497207641601562e-06, 1.1920928955078125e-06, 1.1920928955078125e-06, 8.940696716308594e-08, 1.4901161193847656e-07, 1.4901161193847656e-07, 2.086162567138672e-07, 1.4901161193847656e-07, 2.60770320892334e-07 ], "max_abs": 1.33514404296875e-05, "top_token": 14151 }, "step_empty_vision": { "max_abs_by_output": [ 3.4809112548828125e-05, 1.3113021850585938e-06, 4.76837158203125e-07, 1.430511474609375e-06, 1.430511474609375e-06, 1.9073486328125e-06, 2.1457672119140625e-06, 2.980232238769531e-07, 1.341104507446289e-07, 2.039596438407898e-07, 3.5762786865234375e-07, 1.9371509552001953e-07, 2.551823854446411e-07 ], "max_abs": 3.4809112548828125e-05, "top_token": 2 } } }, "qdq_int8": { "graph": { "bytes": 34647920, "sha256": "7b78d0124b396f8bff136bbf5ac198c8912b61986b69bed0a9d717f75b2b251c", "opsets": { "ai.onnx": 17 }, "operators": { "Add": 57, "Cast": 76, "Concat": 74, "Constant": 541, "ConstantOfShape": 13, "DequantizeLinear": 44, "Div": 38, "Equal": 13, "Expand": 13, "Gather": 77, "LessOrEqual": 1, "MatMul": 56, "Mul": 149, "Neg": 12, "Range": 2, "ReduceMean": 25, "Reshape": 49, "Shape": 88, "Sigmoid": 6, "Slice": 31, "Softmax": 6, "Sqrt": 25, "Sub": 1, "Tanh": 1, "Transpose": 30, "Unsqueeze": 195, "Where": 14 }, "inputs": { "vision_embeds": [ 1, "vision_len", 512 ], "token_one_hot": [ 1, 1, 14630 ], "position_ids": [ 1, "input_len" ], "past_k0": [ 1, 2, "past_len", 64 ], "past_k1": [ 1, 2, "past_len", 64 ], "past_k2": [ 1, 2, "past_len", 64 ], "past_k3": [ 1, 2, "past_len", 64 ], "past_k4": [ 1, 2, "past_len", 64 ], "past_k5": [ 1, 2, "past_len", 64 ], "past_v0": [ 1, 2, "past_len", 64 ], "past_v1": [ 1, 2, "past_len", 64 ], "past_v2": [ 1, 2, "past_len", 64 ], "past_v3": [ 1, 2, "past_len", 64 ], "past_v4": [ 1, 2, "past_len", 64 ], "past_v5": [ 1, 2, "past_len", 64 ] }, "outputs": { "logits": [ "Mullogits_dim_0", "Mullogits_dim_1", 14630 ], "present_k0": [ 1, 2, "total_len", 64 ], "present_k1": [ 1, 2, "total_len", 64 ], "present_k2": [ 1, 2, "total_len", 64 ], "present_k3": [ 1, 2, "total_len", 64 ], "present_k4": [ 1, 2, "total_len", 64 ], "present_k5": [ 1, 2, "total_len", 64 ], "present_v0": [ 1, 2, "total_len", 64 ], "present_v1": [ 1, 2, "total_len", 64 ], "present_v2": [ 1, 2, "total_len", 64 ], "present_v3": [ 1, 2, "total_len", 64 ], "present_v4": [ 1, 2, "total_len", 64 ], "present_v5": [ 1, 2, "total_len", 64 ] } }, "quantization": { "source": "decoder_unified_fp32.onnx", "bytes": 34647920, "sha256": "7b78d0124b396f8bff136bbf5ac198c8912b61986b69bed0a9d717f75b2b251c", "quantized_matmuls": 44, "quantized_gathers": 0, "quantized_elements": 33069056, "operators": { "Add": 57, "Cast": 76, "Concat": 74, "Constant": 541, "ConstantOfShape": 13, "DequantizeLinear": 44, "Div": 38, "Equal": 13, "Expand": 13, "Gather": 77, "LessOrEqual": 1, "MatMul": 56, "Mul": 149, "Neg": 12, "Range": 2, "ReduceMean": 25, "Reshape": 49, "Shape": 88, "Sigmoid": 6, "Slice": 31, "Softmax": 6, "Sqrt": 25, "Sub": 1, "Tanh": 1, "Transpose": 30, "Unsqueeze": 195, "Where": 14 }, "forbidden_webgpu_operators": [] }, "parity": { "prefill_zero_past": { "max_abs_by_output": [ 0.3767728805541992, 0.00906449556350708, 0.010979562997817993, 0.027931302785873413, 0.029697895050048828, 0.02388906478881836, 0.051448822021484375, 0.003086686134338379, 0.0026207156479358673, 0.004904627799987793, 0.009277760982513428, 0.005867749452590942, 0.007260143756866455 ], "max_abs": 0.3767728805541992, "top_token": 14151 }, "step_empty_vision": { "max_abs_by_output": [ 0.3307156562805176, 0.004899594932794571, 0.0077037513256073, 0.022474922239780426, 0.025185763835906982, 0.016876131296157837, 0.02703934907913208, 0.001852184534072876, 0.0019404292106628418, 0.005747482180595398, 0.004208110272884369, 0.003765057772397995, 0.006519559770822525 ], "max_abs": 0.3307156562805176, "top_token": 2 } } } }