[Bug] llama3和Chinese-LLaMA-Alpaca-3 无论是pipeline方式，还是lmdeploy chat方式都无法生成结果（llama2都是好的）

[zhanghui-china]

Checklist

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

Describe the bug

start_chinese_llama3.py

import lmdeploy
from lmdeploy import pipeline
pipe = lmdeploy.pipeline("/home/zhanghui/models/ChineseAlpacaGroup/llama-3-chinese-8b-instruct", model_name='llama3')

response=pipe(['你是谁'])
print(response)

start_meta_llama3.py

from lmdeploy import pipeline

pipe = pipeline('/home/zhanghui/models/LLM-Research/Meta-Llama-3-8B-Instruct', model_name='llama3')

responese = pipe([
    'Who are you?',
    'Hello!'
])

print(responese)

start_chinese_llama2.py

import lmdeploy
from lmdeploy import pipeline
pipe = lmdeploy.pipeline("/mnt/d/models/chinese-alpaca-2-7b-hf", model_name='llama2')

response=pipe(['你是谁'])
print(response)

start_meta_llama2.py

import lmdeploy
from lmdeploy import pipeline
pipe = lmdeploy.pipeline("/mnt/d/models/meta-llama/Llama-2-7b-chat-hf", model_name='llama2')

response=pipe(['你是谁'])
print(response)

Reproduction

llama3代
python start_chinese_llama3.py

python start_meta_llama3.py

llama2代
python start_chinese_llama2.py

python start_meta_llama2.py

Environment

(lmdeploy0430) zhanghui@zhanghui:~/Chinese-LLaMA-Alpaca-3$  lmdeploy check_env
sys.platform: linux
Python: 3.10.14 (main, Mar 21 2024, 16:24:04) [GCC 11.2.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0: NVIDIA GeForce RTX 3080 Laptop GPU
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 11.6, V11.6.124
GCC: gcc (Ubuntu 9.5.0-1ubuntu1~22.04) 9.5.0
PyTorch: 2.1.2+cu121
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 v3.1.1 (Git Hash 64f6bcbcbab628e96f33a62c3e975f8535a7bde4)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - CUDA Runtime 12.1
  - 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_90,code=sm_90
  - CuDNN 8.9.7  (built against CUDA 11.8)
    - Built with CuDNN 8.9.2
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.9.2, 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 -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-invalid-partial-specialization -Wno-unused-private-field -Wno-aligned-allocation-unavailable -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_DISABLE_GPU_ASSERTS=ON, TORCH_VERSION=2.1.2, 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,

LMDeploy: 0.4.0+1e8d7b8
transformers: 4.38.2
gradio: Not Found
fastapi: 0.110.3
pydantic: 2.7.1
triton: 2.1.0

Error traceback

No response

[lvhan028]

Have you tried the official llama3-8b-instruct?

[zhanghui-china]

Have you tried the official llama3-8b-instruct?

我下载的应该是modelscope官方的。https://www.modelscope.cn/models/LLM-Research/Meta-Llama-3-8B-Instruct/summary

这个链接就是官方认证的。

[zhanghui-china]

有5万次下载量·呢！

[lvhan028]

I cannot reproduce your issue

[lvhan028]

Please pass log_level=INFO to the pipeline and run it again. Let's check what the log says

[zhanghui-china]

我从huggingface官网下载了 权重文件。

···
import os

os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com'

command_str = 'huggingface-cli download --token hf_Bxxx --resume-download meta-llama/Meta-Llama-3-8B-Instruct --local-dir ' + os.environ.get('HOME') + '/models/meta-llama/Meta-Llama-3-8B-Instruct1'

os.system(command_str)
···
下载结果如下：

进入lmdeploy 0.4.0环境
conda deactivate
conda activate lmdeploy040
lmdeploy chat /home/zhanghui/models/meta-llama/Meta-Llama-3-8B-Instruct1 --model-name llama3

[zhanghui-china]

修改 start_meta_llama3.py

from lmdeploy import pipeline

#pipe = pipeline('/home/zhanghui/models/LLM-Research/Meta-Llama-3-8B-Instruct', model_name='llama3')
pipe = pipeline('/home/zhanghui/models/meta-llama/Meta-Llama-3-8B-Instruct1', model_name='llama3')

responese = pipe([
    'Who are you?',
    'Hello!'
])

print(responese)

[zhanghui-china]

[zhanghui-china]

加了环境变量之后，好像也一样

[zhanghui-china]

I cannot reproduce your issue

我是在WSL环境的。个人怀疑是WSL环境中 lmdeploy转换有问题。

[lvhan028]

@irexyc may follow up this issue

[lvhan028]

@lzhangzz FYI

[liaoduoduo]

你好，我也遇到这个问题了，请问解决了吗

[lvhan028]

你好，我也遇到这个问题了，请问解决了吗

请问你也是用的wsl环境么？

[liaoduoduo]

你好，我也遇到这个问题了，请问解决了吗

请问你也是用的wsl环境么？

我用的是windows环境

[lvhan028]

裸机上吗？

[liaoduoduo]

裸机上吗？

是的，直接裸机跑，llama2是没问题的，4080显卡

[lvhan028]

那感觉就不是我们讨论的那个问题，可能还在别处。@irexyc @lzhangzz
能不能麻烦把 log-level 设置为 INFO，把日志结果贴出来？
我们手头没有 windows + 24G显卡的机器，debugging这个问题比较棘手

[liaoduoduo]

那感觉就不是我们讨论的那个问题，可能还在别处。@irexyc @lzhangzz 能不能麻烦把 log-level 设置为 INFO，把日志结果贴出来？ 我们手头没有 windows + 24G显卡的机器，debugging这个问题比较棘手
好的，以下是我的日志，推理输出结果为空

[liaoduoduo]

那感觉就不是我们讨论的那个问题，可能还在别处。@irexyc @lzhangzz 能不能麻烦把 log-level 设置为 INFO，把日志结果贴出来？ 我们手头没有 windows + 24G显卡的机器，debugging这个问题比较棘手

加载llama2模型却可正常推理

[liaoduoduo]

Environment
sys.platform: win32
Python: 3.10.14 | packaged by Anaconda, Inc. | (main, Mar 21 2024, 16:20:14) [MSC v.1916 64 bit (AMD64)]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0: NVIDIA GeForce RTX 4080
CUDA_HOME: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4
NVCC: Cuda compilation tools, release 12.4, V12.4.131
GCC: n/a
PyTorch: 2.2.2+cu121
PyTorch compiling details: PyTorch built with:

• C++ Version: 201703
• MSVC 192930151
• Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications
• Intel(R) MKL-DNN v3.3.2 (Git Hash 2dc95a2ad0841e29db8b22fbccaf3e5da7992b01)
• OpenMP 2019
• LAPACK is enabled (usually provided by MKL)
• CPU capability usage: AVX2
• CUDA Runtime 12.1
• NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-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_90,code=sm_90
• CuDNN 8.8.1 (built against CUDA 12.0)
• Magma 2.5.4
• Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.8.1, CXX_COMPILER=C:/actions-runner/_work/pytorch/pytorch/builder/windo
  ws/tmp_bin/sccache-cl.exe, CXX_FLAGS=/DWIN32 /D_WINDOWS /GR /EHsc /Zc:__cplusplus /bigobj /FS /utf-8 -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI
  -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE /wd4624 /wd4068 /wd4067 /wd4267 /wd4661 /wd4717 /wd4244 /wd4804 /wd4273, LAP
  ACK_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=OFF, USE_NNPACK=OFF, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF,

TorchVision: 0.17.2+cu121
LMDeploy: 0.4.1+unknown
transformers: 4.40.0
gradio: Not Found
fastapi: 0.110.3
pydantic: 2.7.1
triton: Not Found

[lvhan028]

那感觉就不是我们讨论的那个问题，可能还在别处。@irexyc @lzhangzz 能不能麻烦把 log-level 设置为 INFO，把日志结果贴出来？ 我们手头没有 windows + 24G显卡的机器，debugging这个问题比较棘手
好的，以下是我的日志，推理输出结果为空

这个日志中显示 invalid infer request，code是6. 表示 too long
日志是全的吗？应该还有一些关于block size，block count的日志。

[liaoduoduo]

那感觉就不是我们讨论的那个问题，可能还在别处。@irexyc @lzhangzz 能不能麻烦把 log-level 设置为 INFO，把日志结果贴出来？ 我们手头没有 windows + 24G显卡的机器，debugging这个问题比较棘手
好的，以下是我的日志，推理输出结果为空

这个日志中显示 invalid infer request，code是6. 表示 too long 日志是全的吗？应该还有一些关于block size，block count的日志。

补上刚刚的日志

应该是显存不足导致的，换了V100 32G显存进行测试，可以正常推理了

[lvhan028]

@zhanghui-china 麻烦看下是不是在你这边也是类似的问题？

[lvhan028]

@liaoduoduo 你的 4080 显卡，内存是多大的？

[liaoduoduo]

@lvhan028 是16GB的

[lvhan028]

16G是不够的。LMDeploy按理说要报 OOM 才对。
@irexyc 总感觉 lmdeploy 在 windows 上磕磕绊绊的，跟进下吧。