[Bug] NPU下Qwen2.5-VL推理报错

[tcye]

Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.

Describe the bug

NPU下，Qwen2.5-VL推理报错，是还有算子未支持吗？但看dlinfer显示已支持Qwen2.5-VL

Reproduction

lmdeploy serve api_server /mnt/cephfs/tiancaiye/r1_model/Qwen2.5-VL-72B-Reasoning/ --tp 8 --server-port 8080 --device ascend --backend pytorch --eager-mode --dtype bfloat16

随后使用OpenAI client对话

Environment

sys.platform: linux
Python: 3.10.15 (main, Nov 27 2024, 06:37:16) [GCC 11.4.0]
CUDA available: False
MUSA available: False
numpy_random_seed: 2147483648
GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.3.1+cpu
PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2022.2-Product Build 20220804 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v3.3.6 (Git Hash 86e6af5974177e513fd3fee58425e1063e7f1361)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=2.3.1, USE_CUDA=0, USE_CUDNN=OFF, USE_CUSPARSELT=OFF, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=OFF, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, 

TorchVision: 0.18.1+cpu
LMDeploy: 0.7.1+
transformers: 4.49.0
gradio: 5.20.0
fastapi: 0.115.11
pydantic: 2.10.6
triton: Not Found

Error traceback

2025-03-10 08:45:28,436 - lmdeploy - ERROR - model_agent.py:391 - Task <ModelAgentLoop> failed
Traceback (most recent call last):
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 386, in _on_finish_callback
    task.result()
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 374, in _async_loop_background
    await self._async_step_background(
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 322, in _async_step_background
    output = await self._async_model_forward(inputs,
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 243, in _async_model_forward
    ret = await __forward(inputs)
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 220, in __forward
    return await self.async_forward(inputs, swap_in_map=swap_in_map, swap_out_map=swap_out_map)
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 538, in async_forward
    output = self._forward_impl(inputs, swap_in_map=swap_in_map, swap_out_map=swap_out_map)
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 521, in _forward_impl
    output = model_forward(
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/engine/model_agent.py", line 75, in model_forward
    output = model(**input_dict)
  File "/opt/lmdeploy/lmdeploy/pytorch/backends/graph_runner.py", line 24, in __call__
    return self.model(**kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/models/qwen2_5_vl.py", line 432, in forward
    image_embeds = self.visual(pixel_values,
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/models/qwen2_5_vl.py", line 359, in forward
    hidden_states = blk(hidden_states, cu_seqlens=cu_seqlens_now, rotary_pos_emb=rotary_pos_emb)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/models/qwen2_5_vl.py", line 195, in forward
    hidden_states = hidden_states + self.attn(
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/models/qwen2_5_vl.py", line 136, in forward
    attn_output = self.proj(attn_output)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "/opt/lmdeploy/lmdeploy/pytorch/nn/linear.py", line 1094, in forward
    return self.impl.forward(x, self.weight, self.bias, all_reduce)
  File "/opt/lmdeploy/lmdeploy/pytorch/backends/dlinfer/linear.py", line 23, in forward
    return linear(x, weight, bias, all_reduce)
  File "/opt/lmdeploy/lmdeploy/pytorch/kernels/dlinfer/linear.py", line 9, in linear
    return ext_ops.linear(x, weight, bias=bias, all_reduce=all_reduce)
  File "/opt/dlinfer/dlinfer/graph/custom_op.py", line 70, in patched_func
    return func_with_default(*args, **kwargs)
  File "/opt/dlinfer/dlinfer/ops/llm.py", line 619, in linear
    return vendor_ops_registry["linear"](x, weight, bias, all_reduce)
  File "/opt/dlinfer/dlinfer/vendor/ascend/torch_npu_ops.py", line 496, in linear
    out = torch.ops.npu.npu_mm_all_reduce_base(
  File "/usr/local/python3.10/lib/python3.10/site-packages/torch/_ops.py", line 854, in __call__
    return self_._op(*args, **(kwargs or {}))
RuntimeError: call aclnnMatmulAllReduce failed, detail:EZ1001: [PID: 13141] 2025-03-10-08:45:28.432.661 x1 not implemented for DT_FLOAT, should be in dtype support list [DT_FLOAT16,DT_BFLOAT16,].
        TraceBack (most recent call last):
        Cannot find bin of op FlashAttentionScore, integral key 0/1/|float/ND/float/ND/bf16/ND/float/ND/float/ND/float/ND/float/ND/.
        Cannot find binary for op FlashAttentionScore.
        Kernel GetWorkspace failed. opType: 39
        x1 not implemented for DT_FLOAT, should be in dtype support list [DT_FLOAT16,DT_BFLOAT16,].