Unsloth无法检测CUDA及str2optimizer32bit错误的解决求助

环境配置 #

• 硬件：HP Z4工作站，Intel Core i7处理器，NVIDIA 1080Ti GPU
• 系统：Ubuntu 22.04
• 软件版本：CUDA 12.1，PyTorch 2.3.0，Libnccl 2.18.3

已执行操作 #

源码编译bitsandbytes #

git clone https://github.com/TimDettmers/bitsandbytes.git && cd bitsandbytes/
pip install -r requirements-dev.txt
cmake -DCOMPUTE_BACKEND=cuda -S .
make
pip install .

Unsloth环境搭建 #

conda create --name unsloth_env python=3.10
conda activate unsloth_env
conda install pytorch-cuda=12.1 pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers
pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps trl peft accelerate bitsandbytes

Bash环境变量配置（.bashrc） #

export BNB_CUDA_VERSION=121
# 添加CUDA至PATH和LD_LIBRARY_PATH
export PATH=/usr/local/cuda-12.1/bin${PATH:+:$PATH}
export LD_LIBRARY_PATH=/usr/local/cuda-12.1/lib64${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}

错误现象 #

Python站点包中仅存在libbitsandbytes_cuda121.so，无CPU版本。运行python -m bitsandbytes或Unsloth训练时均报错，核心错误如下：

1. CUDA库加载失败：找不到libbitsandbytes_cuda121_nocublaslt121.so，尝试加载CPU版本也失败
2. 运行时错误：NameError: name 'str2optimizer32bit' is not defined

完整错误日志：

WARNING: BNB_CUDA_VERSION=121 environment variable detected; loading libbitsandbytes_cuda121_nocublaslt121.so.
This can be used to load a bitsandbytes version that is different from the PyTorch CUDA version.
If this was unintended set the BNB_CUDA_VERSION variable to an empty string: export BNB_CUDA_VERSION=
If you use the manual override make sure the right libcudart.so is in your LD_LIBRARY_PATH
For example by adding the following to your .bashrc: export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path_to_cuda_dir/lib64

Could not find the bitsandbytes CUDA binary at PosixPath('/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cuda121_nocublaslt121.so')
Could not load bitsandbytes native library: /home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cpu.so: cannot open shared object file: No such file or directory
Traceback (most recent call last):
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/cextension.py", line 109, in <module>
    lib = get_native_library()
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/cextension.py", line 96, in get_native_library
    dll = ct.cdll.LoadLibrary(str(binary_path))
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/ctypes/__init__.py", line 452, in LoadLibrary
    return self._dlltype(name)
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/ctypes/__init__.py", line 374, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: /home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cpu.so: cannot open shared object file: No such file or directory

CUDA Setup failed despite CUDA being available. Please run the following command to get more information:

python -m bitsandbytes

Inspect the output of the command and see if you can locate CUDA libraries. You might need to add them
to your LD_LIBRARY_PATH. If you suspect a bug, please take the information from python -m bitsandbytes
and open an issue at: https://github.com/TimDettmers/bitsandbytes/issues

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++ BUG REPORT INFORMATION ++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++ OTHER +++++++++++++++++++++++++++
CUDA specs: CUDASpecs(highest_compute_capability=(6, 1), cuda_version_string='121', cuda_version_tuple=(12, 1))
PyTorch settings found: CUDA_VERSION=121, Highest Compute Capability: (6, 1).
WARNING: BNB_CUDA_VERSION=121 environment variable detected; loading libbitsandbytes_cuda121_nocublaslt121.so.
This can be used to load a bitsandbytes version that is different from the PyTorch CUDA version.
If this was unintended set the BNB_CUDA_VERSION variable to an empty string: export BNB_CUDA_VERSION=
If you use the manual override make sure the right libcudart.so is in your LD_LIBRARY_PATH
For example by adding the following to your .bashrc: export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path_to_cuda_dir/lib64

Library not found: /home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cuda121_nocublaslt121.so. Maybe you need to compile it from source?
If you compiled from source, try again with `make CUDA_VERSION=DETECTED_CUDA_VERSION`,
for example, `make CUDA_VERSION=113`.

The CUDA version for the compile might depend on your conda install, if using conda.
Inspect CUDA version via `conda list | grep cuda`.
To manually override the PyTorch CUDA version please see: https://github.com/TimDettmers/bitsandbytes/blob/main/docs/source/nonpytorchcuda.mdx
WARNING: Compute capability < 7.5 detected! Only slow 8-bit matmul is supported for your GPU!
If you run into issues with 8-bit matmul, you can try 4-bit quantization:
https://huggingface.co/blog/4bit-transformers-bitsandbytes
Found duplicate CUDA runtime files (see below).

We select the PyTorch default CUDA runtime, which is 12.1,
but this might mismatch with the CUDA version that is needed for bitsandbytes.
To override this behavior set the `BNB_CUDA_VERSION=<version string, e.g. 122>` environmental variable.

For example, if you want to use the CUDA version 122,
    BNB_CUDA_VERSION=122 python ...

OR set the environmental variable in your .bashrc:
    export BNB_CUDA_VERSION=122

In the case of a manual override, make sure you set LD_LIBRARY_PATH, e.g.
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.2,
* Found CUDA runtime at: /usr/local/cuda-12.1/lib64/libcudart.so
* Found CUDA runtime at: /usr/local/cuda-12.1/lib64/libcudart.so.12
* Found CUDA runtime at: /usr/local/cuda-12.1/lib64/libcudart.so.12.1.105
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++ DEBUG INFO END ++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Checking that the library is importable and CUDA is callable...
Couldn't load the bitsandbytes library, likely due to missing binaries.
Please ensure bitsandbytes is properly installed.

For source installations, compile the binaries with `cmake -DCOMPUTE_BACKEND=cuda -S .`.
See the documentation for more details if needed.

Trying a simple check anyway, but this will likely fail...
Traceback (most recent call last):
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/diagnostics/main.py", line 66, in main
    sanity_check()
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/diagnostics/main.py", line 40, in sanity_check
    adam.step()
  File "/home/llm/.local/lib/python3.10/site-packages/torch/optim/optimizer.py", line 391, in wrapper
    out = func(*args, **kwargs)
  File "/home/llm/.local/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/optim/optimizer.py", line 287, in step
    self.update_step(group, p, gindex, pindex)
  File "/home/llm/.local/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/optim/optimizer.py", line 496, in update_step
    F.optimizer_update_32bit(
  File "/home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/functional.py", line 1584, in optimizer_update_32bit
    optim_func = str2optimizer32bit[optimizer_name][0]
NameError: name 'str2optimizer32bit' is not defined
Above we output some debug information.
Please provide this info when creating an issue via https://github.com/TimDettmers/bitsandbytes/issues/new/choose
WARNING: Please be sure to sanitize sensitive info from the output before posting it.

解决方案 #

1. 修正bitsandbytes编译参数，生成匹配的CUDA库 #

编译时未指定CUDA版本，导致生成的库文件名与bitsandbytes期望的不匹配。重新编译：

cd bitsandbytes/
make clean
cmake -DCOMPUTE_BACKEND=cuda -DCUDA_VERSION=121 -S .
make CUDA_VERSION=121
pip install --force-reinstall .

编译完成后，若仍找不到目标库，可手动创建软链接：

cd /home/llm/miniconda3/envs/unsloth_env/lib/python3.10/site-packages/bitsandbytes/
ln -s libbitsandbytes_cuda121.so libbitsandbytes_cuda121_nocublaslt121.so

2. 清理冲突的环境变量 #

移除.bashrc中的BNB_CUDA_VERSION=121，让bitsandbytes自动匹配PyTorch的CUDA版本，避免强制加载不存在的库文件。更新后执行：

source ~/.bashrc

3. 修复str2optimizer32bit未定义错误 #

该错误是源码编译的兼容性问题，切换至兼容PyTorch 2.3.0的稳定分支重新编译：

cd bitsandbytes/
git checkout 0.42.0
make clean
cmake -DCOMPUTE_BACKEND=cuda -DCUDA_VERSION=121 -S .
make CUDA_VERSION=121
pip install --force-reinstall .

4. 验证环境是否正常 #

运行以下代码验证兼容性：

import torch
import bitsandbytes as bnb
print("CUDA available:", torch.cuda.is_available())
print("Bitsandbytes version:", bnb.__version__)
# 测试优化器
optimizer = bnb.optim.Adam8bit(torch.nn.Linear(10,10).cuda().parameters(), lr=1e-3)
optimizer.step()
print("Optimizer test passed")

内容的提问来源于stack exchange，提问作者cerebro007