Masaki Kozuki
Masaki Kozuki
thank you for your clarification. I misunderstood some bits. could you try commenting out https://github.com/NVIDIA/apex/blob/05091d498d21058a0fe736b828c43431d4f0dda2/setup.py#L130 ? since your install command doesn't depend on any custom extensions, I don't think the...
@xiawu out of curiosity, which features would you use after CPU-only installation?
> log: [9_docker (pytorch-1.9.0torchvision-0.10.0cuda-10.2ffmpeg-4.2.txt](https://github.com/NVIDIA/apex/files/11556737/9_docker.pytorch-1.9.0torchvision-0.10.0cuda-10.2ffmpeg-4.2.txt) would this mean pytorch 1.9 is installed in your environment? if so, the pytorch itself is a bit too old to compile apex
> torch.__version__ = 1.5.1 the installed pytorch looks a bit too old
> Even with `torch 1.13.1`, the apex failed to install. with the same error message? where is cuda installed in your environment? could you try setting the environment variable of...
> ```python > /home/celso/projects/venvs/LightXML/lib/python3.8/site-packages/torch/include/torch/csrc/python_headers.h:12:10: fatal error: Python.h: No such file or directory > 12 | #include > | ^~~~~~~~~~ > compilation terminated. > error: command 'x86_64-linux-gnu-gcc' failed with exit status...
LayerNorm vs RMSNorm [GptNeoxMLP](https://github.com/Lightning-AI/litgpt/blob/9a1b82b66d29a08e6682e3c6095ef76f25ee3245/litgpt/model.py#L346) vs [LlamaMLP](https://github.com/Lightning-AI/litgpt/blob/9a1b82b66d29a08e6682e3c6095ef76f25ee3245/litgpt/model.py#L360) fwiw, the mlp used in stablecode is not benchmarked as per #742 ``` # stablecode-completion-alpha-3b GPT( (lm_head): Linear(in_features=2560, out_features=49152, bias=False) (transformer): ModuleDict( (wte): Embedding(49152,...
This also seems to boil down to `parallel_residual` If I manually change turn `parallel_residual` off, then the memory consumption of fsdp2 & ThunderCompiler for stablecode-completion-alpha-3b is smaller than `thunder.distributed.fsdp`, and...
would this repo somehow happen to disallow [`--allow-empty`](https://git-scm.com/docs/git-commit/2.31.0#Documentation/git-commit.txt---allow-empty)?
wouldn't `python setup.py install --fast_multihead_attn` work?