转换qwen模型的时候，提示atten_mask:5 error.

[louwangzhiyuY]

odeling_qwen.py:1297: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! self._seq_len_cached = max(2 * seqlen, 16) C:\Users\ML.cache\huggingface\modules\transformers_modules\Qwen\Qwen-7B-Chat\218aa3240fd5a5d1e80bb6c47d5d774361913706\modeling_qwen.py:482: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! if key_size > self.seq_length and self.use_logn_attn and not self.training: C:\Users\ML.cache\huggingface\modules\transformers_modules\Qwen\Qwen-7B-Chat\218aa3240fd5a5d1e80bb6c47d5d774361913706\modeling_qwen.py:502: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! if query.size(1) == key_size: In-place op on output of tensor.shape. See https://pytorch.org/docs/master/onnx.html#avoid-inplace-operations-when-using-tensor-shape-in-tracing-mode Traceback (most recent call last): File "c:\work\AI\directml\export_llama_to_onnx\export_qwen_naive.py", line 173, in export_qwen(args) File "c:\work\AI\directml\export_llama_to_onnx\export_qwen_naive.py", line 151, in export_qwen export_qwen_to_single_onnx(model, config, dtype, args, "qwen_onnx") File "c:\work\AI\directml\export_llama_to_onnx\export_qwen_naive.py", line 109, in export_qwen_to_single_onnx torch.onnx.export( File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\utils.py", line 516, in export _export( File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\utils.py", line 1596, in _export graph, params_dict, torch_out = _model_to_graph( ^^^^^^^^^^^^^^^^ File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\utils.py", line 1139, in _model_to_graph graph = _optimize_graph( ^^^^^^^^^^^^^^^^ File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\utils.py", line 677, in _optimize_graph graph = _C._jit_pass_onnx(graph, operator_export_type) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\utils.py", line 1940, in _run_symbolic_function return symbolic_fn(graph_context, *inputs, **attrs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\symbolic_helper.py", line 306, in wrapper return fn(g, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^ File "c:\work\AI\directml\export_llama_to_onnx\export.ven\Lib\site-packages\torch\onnx\symbolic_opset14.py", line 197, in scaled_dot_product_attention raise ValueError( ValueError: Unsupported type for attn_mask: 5

[alphaRGB]

遇到相同的转换错误

[BeStrongok]

已解决， 将 attention_mask = torch.ones([batch, sumN], dtype=torch.int64).to(args.device) 改为 attention_mask = torch.ones([batch, sumN], dtype=torch.bool).to(args.device) 同时使用如下命令运行脚本， python3 export_qwen_naive.py -m qwen -o qwen.onnx -d cpu -p float32