error when run finetune.py

[mazzzystar]

When I run python finetune.py, it will first shows error like this:

  File "/home/ubuntu/miniconda3/envs/lora/lib/python3.10/site-packages/bitsandbytes/functional.py", line 1505, in get_colrow_absmax
    lib.cget_col_row_stats(ptrA, ptrRowStats, ptrColStats, ptrNnzrows, ct.c_float(threshold), rows, cols)
  File "/home/ubuntu/miniconda3/envs/lora/lib/python3.10/ctypes/__init__.py", line 387, in __getattr__
    func = self.__getitem__(name)
  File "/home/ubuntu/miniconda3/envs/lora/lib/python3.10/ctypes/__init__.py", line 392, in __getitem__
    func = self._FuncPtr((name_or_ordinal, self))
AttributeError: /home/ubuntu/miniconda3/envs/lora/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cpu.so: undefined symbol: cget_col_row_stats

Then I tried as this: bitsandbytes-foundation/bitsandbytes#156 (comment)

The error then become:

Loading checkpoint shards:   0%|                                                                                              | 0/33 [00:00<?, ?it/s]
Error named symbol not found at line 508 in file /mmfs1/gscratch/zlab/timdettmers/git/bitsandbytes/csrc/ops.cu

I'm using Python 3.10 + CUDA 11.7.

[ParaNoth]

Which architecture graphics card are you using? If your graphics card's architecture is not Turing or Ampere, you should use libbitsandbytes_cuda{your_cuda_version}_nocublaslt.so instead of libbitsandbytes_cuda{your_cuda_version}.so

It means you should use this command:

cp libbitsandbytes_cuda117_nocublaslt.so libbitsandbytes_cpu.so

[mazzzystar]

@ParaNoth Thank you very much!

My graphics is NVIDIA V100, I figured out it was Volta architecture. I've changed to libbitsandbytes_cuda{cuda_version}_nocublaslt.so and it works!

[herobrine19]

It works same to me! Thank you!