[Bug] tp=4 tp=8 no response

zeroleavebaoyang commented 3 weeks ago

Checklist

[X] 1. I have searched related issues but cannot get the expected help.
[X] 2. The bug has not been fixed in the latest version.

Describe the bug

发现一个问题，在rtx4090 * 8 环境，针对qwen1.5-110b-awq设置--tp 8 或者 qwen2-72b-awq 设置--tp 4 都会卡死一直无响应，张量并行设置大了好像基本都会有这样的卡死情况。

Reproduction

CUDA_VISIBLE_DEVICES=0,1,2,3 lmdeploy serve api_server /home/nlp/pretrain_models/Qwen2-72B-Instruct-AWQ \ --model-name qwen \ --server-name 0.0.0.0 \ --server-port 23334 \ --tp 4 \ --cache-max-entry-count 0.1 \ --quant-policy 4 \ --model-format awq

Environment

sys.platform: linux
Python: 3.10.14 (main, May  6 2024, 19:42:50) [GCC 11.2.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3,4,5,6,7,8,9: NVIDIA GeForce RTX 4090
CUDA_HOME: /usr/local/cuda-11.8
NVCC: Cuda compilation tools, release 11.8, V11.8.89
GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.2.2+cu118
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 v3.3.2 (Git Hash 2dc95a2ad0841e29db8b22fbccaf3e5da7992b01)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - CUDA Runtime 11.8
  - NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_90,code=sm_90
  - CuDNN 8.7
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.8, CUDNN_VERSION=8.7.0, 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_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=2.2.2, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF,

TorchVision: 0.17.2+cu118
LMDeploy: 0.4.2+
transformers: 4.41.2
gradio: Not Found
fastapi: 0.111.0
pydantic: 2.7.3
triton: 2.2.0

Error traceback

No response

lvhan028 commented 3 weeks ago

可能和 #1750 遇到的是同一类问题。尝试下下面的方法，看看能不能解决问题

export NCCL_P2P_DISABLE=1

如果不能解决的话，麻烦在启动命令中加入 --log-level INFO，把日志贴上来吧。

zeroleavebaoyang commented 3 weeks ago

@

可能和 #1750 遇到的是同一类问题。尝试下下面的方法，看看能不能解决问题
export NCCL_P2P_DISABLE=1
如果不能解决的话，麻烦在启动命令中加入 --log-level INFO，把日志贴上来吧。

如图所示，加入了 export NCCL_P2P_DISABLE=1 之后也是一样，一直卡死，并且最后一张卡 100%

lvhan028 commented 3 weeks ago

I haven't reproduced this issue. My device is A100-80G(x8) Could you try the docker image openmmlab/lmdeploy:v0.4.2?

lvhan028 commented 3 weeks ago

我感觉得用 gdb 来debug问题所在。在 hang 住之后，开另一个窗口，执行下面的命令

gdb attach <pid> # pid 是服务进程 id，可以通过 nvidia-smi 查看
set logging on
thread apply all bt
# 按 c，会显示所有的堆栈信息，这些信息会写到日志 gdb.txt 中
set logging off
q

执行完上述操作后，会在当前工作目录产生一个 gdb.txt 文件，麻烦把这个文件传到issue中来吧。

CocaColaKing commented 2 weeks ago

我感觉得用 gdb 来debug问题所在。在 hang 住之后，开另一个窗口，执行下面的命令
gdb attach <pid> # pid 是服务进程 id，可以通过 nvidia-smi 查看
set logging on
thread apply all bt
# 按 c，会显示所有的堆栈信息，这些信息会写到日志 gdb.txt 中
set logging off
q
执行完上述操作后，会在当前工作目录产生一个 gdb.txt 文件，麻烦把这个文件传到issue中来吧。

gdb.txt

InternLM / lmdeploy