meta-llama
Please help: failed to create process
Hi,
Apologies if the solution is obvious but I'm new to this. When running the example infilling script:
torchrun --nproc_per_node 1 example_infilling.py --ckpt_dir CodeLlama-7b/ --tokenizer_path CodeLlama-7b/tokenizer.model --max_seq_len 192 --max_batch_size 4
I get the following error with no additional details:
failed to create process
In fact anything with torchrun returns the same error.
I tried:
- Different versions of CUDA and CPU-only Pytorch
- Checked that
tokenizer_pathis correct andnproc_per_nodeis set to the right MP value - A comment on another post suggested using
python -m torch.distributed.runinstead oftorchrun. I get a different error when I do this (happy to give more info)
Any help would be greatly appreciated!
I'm facing the same issue. Tried with python 3.8 and 3.7. I'm just an enthusiastic person trying to play with an AI , I have no experience with Python but I would love a tip to solve this. Trying researching and even ChatGPT but no success so far.
I created my Conda env with the following conda create -conda create --name llama-code-p37 python=3.8 pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c conda-forge. Am I missing something?
Sto affrontando lo stesso problema. Provato con Python 3.8 e 3.7. Sono solo una persona entusiasta che cerca di giocare con un'intelligenza artificiale, non ho esperienza con Python ma mi piacerebbe un suggerimento per risolvere questo problema. Ho provato a fare ricerche e persino a ChatGPT, ma finora senza successo.
Ho creato il mio Conda env con il seguente file
conda create -conda create --name llama-code-p37 python=3.8 pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c conda-forge. Mi sto perdendo qualcosa?
Serve python3.10 io uso python3.12 deve essere superiore a pyton10 prova a vedere se risolvi il problema
Solved from stackoverflow! you can try
https://stackoverflow.com/questions/77425569/llama2-running-pytorch-produces-a-failed-to-create-process
