910b上infer一直报错，ERR00007 PTA feature not supported

问题命令：
swift infer \ --adapters /cache/model/qwen8b-lora/sft-1120-1/v0-20251120-152355/checkpoint-80 \ --stream true \ --infer_backend vllm \ --vllm_max_model_len 8192 \ --temperature 0 \ --max_new_tokens 2048

报错：
RuntimeError: Cannot run aclop operators during NPU graph capture. Current working aclop is MatMul. If you need this call to be captured, please try to set torch.npu.config.allow_internal_format = False. If still fail, the operator needs aclnn implementation and please file an issue. Current npuStreamCaptureStatus: npuStreamCaptureStatusActive

具体报错：
npuStreamCaptureStatusActive
(EngineCore_DP0 pid=2011231) ERROR 11-20 20:48:03 [core.py:708] [ERROR] 2025-11-20-20:48:03 (PID:2011231, Device:0, RankID:-1) ERR00007 PTA feature not supported
(EngineCore_DP0 pid=2011231) Process EngineCore_DP0:
(EngineCore_DP0 pid=2011231) Traceback (most recent call last):
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/multiprocessing/process.py", line 314, in bootstrap
(EngineCore_DP0 pid=2011231) self.run()
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/multiprocessing/process.py", line 108, in run
(EngineCore_DP0 pid=2011231) self.target(*self.args, **self.kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 712, in run_engine_core
(EngineCore_DP0 pid=2011231) raise e
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 699, in run_engine_core
(EngineCore_DP0 pid=2011231) engine_core = EngineCoreProc(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 498, in init
(EngineCore_DP0 pid=2011231) super().init(vllm_config, executor_class, log_stats,
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 92, in init
(EngineCore_DP0 pid=2011231) self.initialize_kv_caches(vllm_config)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 207, in initialize_kv_caches
(EngineCore_DP0 pid=2011231) self.model_executor.initialize_from_config(kv_cache_configs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/executor/abstract.py", line 75, in initialize_from_config
(EngineCore_DP0 pid=2011231) self.collective_rpc("compile_or_warm_up_model")
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/executor/uniproc_executor.py", line 83, in collective_rpc
(EngineCore_DP0 pid=2011231) return [run_method(self.driver_worker, method, args, kwargs)]
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/utils/init.py", line 3122, in run_method
(EngineCore_DP0 pid=2011231) return func(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/worker_v1.py", line 329, in compile_or_warm_up_model
(EngineCore_DP0 pid=2011231) self.model_runner.capture_model()
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/model_runner_v1.py", line 3561, in capture_model
(EngineCore_DP0 pid=2011231) self.capture_model()
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/model_runner_v1.py", line 3499, in capture_model
(EngineCore_DP0 pid=2011231) self.capture_aclgraphs(
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/model_runner_v1.py", line 3473, in capture_aclgraphs
(EngineCore_DP0 pid=2011231) self.dummy_run(num_tokens,
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/utils/contextlib.py", line 116, in decorate_context
(EngineCore_DP0 pid=2011231) return func(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/model_runner_v1.py", line 2494, in dummy_run
(EngineCore_DP0 pid=2011231) hidden_states = self.generate_dummy_run_hidden_states(
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/worker/model_runner_v1.py", line 2315, in generate_dummy_run_hidden_states
(EngineCore_DP0 pid=2011231) hidden_states = self.model(input_ids=input_ids,
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in wrapped_call_impl
(EngineCore_DP0 pid=2011231) return self.call_impl(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in call_impl
(EngineCore_DP0 pid=2011231) return forward_call(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/model_executor/models/qwen3.py", line 323, in forward
(EngineCore_DP0 pid=2011231) hidden_states = self.model(input_ids, positions, intermediate_tensors,
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/compilation/decorators.py", line 317, in call
(EngineCore_DP0 pid=2011231) model_output = self.forward(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/model_executor/models/qwen2.py", line 341, in forward
(EngineCore_DP0 pid=2011231) def forward(
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in wrapped_call_impl
(EngineCore_DP0 pid=2011231) return self.call_impl(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in call_impl
(EngineCore_DP0 pid=2011231) return forward_call(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/dynamo/eval_frame.py", line 838, in fn
(EngineCore_DP0 pid=2011231) return fn(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/fx/graph_module.py", line 830, in call_wrapped
(EngineCore_DP0 pid=2011231) return self.wrapped_call(self, *args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/fx/graph_module.py", line 406, in call
(EngineCore_DP0 pid=2011231) raise e
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/fx/graph_module.py", line 393, in call
(EngineCore_DP0 pid=2011231) return super(self.cls, obj).call(*args, **kwargs) # type: ignore[misc]
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in wrapped_call_impl
(EngineCore_DP0 pid=2011231) return self.call_impl(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in call_impl
(EngineCore_DP0 pid=2011231) return forward_call(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "<eval_with_key>.74", line 804, in forward
(EngineCore_DP0 pid=2011231) submod_0 = self.submod_0(l_input_ids, s0, l_self_modules_embed_tokens_parameters_weight, l_self_modules_layers_modules_0_modules_input_layernorm_parameters_weight, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_modules_base_layer_parameters_weight, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_punica_wrapper_seq_lengths, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_punica_wrapper_lora_indices_per_batch, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_0, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_1, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_2, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_0, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_1, l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_2, l_self_modules_layers_modules_0_modules_self_attn_modules_q_norm_parameters_weight, l_self_modules_layers_modules_0_modules_self_attn_modules_k_norm_parameters_weight, l_self_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache, l_positions, s1); l_input_ids = l_self_modules_embed_tokens_parameters_weight = l_self_modules_layers_modules_0_modules_input_layernorm_parameters_weight = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_modules_base_layer_parameters_weight = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_0 = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_1 = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_a_stacked_2 = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_0 = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_1 = l_self_modules_layers_modules_0_modules_self_attn_modules_qkv_proj_lora_b_stacked_2 = l_self_modules_layers_modules_0_modules_self_attn_modules_q_norm_parameters_weight = l_self_modules_layers_modules_0_modules_self_attn_modules_k_norm_parameters_weight = l_self_modules_layers_modules_0_modules_self_attn_modules_rotary_emb_buffers_cos_sin_cache = l_positions_ = None
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm_ascend/compilation/acl_graph.py", line 155, in call
(EngineCore_DP0 pid=2011231) output = self.runnable(*args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/compilation/cuda_piecewise_backend.py", line 96, in call
(EngineCore_DP0 pid=2011231) return self.compiled_graph_for_general_shape(*args)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/fx/graph_module.py", line 830, in call_wrapped
(EngineCore_DP0 pid=2011231) return self._wrapped_call(self, *args, **kwargs)
(EngineCore_DP0 pid=2011231) File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/torch/fx/graph_module.py", line 404, in call
(EngineCore_DP0 pid=2011231) raise e.with_traceback(None) # noqa: B904
(EngineCore_DP0 pid=2011231) RuntimeError: Cannot run aclop operators during NPU graph capture. Current working aclop is MatMul. If you need this call to be captured, please try to set torch.npu.config.allow_internal_format = False. If still fail, the operator needs aclnn implementation and please file an issue. Current npuStreamCaptureStatus: npuStreamCaptureStatusActive
(EngineCore_DP0 pid=2011231) [ERROR] 2025-11-20-20:48:03 (PID:2011231, Device:0, RankID:-1) ERR00007 PTA feature not supported
Traceback (most recent call last):
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/cli/infer.py", line 5, in
infer_main()
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/llm/infer/infer.py", line 301, in infer_main
return SwiftInfer(args).main()
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/llm/infer/infer.py", line 39, in init
self.infer_engine = self.get_infer_engine(args, self.template)
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/llm/infer/infer.py", line 86, in get_infer_engine
return infer_engine_cls(**kwargs)
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/llm/infer/infer_engine/vllm_engine.py", line 140, in init
self._prepare_engine()
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/swift/llm/infer/infer_engine/vllm_engine.py", line 150, in _prepare_engine
engine = llm_engine_cls.from_engine_args(self.engine_args)
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/llm_engine.py", line 177, in from_engine_args
return cls(vllm_config=vllm_config,
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/llm_engine.py", line 114, in init
self.engine_core = EngineCoreClient.make_client(
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core_client.py", line 80, in make_client
return SyncMPClient(vllm_config, executor_class, log_stats)
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core_client.py", line 602, in init
super().init(
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/core_client.py", line 448, in init
with launch_core_engines(vllm_config, executor_class,
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/contextlib.py", line 142, in exit
next(self.gen)
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/utils.py", line 732, in launch_core_engines
wait_for_engine_startup(
File "/home/ma-user/anaconda3/envs/vllm0.11.0/lib/python3.10/site-packages/vllm/v1/engine/utils.py", line 785, in wait_for_engine_startup
raise RuntimeError("Engine core initialization failed. "
RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}
[ERROR] 2025-11-20-20:48:14 (PID:2011056, Device:-1, RankID:-1) ERR99999 UNKNOWN applicaiton exceptio`

Your hardware and system info
CANN版本8.2
vllm版本0.11.0
torch-npu版本2.7.1
ms_swift版本3.10.1

Reproduce the supplied swift infer command and begin with the traceback locations in vllm_ascend/worker/model_runner_v1.py, especially capture_model, _capture_aclgraphs, and _dummy_run. Compare the NPU graph-capture path with the Qwen3 call in vllm/model_executor/models/qwen3.py; done means inference no longer fails with ERR00007 PTA feature not supported.