Matthew Douglas

Hi all, This seems to be an issue for [LangChain](https://github.com/langchain-ai/langchain), and not bitsandbytes.

@Wouter1 Since you've built with CUDA 12.4, try with `export BNB_CUDA_VERSION=124`. In this case it should look for the build library at `/tmp/bitsandbytes/bitsandbytes/libbitsandbytes_cuda124_nocublaslt.so` .

@NavinKumarMNK Given your GPU compute capability (7.0) you'll want to compile with `-DNO_CUBLASLT`. I'm curious to know if this works on ppc64le after that adjustment! @shigabeev It looks like you're...

Hi @Edenzzzz, Make sure that this chunk is contiguous as F.optimizer_update_32bit ultimately treats it as 1D. ```python dist_low_rank_grad = torch.load("dist_low_rank_grad.pt").contiguous() ``` I was able to reproduce your results, and after...

> I wonder if that's due to the c++ kernel not considering the reshaped strides and assuming row-major format? Yes, exactly. The C++ kernel assumes it's row-major and only knows...

Related: #1126 I agree; this could use some improvement.

Hi @bryanhughes, Despite the wheel's tag, `bitsandbytes-0.42.0-py3-none-any.whl`, it is only built for x86-64 platforms. There has not been an aarch64 release (yet). With that said, I'm personally interested in supporting...

@yurisalesc I'm not sure that's the same issue; as far as I know MX350 isn't found in aarch64 devices. Can you open a separate issue with more information? @bryanhughes It's...

Hi @Harsh-raj, What version of `bitsandbytes` are you using?

@jeetendraabvv Windows support requires `bitsandbytes >= 0.43.0`. Please try upgrading using `pip install -U bitsandbytes`.