序言

之前已经介绍了在超算互联网平台SCNet上使用异构加速卡AI 显存64GB PCIE，私有化部署Llama3模型，并对 Llama3-8B-Instruct 模型进行 LoRA 微调、推理和合并 ，详细内容请参考另一篇博客：快速体验LLaMA-Factory 私有化部署和高效微调Llama3模型（曙光超算互联网平台异构加速卡DCU）。

由于博主调试过程中遇到较多问题，本文记录FAQ相关问题，仅提供解决思路。

一、参考资料

曙光超算互联网平台SCNet之国产异构加速卡DCU

Llama3本地部署与高效微调入门

二、重要说明

当遇到包冲突时，通常使用 pip install --no-deps -e . 可解决绝大多数问题。

三、FAQ

Q：ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.1.0-git782048c.abi0.dtk2404.torch2.1. requires transformers==4.33.2, but you have transformers 4.43.3 which is incompatible.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
transformers 4.33.2 requires tokenizers!=0.11.3,<0.14,>=0.11.1, but you have tokenizers 0.15.0 which is incompatible.
vllm 0.3.3+git3380931.abi0.dtk2404.torch2.1 requires transformers>=4.38.0, but you have transformers 4.33.2 which is incompatible.

错误原因：错误一要求安装 transformers==4.33.2，安装该版本后，出现错误二。错误二要求安装 transformers>=4.38.0，与错误一相矛盾。

解决方法：解决该问题的思路，请参考下文的FAQ。

Q：pip._vendor.packaging.version.InvalidVersion: Invalid version: '0.1.0-git782048c.abi0.dtk2404.torch2.1.'

ERROR: Exception:
Traceback (most recent call last):File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/base_command.py", line 105, in _run_wrapperstatus = _inner_run()File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/base_command.py", line 96, in _inner_runreturn self.run(options, args)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/req_command.py", line 67, in wrapperreturn func(self, options, args)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/commands/install.py", line 483, in runinstalled_versions[distribution.canonical_name] = distribution.versionFile "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/metadata/pkg_resources.py", line 192, in versionreturn parse_version(self._dist.version)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_vendor/packaging/version.py", line 56, in parsereturn Version(version)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_vendor/packaging/version.py", line 202, in __init__raise InvalidVersion(f"Invalid version: '{version}'")
pip._vendor.packaging.version.InvalidVersion: Invalid version: '0.1.0-git782048c.abi0.dtk2404.torch2.1.'

(llama_factory_torch) root@notebook-1813389960667746306-scnlbe5oi5-50216:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory# pip install tokenizers==0.13
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Collecting tokenizers==0.13Downloading https://pypi.tuna.tsinghua.edu.cn/packages/cc/67/4c05eb8cbe8d20e52f5f47a9c591738d8cbc2a29e918813b7fcc431ec3db/tokenizers-0.13.0-cp310-cp310-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (7.0 MB)━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 7.0/7.0 MB 37.4 MB/s eta 0:00:00
WARNING: Error parsing dependencies of lmdeploy: Invalid version: '0.1.0-git782048c.abi0.dtk2404.torch2.1.'
WARNING: Error parsing dependencies of mmcv: Invalid version: '2.0.1-gitc0ccf15.abi0.dtk2404.torch2.1.'
Installing collected packages: tokenizersAttempting uninstall: tokenizersFound existing installation: tokenizers 0.15.0Uninstalling tokenizers-0.15.0:Successfully uninstalled tokenizers-0.15.0
ERROR: Exception:
Traceback (most recent call last):File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/base_command.py", line 105, in _run_wrapperstatus = _inner_run()File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/base_command.py", line 96, in _inner_runreturn self.run(options, args)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/cli/req_command.py", line 67, in wrapperreturn func(self, options, args)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/commands/install.py", line 483, in runinstalled_versions[distribution.canonical_name] = distribution.versionFile "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_internal/metadata/pkg_resources.py", line 192, in versionreturn parse_version(self._dist.version)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_vendor/packaging/version.py", line 56, in parsereturn Version(version)File "/opt/conda/envs/llama3/lib/python3.10/site-packages/pip/_vendor/packaging/version.py", line 202, in __init__raise InvalidVersion(f"Invalid version: '{version}'")
pip._vendor.packaging.version.InvalidVersion: Invalid version: '0.1.0-git782048c.abi0.dtk2404.torch2.1.'

错误原因：lmdeploy版本问题。

解决方法：解决该问题的思路，请参考下文的FAQ。

Q：版本匹配问题

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install -r requirements.txt
...
Installing collected packages: pydub, websockets, urllib3, tomlkit, shtab, semantic-version, scipy, ruff, importlib-resources, ffmpy, docstring-parser, aiofiles, tyro, sse-starlette, tokenizers, gradio-client, transformers, trl, peft, gradioAttempting uninstall: websocketsFound existing installation: websockets 12.0Uninstalling websockets-12.0:Successfully uninstalled websockets-12.0Attempting uninstall: urllib3Found existing installation: urllib3 1.26.13Uninstalling urllib3-1.26.13:Successfully uninstalled urllib3-1.26.13Attempting uninstall: tokenizersFound existing installation: tokenizers 0.15.0Uninstalling tokenizers-0.15.0:Successfully uninstalled tokenizers-0.15.0Attempting uninstall: transformersFound existing installation: transformers 4.38.0Uninstalling transformers-4.38.0:Successfully uninstalled transformers-4.38.0
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.1.0-git782048c.abi0.dtk2404.torch2.1. requires transformers==4.33.2, but you have transformers 4.43.3 which is incompatible.

错误原因：lmdeploy 0.1.0-git782048c.abi0.dtk2404.torch2.1. 与 transformers 版本冲突，要求 transformers==4.33.2。由于LLaMA-Factory项目要求 transformers>=4.41.2，因此选择升级 lmdeploy 以匹配 transformers 版本。

解决方法：在光合社区中查询并下载安装lmdeploy。以 lmdeploy-0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0-cp310-cp310-manylinux_2_31_x86_64.whl 为例，尝试安装 lmdeploy-0.2.6：

root@notebook-1813389960667746306-scnlbe5oi5-17811:~# pip list | grep lmdeploy
lmdeploy                       0.1.0-git782048c.abi0.dtk2404.torch2.1.
(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/cache# pip install  lmdeploy-0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0-cp310-cp310-manylinux_2_31_x86_64.whl
...
Installing collected packages: shortuuid, tokenizers, transformers, peft, lmdeployAttempting uninstall: tokenizersFound existing installation: tokenizers 0.19.1Uninstalling tokenizers-0.19.1:Successfully uninstalled tokenizers-0.19.1Attempting uninstall: transformersFound existing installation: transformers 4.43.3Uninstalling transformers-4.43.3:Successfully uninstalled transformers-4.43.3Attempting uninstall: peftFound existing installation: peft 0.12.0Uninstalling peft-0.12.0:Successfully uninstalled peft-0.12.0Attempting uninstall: lmdeployFound existing installation: lmdeploy 0.1.0-git782048c.abi0.dtk2404.torch2.1.Uninstalling lmdeploy-0.1.0-git782048c.abi0.dtk2404.torch2.1.:Successfully uninstalled lmdeploy-0.1.0-git782048c.abi0.dtk2404.torch2.1.
Successfully installed lmdeploy-0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0 peft-0.9.0 shortuuid-1.0.13 tokenizers-0.15.2 transformers-4.38.1

lmdeploy-0.2.6 安装成功，且没有报错，但是transformers版本降低为transformers-4.38.1。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py \
>     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct" \
>     --template llama3 \
>     --infer_backend vllm \
>     --vllm_enforce_eager
Traceback (most recent call last):File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/webui.py", line 17, in <module>from llamafactory.webui.interface import create_uiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/__init__.py", line 38, in <module>from .cli import VERSIONFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/cli.py", line 21, in <module>from . import launcherFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/launcher.py", line 15, in <module>from llamafactory.train.tuner import run_expFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/train/tuner.py", line 25, in <module>from ..hparams import get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/__init__.py", line 20, in <module>from .parser import get_eval_args, get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/parser.py", line 45, in <module>check_dependencies()File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/extras/misc.py", line 82, in check_dependenciesrequire_version("transformers>=4.41.2", "To fix: pip install transformers>=4.41.2")File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 111, in require_version_compare_versions(op, got_ver, want_ver, requirement, pkg, hint)File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 44, in _compare_versionsraise ImportError(
ImportError: transformers>=4.41.2 is required for a normal functioning of this module, but found transformers==4.38.1.
To fix: pip install transformers>=4.41.2

解决方法：升级 transformers，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/cache# pip install -U transformers
...
Installing collected packages: tokenizers, transformersAttempting uninstall: tokenizersFound existing installation: tokenizers 0.15.2Uninstalling tokenizers-0.15.2:Successfully uninstalled tokenizers-0.15.2Attempting uninstall: transformersFound existing installation: transformers 4.38.1Uninstalling transformers-4.38.1:Successfully uninstalled transformers-4.38.1
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0 requires transformers<=4.38.1,>=4.33.0, but you have transformers 4.43.3 which is incompatible.
Successfully installed tokenizers-0.19.1 transformers-4.43.3

错误原因：lmdeploy 0.2.6 与 transformers 版本冲突，要求 transformers<=4.38.1,>=4.33.0。由于LLaMA-Factory项目要求 transformers>=4.41.2，因此选择继续升级 lmdeploy 以匹配 transformers 版本。

解决方法：升级 lmdeploy：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/cache# pip install -U lmdeploy
...
Installing collected packages: nvidia-nccl-cu12, nvidia-curand-cu12, nvidia-cuda-runtime-cu12, nvidia-cublas-cu12, lmdeployAttempting uninstall: lmdeployFound existing installation: lmdeploy 0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0Uninstalling lmdeploy-0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0:Successfully uninstalled lmdeploy-0.2.6+das1.1.git6ba90df.abi1.dtk2404.torch2.1.0
Successfully installed lmdeploy-0.5.2.post1 nvidia-cublas-cu12-12.5.3.2 nvidia-cuda-runtime-cu12-12.5.82 nvidia-curand-cu12-10.3.6.82 nvidia-nccl-cu12-2.22.3

lmdeploy-0.5.2 安装成功，且没有报错。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct"     --template llama3     --infer_backend vllm     --vllm_enforce_eager
Traceback (most recent call last):File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/webui.py", line 17, in <module>from llamafactory.webui.interface import create_uiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/__init__.py", line 38, in <module>from .cli import VERSIONFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/cli.py", line 21, in <module>from . import launcherFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/launcher.py", line 15, in <module>from llamafactory.train.tuner import run_expFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/train/tuner.py", line 25, in <module>from ..hparams import get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/__init__.py", line 20, in <module>from .parser import get_eval_args, get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/parser.py", line 45, in <module>check_dependencies()File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/extras/misc.py", line 85, in check_dependenciesrequire_version("peft>=0.11.1", "To fix: pip install peft>=0.11.1")File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 111, in require_version_compare_versions(op, got_ver, want_ver, requirement, pkg, hint)File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 44, in _compare_versionsraise ImportError(
ImportError: peft>=0.11.1 is required for a normal functioning of this module, but found peft==0.9.0.
To fix: pip install peft>=0.11.1

解决方法：安装 peft==0.11.1：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/cache# pip install peft==0.11.1
...
Installing collected packages: peftAttempting uninstall: peftFound existing installation: peft 0.12.0Uninstalling peft-0.12.0:Successfully uninstalled peft-0.12.0
Successfully installed peft-0.11.1

peft==0.11.1 安装成功，且没有报错。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct"     --template llama3     --infer_backend vllm     --vllm_enforce_eager
[2024-07-31 15:23:04,562] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Traceback (most recent call last):File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/webui.py", line 17, in <module>from llamafactory.webui.interface import create_uiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/__init__.py", line 38, in <module>from .cli import VERSIONFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/cli.py", line 22, in <module>from .api.app import run_apiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/api/app.py", line 21, in <module>from ..chat import ChatModelFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/__init__.py", line 16, in <module>from .chat_model import ChatModelFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/chat_model.py", line 26, in <module>from .vllm_engine import VllmEngineFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/vllm_engine.py", line 37, in <module>from vllm.sequence import MultiModalData
ImportError: cannot import name 'MultiModalData' from 'vllm.sequence' (/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/sequence.py)

该问题的解决方法，请参考下文的FAQ。

Q：ImportError: cannot import name 'MultiModalData' from 'vllm.sequence'

最新取用的代码，运行api.py（或者是webui.py）报错，错误信息均是：ImportError: cannot import name ‘MultiModalData’ from ‘vllm.sequence’ (/usr/local/lib/python3.10/dist-packages/vllm/sequence.py) #3645

ImportError: cannot import name 'MultiModalData' from 'vllm.sequence'

错误原因：vllm版本过高或者版本过低，而LLaMA-Factory项目要求最低版本 vllm==0.4.3。

解决方法：以版本过高为例，将vllm版本从 vllm==0.5.0 降低到 vllm==0.4.3，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm==0.4.3
...
Installing collected packages: nvidia-ml-py, triton, nvidia-nvtx-cu12, nvidia-nvjitlink-cu12, nvidia-nccl-cu12, nvidia-curand-cu12, nvidia-cufft-cu12, nvidia-cuda-runtime-cu12, nvidia-cuda-nvrtc-cu12, nvidia-cuda-cupti-cu12, nvidia-cublas-cu12, llvmlite, lark, joblib, interegular, distro, diskcache, cmake, cloudpickle, nvidia-cusparse-cu12, nvidia-cudnn-cu12, numba, prometheus-fastapi-instrumentator, openai, nvidia-cusolver-cu12, lm-format-enforcer, torch, xformers, vllm-flash-attn, outlines, vllmAttempting uninstall: tritonFound existing installation: triton 2.1.0+git3841f975.abi0.dtk2404Uninstalling triton-2.1.0+git3841f975.abi0.dtk2404:Successfully uninstalled triton-2.1.0+git3841f975.abi0.dtk2404Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.22.3Uninstalling nvidia-nccl-cu12-2.22.3:Successfully uninstalled nvidia-nccl-cu12-2.22.3Attempting uninstall: nvidia-curand-cu12Found existing installation: nvidia-curand-cu12 10.3.6.82Uninstalling nvidia-curand-cu12-10.3.6.82:Successfully uninstalled nvidia-curand-cu12-10.3.6.82Attempting uninstall: nvidia-cuda-runtime-cu12Found existing installation: nvidia-cuda-runtime-cu12 12.5.82Uninstalling nvidia-cuda-runtime-cu12-12.5.82:Successfully uninstalled nvidia-cuda-runtime-cu12-12.5.82Attempting uninstall: nvidia-cublas-cu12Found existing installation: nvidia-cublas-cu12 12.5.3.2Uninstalling nvidia-cublas-cu12-12.5.3.2:Successfully uninstalled nvidia-cublas-cu12-12.5.3.2Attempting uninstall: torchFound existing installation: torch 2.1.0+git00661e0.abi0.dtk2404Uninstalling torch-2.1.0+git00661e0.abi0.dtk2404:Successfully uninstalled torch-2.1.0+git00661e0.abi0.dtk2404Attempting uninstall: xformersFound existing installation: xformers 0.0.25+gitd11e899.abi0.dtk2404.torch2.1Uninstalling xformers-0.0.25+gitd11e899.abi0.dtk2404.torch2.1:Successfully uninstalled xformers-0.0.25+gitd11e899.abi0.dtk2404.torch2.1Attempting uninstall: vllmFound existing installation: vllm 0.3.3+git3380931.abi0.dtk2404.torch2.1Uninstalling vllm-0.3.3+git3380931.abi0.dtk2404.torch2.1:Successfully uninstalled vllm-0.3.3+git3380931.abi0.dtk2404.torch2.1
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.5.2.post1 requires torch<=2.2.2,>=2.0.0, but you have torch 2.3.0 which is incompatible.
lmdeploy 0.5.2.post1 requires triton<=2.2.0,>=2.1.0; sys_platform == "linux", but you have triton 2.3.0 which is incompatible.
Successfully installed cloudpickle-3.0.0 cmake-3.30.1 diskcache-5.6.3 distro-1.9.0 interegular-0.3.3 joblib-1.4.2 lark-1.1.9 llvmlite-0.43.0 lm-format-enforcer-0.10.1 numba-0.60.0 nvidia-cublas-cu12-12.1.3.1 nvidia-cuda-cupti-cu12-12.1.105 nvidia-cuda-nvrtc-cu12-12.1.105 nvidia-cuda-runtime-cu12-12.1.105 nvidia-cudnn-cu12-8.9.2.26 nvidia-cufft-cu12-11.0.2.54 nvidia-curand-cu12-10.3.2.106 nvidia-cusolver-cu12-11.4.5.107 nvidia-cusparse-cu12-12.1.0.106 nvidia-ml-py-12.555.43 nvidia-nccl-cu12-2.20.5 nvidia-nvjitlink-cu12-12.5.82 nvidia-nvtx-cu12-12.1.105 openai-1.37.1 outlines-0.0.34 prometheus-fastapi-instrumentator-7.0.0 torch-2.3.0 triton-2.3.0 vllm-0.4.3 vllm-flash-attn-2.5.8.post2 xformers-0.0.26.post1

解决方法：将torch版本从 torch 2.3.0 降低到 torch 2.1.0，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install torch==2.1.0
...
Installing collected packages: triton, nvidia-nccl-cu12, torchAttempting uninstall: tritonFound existing installation: triton 2.3.0Uninstalling triton-2.3.0:Successfully uninstalled triton-2.3.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.20.5Uninstalling nvidia-nccl-cu12-2.20.5:Successfully uninstalled nvidia-nccl-cu12-2.20.5Attempting uninstall: torchFound existing installation: torch 2.3.0Uninstalling torch-2.3.0:Successfully uninstalled torch-2.3.0
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
vllm 0.4.3 requires torch==2.3.0, but you have torch 2.1.0 which is incompatible.
vllm-flash-attn 2.5.8.post2 requires torch==2.3.0, but you have torch 2.1.0 which is incompatible.
xformers 0.0.26.post1 requires torch==2.3.0, but you have torch 2.1.0 which is incompatible.
Successfully installed nvidia-nccl-cu12-2.18.1 torch-2.1.0 triton-2.1.0

解决方法：将vllm版本从 vllm 0.4.3 降低到 vllm 0.4.2，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm==0.4.2
...
Installing collected packages: vllm-nccl-cu12, triton, nvidia-nccl-cu12, tiktoken, torch, lm-format-enforcer, vllmAttempting uninstall: tritonFound existing installation: triton 2.1.0Uninstalling triton-2.1.0:Successfully uninstalled triton-2.1.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.18.1Uninstalling nvidia-nccl-cu12-2.18.1:Successfully uninstalled nvidia-nccl-cu12-2.18.1Attempting uninstall: tiktokenFound existing installation: tiktoken 0.7.0Uninstalling tiktoken-0.7.0:Successfully uninstalled tiktoken-0.7.0Attempting uninstall: torchFound existing installation: torch 2.1.0Uninstalling torch-2.1.0:Successfully uninstalled torch-2.1.0Attempting uninstall: lm-format-enforcerFound existing installation: lm-format-enforcer 0.10.1Uninstalling lm-format-enforcer-0.10.1:Successfully uninstalled lm-format-enforcer-0.10.1Attempting uninstall: vllmFound existing installation: vllm 0.4.3Uninstalling vllm-0.4.3:Successfully uninstalled vllm-0.4.3
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.5.2.post1 requires torch<=2.2.2,>=2.0.0, but you have torch 2.3.0 which is incompatible.
lmdeploy 0.5.2.post1 requires triton<=2.2.0,>=2.1.0; sys_platform == "linux", but you have triton 2.3.0 which is incompatible.
Successfully installed lm-format-enforcer-0.9.8 nvidia-nccl-cu12-2.20.5 tiktoken-0.6.0 torch-2.3.0 triton-2.3.0 vllm-0.4.2 vllm-nccl-cu12-2.18.1.0.4.0

解决方法：将vllm版本从 vllm 0.4.2 降低到 vllm 0.4.1，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm==0.4.1
...
Installing collected packages: triton, nvidia-nccl-cu12, torch, xformers, vllmAttempting uninstall: tritonFound existing installation: triton 2.3.0Uninstalling triton-2.3.0:Successfully uninstalled triton-2.3.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.20.5Uninstalling nvidia-nccl-cu12-2.20.5:Successfully uninstalled nvidia-nccl-cu12-2.20.5Attempting uninstall: torchFound existing installation: torch 2.3.0Uninstalling torch-2.3.0:Successfully uninstalled torch-2.3.0Attempting uninstall: xformersFound existing installation: xformers 0.0.26.post1Uninstalling xformers-0.0.26.post1:Successfully uninstalled xformers-0.0.26.post1Attempting uninstall: vllmFound existing installation: vllm 0.4.2Uninstalling vllm-0.4.2:Successfully uninstalled vllm-0.4.2
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
vllm-flash-attn 2.5.8.post2 requires torch==2.3.0, but you have torch 2.2.1 which is incompatible.
Successfully installed nvidia-nccl-cu12-2.19.3 torch-2.2.1 triton-2.2.0 vllm-0.4.1 xformers-0.0.25

解决方法：将vllm版本从 vllm-flash-attn 2.5.8.post2 降低到 vllm-flash-attn-2.5.6，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm-flash-attn==2.5.6
...
Installing collected packages: triton, nvidia-nccl-cu12, torch, vllm-flash-attnAttempting uninstall: tritonFound existing installation: triton 2.2.0Uninstalling triton-2.2.0:Successfully uninstalled triton-2.2.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.19.3Uninstalling nvidia-nccl-cu12-2.19.3:Successfully uninstalled nvidia-nccl-cu12-2.19.3Attempting uninstall: torchFound existing installation: torch 2.2.1Uninstalling torch-2.2.1:Successfully uninstalled torch-2.2.1Attempting uninstall: vllm-flash-attnFound existing installation: vllm-flash-attn 2.5.8.post2Uninstalling vllm-flash-attn-2.5.8.post2:Successfully uninstalled vllm-flash-attn-2.5.8.post2
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
vllm 0.4.1 requires torch==2.2.1, but you have torch 2.1.2 which is incompatible.
xformers 0.0.25 requires torch==2.2.1, but you have torch 2.1.2 which is incompatible.
Successfully installed nvidia-nccl-cu12-2.18.1 torch-2.1.2 triton-2.1.0 vllm-flash-attn-2.5.6

解决方法：将vllm版本从 vllm 0.4.1 降低到 vllm 0.4.0：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm==0.4.0
...
Installing collected packages: xformers, vllmAttempting uninstall: xformersFound existing installation: xformers 0.0.25Uninstalling xformers-0.0.25:Successfully uninstalled xformers-0.0.25Attempting uninstall: vllmFound existing installation: vllm 0.4.1Uninstalling vllm-0.4.1:Successfully uninstalled vllm-0.4.1
Successfully installed vllm-0.4.0 xformers-0.0.23.post1

vllm 0.4.0 安装成功，且没有报错。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py \
>     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct" \
>     --template llama3 \
>     --infer_backend vllm \
>     --vllm_enforce_eager
No ROCm runtime is found, using ROCM_HOME='/opt/dtk'
/opt/conda/envs/llama_factory/lib/python3.10/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: 'libc10_hip.so: cannot open shared object file: No such file or directory'If you don't plan on using image functionality from `torchvision.io`, you can ignore this warning. Otherwise, there might be something wrong with your environment. Did you have `libjpeg` or `libpng` installed before building `torchvision` from source?warn(
[2024-07-31 15:52:48,647] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Traceback (most recent call last):File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/webui.py", line 17, in <module>from llamafactory.webui.interface import create_uiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/__init__.py", line 38, in <module>from .cli import VERSIONFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/cli.py", line 22, in <module>from .api.app import run_apiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/api/app.py", line 21, in <module>from ..chat import ChatModelFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/__init__.py", line 16, in <module>from .chat_model import ChatModelFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/chat_model.py", line 26, in <module>from .vllm_engine import VllmEngineFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/chat/vllm_engine.py", line 29, in <module>from vllm import AsyncEngineArgs, AsyncLLMEngine, RequestOutput, SamplingParamsFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/__init__.py", line 4, in <module>from vllm.engine.async_llm_engine import AsyncLLMEngineFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 12, in <module>from vllm.engine.llm_engine import LLMEngineFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 16, in <module>from vllm.model_executor.model_loader import get_architecture_class_nameFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/model_executor/model_loader.py", line 10, in <module>from vllm.model_executor.models.llava import LlavaForConditionalGenerationFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/model_executor/models/llava.py", line 11, in <module>from vllm.model_executor.layers.activation import get_act_fnFile "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/vllm/model_executor/layers/activation.py", line 9, in <module>from vllm._C import ops
ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

该问题的解决方法，请参考下文的FAQ。

Q：ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

查找 libcuda.so.1 文件：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# find / -name "libcuda.so.1"
find: '/proc/1/map_files': Operation not permitted
find: '/proc/13/map_files': Operation not permitted
find: '/proc/45/map_files': Operation not permitted
find: '/proc/116/map_files': Operation not permitted
find: '/proc/118/map_files': Operation not permitted
find: '/proc/120/map_files': Operation not permitted
find: '/proc/121/map_files': Operation not permitted
find: '/proc/5527/map_files': Operation not permitted
find: '/proc/5529/map_files': Operation not permitted
find: '/proc/5531/map_files': Operation not permitted
find: '/proc/6148/map_files': Operation not permitted
find: '/proc/24592/map_files': Operation not permitted
find: '/proc/24970/map_files': Operation not permitted
find: '/proc/24971/map_files': Operation not permitted

错误原因：没有找到该文件，猜测是vllm的版本问题。

解决方法：重新安装 llvm 0.4.3，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install vllm==0.4.3
...
Installing collected packages: triton, nvidia-nccl-cu12, torch, lm-format-enforcer, xformers, vllm-flash-attn, vllmAttempting uninstall: tritonFound existing installation: triton 2.1.0Uninstalling triton-2.1.0:Successfully uninstalled triton-2.1.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.18.1Uninstalling nvidia-nccl-cu12-2.18.1:Successfully uninstalled nvidia-nccl-cu12-2.18.1Attempting uninstall: torchFound existing installation: torch 2.1.2Uninstalling torch-2.1.2:Successfully uninstalled torch-2.1.2Attempting uninstall: lm-format-enforcerFound existing installation: lm-format-enforcer 0.9.8Uninstalling lm-format-enforcer-0.9.8:Successfully uninstalled lm-format-enforcer-0.9.8Attempting uninstall: xformersFound existing installation: xformers 0.0.23.post1Uninstalling xformers-0.0.23.post1:Successfully uninstalled xformers-0.0.23.post1Attempting uninstall: vllm-flash-attnFound existing installation: vllm-flash-attn 2.5.6Uninstalling vllm-flash-attn-2.5.6:Successfully uninstalled vllm-flash-attn-2.5.6Attempting uninstall: vllmFound existing installation: vllm 0.4.0Uninstalling vllm-0.4.0:Successfully uninstalled vllm-0.4.0
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.5.2.post1 requires torch<=2.2.2,>=2.0.0, but you have torch 2.3.0 which is incompatible.
lmdeploy 0.5.2.post1 requires triton<=2.2.0,>=2.1.0; sys_platform == "linux", but you have triton 2.3.0 which is incompatible.
Successfully installed lm-format-enforcer-0.10.1 nvidia-nccl-cu12-2.20.5 torch-2.3.0 triton-2.3.0 vllm-0.4.3 vllm-flash-attn-2.5.8.post2 xformers-0.0.26.post1

错误原因：lmdeploy 0.5.2.post1 与 torch 版本冲突，要求 torch<=2.2.2,>=2.0.0，而当前版本为torch 2.3.0。lmdeploy 0.5.2.post1 与 triton 版本冲突，要求 triton<=2.2.0,>=2.1.0，而当前版本为triton 2.3.0。理论上，应该升级lmdeploy 版本以匹配torch版本，但是lmdeploy已经是最新版本了。因此，尝试降低lmdeploy版本。

解决方法：将lmdeploy版本从 lmdeploy 0.5.2.post1 降低到 lmdeploy 0.5.0，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install lmdeploy==0.5.0
...
Installing collected packages: triton, nvidia-nccl-cu12, torch, lmdeployAttempting uninstall: tritonFound existing installation: triton 2.3.0Uninstalling triton-2.3.0:Successfully uninstalled triton-2.3.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.20.5Uninstalling nvidia-nccl-cu12-2.20.5:Successfully uninstalled nvidia-nccl-cu12-2.20.5Attempting uninstall: torchFound existing installation: torch 2.3.0Uninstalling torch-2.3.0:Successfully uninstalled torch-2.3.0Attempting uninstall: lmdeployFound existing installation: lmdeploy 0.5.2.post1Uninstalling lmdeploy-0.5.2.post1:Successfully uninstalled lmdeploy-0.5.2.post1
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
vllm 0.4.3 requires torch==2.3.0, but you have torch 2.2.2 which is incompatible.
vllm-flash-attn 2.5.8.post2 requires torch==2.3.0, but you have torch 2.2.2 which is incompatible.
xformers 0.0.26.post1 requires torch==2.3.0, but you have torch 2.2.2 which is incompatible.

解决方法：将torch版本从 torch 2.2.2 升级到 torch 2.3.0，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install torch==2.3.0
...
Installing collected packages: triton, nvidia-nccl-cu12, torchAttempting uninstall: tritonFound existing installation: triton 2.2.0Uninstalling triton-2.2.0:Successfully uninstalled triton-2.2.0Attempting uninstall: nvidia-nccl-cu12Found existing installation: nvidia-nccl-cu12 2.19.3Uninstalling nvidia-nccl-cu12-2.19.3:Successfully uninstalled nvidia-nccl-cu12-2.19.3Attempting uninstall: torchFound existing installation: torch 2.2.2Uninstalling torch-2.2.2:Successfully uninstalled torch-2.2.2
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lmdeploy 0.5.0 requires torch<=2.2.2,>=2.0.0, but you have torch 2.3.0 which is incompatible.
lmdeploy 0.5.0 requires triton<=2.2.0,>=2.1.0; sys_platform == "linux", but you have triton 2.3.0 which is incompatible.
Successfully installed nvidia-nccl-cu12-2.20.5 torch-2.3.0 triton-2.3.0

解决方法：把lmdeploy版本从 lmdeploy 0.5.0 升级到 lmdeploy 0.5.1。

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# pip install lmdeploy==0.5.1
...
Installing collected packages: lmdeployAttempting uninstall: lmdeployFound existing installation: lmdeploy 0.5.0Uninstalling lmdeploy-0.5.0:Successfully uninstalled lmdeploy-0.5.0
Successfully installed lmdeploy-0.5.1

lmdeploy-0.5.1 安装成功，且没有报错。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py \
>     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct" \
>     --template llama3 \
>     --infer_backend vllm \
>     --vllm_enforce_eager
Traceback (most recent call last):File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/webui.py", line 17, in <module>from llamafactory.webui.interface import create_uiFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/__init__.py", line 38, in <module>from .cli import VERSIONFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/cli.py", line 21, in <module>from . import launcherFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/launcher.py", line 15, in <module>from llamafactory.train.tuner import run_expFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/train/tuner.py", line 25, in <module>from ..hparams import get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/__init__.py", line 20, in <module>from .parser import get_eval_args, get_infer_args, get_train_argsFile "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/hparams/parser.py", line 45, in <module>check_dependencies()File "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/src/llamafactory/extras/misc.py", line 85, in check_dependenciesrequire_version("peft>=0.11.1", "To fix: pip install peft>=0.11.1")File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 111, in require_version_compare_versions(op, got_ver, want_ver, requirement, pkg, hint)File "/opt/conda/envs/llama_factory/lib/python3.10/site-packages/transformers/utils/versions.py", line 44, in _compare_versionsraise ImportError(
ImportError: peft>=0.11.1 is required for a normal functioning of this module, but found peft==0.9.0.
To fix: pip install peft>=0.11.1

解决方法：升级 peft==0.11.1。

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/cache# pip install peft==0.11.1
...
Installing collected packages: peftAttempting uninstall: peftFound existing installation: peft 0.9.0Uninstalling peft-0.9.0:Successfully uninstalled peft-0.9.0
Successfully installed peft-0.11.1

peft-0.11.1 安装成功，且没有报错。

重新启动服务，又出现新问题：

(llama_factory) root@notebook-1813389960667746306-scnlbe5oi5-17811:/public/home/scnlbe5oi5/Downloads/models/LLaMA-Facto
ry# python src/webui.py     --model_name_or_path "/public/home/scnlbe5oi5/Downloads/models/LLaMA-Factory/models/Meta-Llama-3-8B-Instruct"     --template llama3     --infer_backend vllm     --vllm_enforce_eager
No ROCm runtime is found, using ROCM_HOME='/opt/dtk'
/opt/conda/envs/llama_factory/lib/python3.10/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: 'libc10_hip.so: cannot open shared object file: No such file or directory'If you don't plan on using image functionality from `torchvision.io`, you can ignore this warning. Otherwise, there might be something wrong with your environment. Did you have `libjpeg` or `libpng` installed before building `torchvision` from source?warn(
[2024-07-31 16:58:35,443] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect)
gradio_share: False
Running on local URL:  http://127.0.0.1:7860Could not create share link. Missing file: /opt/conda/envs/llama_factory/lib/python3.10/site-packages/gradio/frpc_linux_amd64_v0.2.Please check your internet connection. This can happen if your antivirus software blocks the download of this file. You can install manually by following these steps:1. Download this file: https://cdn-media.huggingface.co/frpc-gradio-0.2/frpc_linux_amd64
2. Rename the downloaded file to: frpc_linux_amd64_v0.2
3. Move the file to this location: /opt/conda/envs/llama_factory/lib/python3.10/site-packages/gradio

该问题的解决方法，请参考下文的FAQ。

Q. Could not create share link. Missing file:/PATH/TO/gradio/frpc_linux_amd64_v0.2

【Gradio】Could not create share link

Could not create share link. Missing file: /opt/conda/envs/llama_factory_torch/lib/python3.11/site-packages/gradio/frpc_linux_amd64_v0.2. Please check your internet connection. This can happen if your antivirus software blocks the download of this file. You can install manually by following these steps: 1. Download this file: https://cdn-media.huggingface.co/frpc-gradio-0.2/frpc_linux_amd64
2. Rename the downloaded file to: frpc_linux_amd64_v0.2
3. Move the file to this location: /opt/conda/envs/llama_factory_torch/lib/python3.11/site-packages/gradio

# 解决方法
1. 下载 https://cdn-media.huggingface.co/frpc-gradio-0.2/frpc_linux_amd642. 重命名
mv frpc_linux_amd64 frpc_linux_amd64_v0.23. 移动到指定目录
cp frpc_linux_amd64_v0.2 /opt/conda/envs/llama_factory_torch/lib/python3.10/site-packages/gradio4. 修改权限
chmod +x /opt/conda/envs/llama_factory_torch/lib/python3.10/site-packages/gradio/frpc_linux_amd64_v0.2

Q. Could not create share link.. Please check your internet connection or our status page

Could not create share link. Please check your internet connection or our status page: https://status.gradio.app

解决方法：修改 frpc_linux_amd64_v0.2文件权限。

chmod +x /opt/conda/envs/llama_factory_torch/lib/python3.11/site-packages/gradio/frpc_linux_amd64_v0.2