[Bug] 用lite命令auto_awq量化的QwQ-32B，推理时在单节点多卡机子上指定tp=2报错，在ascend和cuda环境下一样的报错。

[zzhaowendao]

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

用lite命令量化的QwQ-32B成功，在单节点多卡机子上进行推理时指定tp=2报错。在ascend和cuda环境下都试了，一样的报错。

Reproduction

量化命令：
lmdeploy lite auto_awq QwQ-32B --device npu(或者cuda)

量化后推理命令：
lmdeploy serve api_server qwq-32b-w4a16 --server-port 23334 --cache-max-entry-count 0.05 --quant-policy 4 --log-level DEBUG --backend pytorch --device ascend(或者cuda) --tp 2

Environment

sys.platform: linux
Python: 3.9.21 (main, Dec 11 2024, 16:27:47) [GCC 11.2.0]
CUDA available: False
MUSA available: False
numpy_random_seed: 2147483648
GCC: gcc (GCC) 7.3.0
PyTorch: 2.3.1
PyTorch compiling details: PyTorch built with:
  - GCC 10.2
  - C++ Version: 201703
  - 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: NO AVX
  - Build settings: BLAS_INFO=open, BUILD_TYPE=Release, CXX_COMPILER=/opt/rh/devtoolset-10/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-ver
sion=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DLIBKINETO_NOROCTRACER -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-stric
t-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdia
gnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-ov
erflow, LAPACK_INFO=open, TORCH_VERSION=2.3.1, USE_CUDA=OFF, USE_CUDNN=OFF, USE_CUSPARSELT=OFF, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=
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
LMDeploy: 0.7.2.post1+dac197c
transformers: 4.50.0
gradio: Not Found
fastapi: 0.115.12
pydantic: 2.10.6
triton: Not Found

Error traceback

RuntimeError: The expanded size of the tensor (5120) must match the existing size (2560) at non-singleton dimension 0.  Target sizes: [5120].  Tensor sizes:
 [2560]

[jinminxi104]

我印象里面做量化不需要tp。看看tp=1？

[zzhaowendao]

量化目前不支持tp(希望后续能够增加这个功能，毕竟对于很大的模型如ds六百多B，多卡进行量化是非常有必要的)。
推理可以指定tp。
又测试了几个例子，总结如下：

推理命令大致这样：lmdeploy serve api_server qwq-32b-w4a16 --server-port 23334 --cache-max-entry-count 0.05 --quant-policy 4 --log-level DEBUG --backend pytorch --device ascend(或者cuda) --tp 2

cuda环境：
lite auto_awq w4a16量化：
推理运行结果：

1. tp=1正常不报错。
2. tp=2报错RuntimeError: The expanded size of the tensor (5120) must match the existing size (2560) at non-singleton dimension 0. Target sizes: [5120]. Tensor sizes: [2560]

lite smooth_quant w8a8量化：
推理运行结果：
tp=1和tp=2都正常。

ascend环境：
lite auto_awq w4a16量化：
推理运行结果：

1. tp=1报错AttributeError: 'MergedAwqLinear' object has no attribute 'is_tp'.
2. tp=2报错RuntimeError: The expanded size of the tensor (5120) must match the existing size (2560) at non-singleton dimension 0. Target sizes: [5120]. Tensor sizes: [2560]

lite smooth_quant w8a8量化：直接报错，无法量化成功。
但是如果用在cuda环境下lite smooth_quant w8a8量化成功的模型，进行推理，tp=1和tp=2都正常。

[jinminxi104]

目前ascend只支持awq w4a16的 eager下的推理。
smooth_quant w8a8在ascend上的图模式正在pr中

[zzhaowendao]

我测试的都是eager模式，还没开始测图模式。

麻烦问下上面的各种情况下的报错可以解决吗？

[lvhan028]

cc @AllentDan

[jinminxi104]

我测试的都是eager模式，还没开始测图模式。

麻烦问下上面的各种情况下的报错可以解决吗？

tp1的那个错误我们会修复一下

[zzhaowendao]

好的，麻烦了。

tp=2这个cuda和ascend都报一样的错，这个看起来是框架整体的问题，后续可以修复吗？