cuda版本冲突问题

原创已于 2026-05-08 15:28:33 修改 · 97 阅读

本内容遵循CC 4.0 BY-SA版权协议

于 2026-04-28 10:44:33 首次发布

1. cuda版本与nvidia驱动版本冲突问题

#python test_env.py
PyTorch version: 2.11.0+cu130
/envs/audio/lib/python3.10/site-packages/torch/cuda/init.py:180: UserWarning: CUDA initialization: The NVIDIA driver on your system is too old (found version 12040). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver. (Triggered internally at /pytorch/c10/cuda/CUDAFunctions.cpp:119.)
return torch._C._cuda_getDeviceCount() > 0
CUDA available: False
Traceback (most recent call last):
File "/data/test_env.py", line 4, in <module>
print("Current device:", torch.cuda.current_device())
File "/envs/audio/lib/python3.10/site-packages/torch/cuda/init.py", line 1148, in current_device
_lazy_init()
File "/envs/audio/lib/python3.10/site-packages/torch/cuda/init.py", line 478, in _lazy_init
torch._C._cuda_init()
RuntimeError: The NVIDIA driver on your system is too old (found version 12040). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver.
(audio) root@online-dev-code-server-147-869f69f576-dl4dw:/data/# nvidia-smi
Tue Apr 28 02:19:15 2026

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.154.05             Driver Version: 535.154.05   CUDA Version: 12.4     |

# test_env.py
import torch
print("PyTorch version:", torch.__version__)
print("CUDA available:", torch.cuda.is_available())
print("Current device:", torch.cuda.current_device())
print("Device name:", torch.cuda.get_device_name(0))

1.1 分析问题

你遇到的错误是因为 PyTorch 2.11.0 编译时使用了较新的 CUDA 版本（比如 CUDA 13.0），而你的 NVIDIA 驱动版本是 535.154.05，只支持到 CUDA 12.4，导致 PyTorch 无法正确初始化 CUDA。 🔍 错误信息总结： PyTorch version: 2.11.0+cu130 （编译时使用的 CUDA 是 13.0） NVIDIA driver version: 535.154.05 ，对应支持的 CUDA 最高为 12.4 Error message: The NVIDIA driver on your system is too old (found version 12040). Please update your GPU driver..

1.2 解决问题

✅ 正确的安装命令（推荐）

你可以安装如下组合：
pip install torch==2.0.0+cu121 torchvision==0.15.1+cu121 torchaudio==2.0.1 --extra-index-url https://download.pytorch.org/whl/cu121 -i https://pypi.tuna.tsinghua.edu.cn/simple

或者如果你可以接受稍微更新一点的版本（比如 2.1.x），那可以这样安装：
pip install torch==2.1.0+cu121 torchvision==0.16.0+cu121 torchaudio==2.1.0 --extra-index-url https://download.pytorch.org/whl/cu121 -i https://pypi.tuna.tsinghua.edu.cn/simple

(audio) root@# python test_env.py 
PyTorch version: 2.1.0+cu121
CUDA available: True
Current device: 0
Device name: NVIDIA L20

2. torch版本太低问题

#test_model.py
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat")
tokenizer = AutoTokenizer.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat")

print(model)
print(tokenizer)

(audio) root# python test_model.py
[transformers] Disabling PyTorch because PyTorch >= 2.4 is required but found 2.1.0+cu121
[transformers] PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
Traceback (most recent call last):
File "/data/test_model.py", line 4, in <module>
model = AutoModelForCausalLM.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat")
File "/dataenvs/audio/lib/python3.10/site-packages/transformers/utils/import_utils.py", line 2040, in getattribute
requires_backends(cls, cls._backends)
File "/dataenvs/audio/lib/python3.10/site-packages/transformers/utils/import_utils.py", line 2026, in requires_backends
raise ImportError("".join(failed))
ImportError:
AutoModelForCausalLM requires the PyTorch library but it was not found in your environment. Check out the instructions on the
installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
Please note that you may need to restart your runtime after installation.

2.1 分析问题

你遇到的错误是因为： AutoModelForCausalLM 需要 PyTorch >= 2.4，但你的环境中安装的是 torch==2.1.0+cu121

✅ 错误信息总结 [transformers] Disabling PyTorch because PyTorch >= 2.4 is required but found 2.1.0+cu121

这意味着你当前使用的 PyTorch 版本太旧，无法加载模型。

2.2 解决问题

去官网

https://pytorch.org/get-started/previous-versions

找到2.5.1版本

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124 -i https://pypi.tuna.tsinghua.edu.cn/simple

# test_env.py
import torch, torchvision, torchaudio
print("PyTorch version:", torch.__version__)
print("torchvision:", torchvision.__version__)
print("torchaudio:", torchaudio.__version__)
print("CUDA available:", torch.cuda.is_available())
print("Current device:", torch.cuda.current_device())
print("Device name:", torch.cuda.get_device_name(0))

# python test_env.py 
PyTorch version: 2.5.1+cu124
torchvision: 0.20.1+cu124
torchaudio: 2.5.1+cu124
CUDA available: True
Current device: 0
Device name: NVIDIA L20

3. transformer版本太低问题

# python test_model.py

The repository /data/model_llm/Qwen/Qwen-Audio-Chat contains custom code which must be executed to correctly load the model. You can inspect the repository content at /data/model_llm/Qwen/Qwen-Audio-Chat .
You can inspect the repository content at https:/hf.co/data/model_llm/Qwen/Qwen-Audio-Chat.
You can avoid this prompt in future by passing the argument 
trust_remote_code=True
.

Do you wish to run the custom code? [y/N] y
The repository /data/model_llm/Qwen/Qwen-Audio-Chat contains custom code which must be executed to correctly load the model. You can inspect the repository content at /data/model_llm/Qwen/Qwen-Audio-Chat .
You can inspect the repository content at https:/hf.co/data/model_llm/Qwen/Qwen-Audio-Chat.
You can avoid this prompt in future by passing the argument 
trust_remote_code=True
.

Do you wish to run the custom code? [y/N] y
[transformers] Encountered exception while importing transformers_stream_generator: cannot import name 'DisjunctiveConstraint' from 'transformers' (/data/envs/audio/lib/python3.10/site-packages/transformers/init.py)
Traceback (most recent call last):
File "/data/test_model.py", line 4, in <module>
model = AutoModelForCausalLM.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat")
File "/data/envs/audio/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 369, in from_pretrained
model_class = get_class_from_dynamic_module(
File "/data/envs/audio/lib/python3.10/site-packages/transformers/dynamic_module_utils.py", line 572, in get_class_from_dynamic_module
final_module = get_cached_module_file(
File "/data/envs/audio/lib/python3.10/site-packages/transformers/dynamic_module_utils.py", line 410, in get_cached_module_file
modules_needed = check_imports(resolved_module_file)
File "/data/envs/audio/lib/python3.10/site-packages/transformers/dynamic_module_utils.py", line 246, in check_imports
importlib.import_module(imp)
File "/data/envs/audio/lib/python3.10/importlib/init.py", line 126, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "<frozen importlib._bootstrap>", line 1050, in _gcd_import
File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
File "<frozen importlib._bootstrap_external>", line 883, in exec_module
File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
File "/data/envs/audio/lib/python3.10/site-packages/transformers_stream_generator/init.py", line 1, in <module>
from .main import init_stream_support
File "/data/envs/audio/lib/python3.10/site-packages/transformers_stream_generator/main.py", line 1, in <module>
from transformers import (
ImportError: cannot import name 'DisjunctiveConstraint' from 'transformers' (/data/envs/audio/lib/python3.10/site-packages/transformers/init.py)

3.1 分析问题

你遇到的错误是因为：

模型使用了 transformers_stream_generator 库，但该库依赖于 DisjunctiveConstraint 这个类，而它在你当前安装的 transformers 版本中不存在。

✅ 错误信息总结 ImportError: cannot import name 'DisjunctiveConstraint' from 'transformers'

说明你的 transformers 版本太旧，缺少 DisjunctiveConstraint 。该功能通常在 transformers >= 4.37.0 中引入。

3.2 解决问题

✅ 解决方法一：升级 transformers（推荐）
你需要将 transformers 升级到 至少 4.37.0 或更高版本。

🔧 安装命令
pip install transformers>=4.37.0

或者指定具体版本：
pip install transformers==4.39.0

还要安装

pip install matplotlib tiktoken -i https://pypi.tuna.tsinghua.edu.cn/simple

测试：

# test_transformer.py
from transformers import DisjunctiveConstraint
print("Imported DisjunctiveConstraint successfully.")

# python test_transformer.py 
Imported DisjunctiveConstraint successfully.

测试2：

# test_model.py
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# model = AutoModelForCausalLM.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat")
model = AutoModelForCausalLM.from_pretrained(
    "/data/model_llm/Qwen/Qwen-Audio-Chat",
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("/data/model_llm/Qwen/Qwen-Audio-Chat", trust_remote_code=True)

print(model)
print(tokenizer)

# python test_model.py 
The model is automatically converting to bf16 for faster inference. If you want to disable the automatic precision, please manually add bf16/fp16/fp32=True to "AutoModelForCausalLM.from_pretrained".
Try importing flash-attention for faster inference...
Warning: import flash_attn rotary fail, please install FlashAttention rotary to get higher efficiency https://github.com/Dao-AILab/flash-attention/tree/main/csrc/rotary
Warning: import flash_attn rms_norm fail, please install FlashAttention layer_norm to get higher efficiency https://github.com/Dao-AILab/flash-attention/tree/main/csrc/layer_norm
Warning: import flash_attn fail, please install FlashAttention to get higher efficiency https://github.com/Dao-AILab/flash-attention
Loading checkpoint shards: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:01<00:00,  5.25it/s]
audio_start_id: 155163, audio_end_id: 155164, audio_pad_id: 151851.
QWenLMHeadModel(
  (transformer): QWenModel(
    (wte): Embedding(155947, 4096)
    (drop): Dropout(p=0.0, inplace=False)
    (rotary_emb): RotaryEmbedding()
    (h): ModuleList(
      (0-31): 32 x QWenBlock(
        (ln_1): RMSNorm()
        (attn): QWenAttention(
          (c_attn): Linear(in_features=4096, out_features=12288, bias=True)
          (c_proj): Linear(in_features=4096, out_features=4096, bias=False)
          (attn_dropout): Dropout(p=0.0, inplace=False)
        )
        (ln_2): RMSNorm()
        (mlp): QWenMLP(
          (w1): Linear(in_features=4096, out_features=11008, bias=False)
          (w2): Linear(in_features=4096, out_features=11008, bias=False)
          (c_proj): Linear(in_features=11008, out_features=4096, bias=False)
        )
      )
    )
    (ln_f): RMSNorm()
    (audio): AudioEncoder(
      (conv1): Conv1d(80, 1280, kernel_size=(3,), stride=(1,), padding=(1,))
      (conv2): Conv1d(1280, 1280, kernel_size=(3,), stride=(2,), padding=(1,))
      (blocks): ModuleList(
        (0-31): 32 x ResidualAttentionBlock(
          (attn): MultiHeadAttention(
            (query): Linear(in_features=1280, out_features=1280, bias=True)
            (key): Linear(in_features=1280, out_features=1280, bias=False)
            (value): Linear(in_features=1280, out_features=1280, bias=True)
            (out): Linear(in_features=1280, out_features=1280, bias=True)
          )
          (attn_ln): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (0): Linear(in_features=1280, out_features=5120, bias=True)
            (1): GELU(approximate='none')
            (2): Linear(in_features=5120, out_features=1280, bias=True)
          )
          (mlp_ln): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
        )
      )
      (ln_post): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
      (avg_pooler): AvgPool1d(kernel_size=(2,), stride=(2,), padding=(0,))
      (proj): Linear(in_features=1280, out_features=4096, bias=True)
      (audio_bos_eos_token): Embedding(2, 4096)
    )
  )
  (lm_head): Linear(in_features=4096, out_features=155947, bias=False)
)
QWenTokenizer(name_or_path='/data/model_llm/Qwen/Qwen-Audio-Chat', vocab_size=155165, model_max_length=2048, is_fast=False, padding_side='right', truncation_side='right', special_tokens={}, clean_up_tokenization_spaces=True),  added_tokens_decoder={

}

标签

#人工智能 #python