Theo

Results 9 comments of Theo

Same here. Did you add the option?

Sounds good, here is a PR: https://github.com/rom1504/img2dataset/pull/397 Great work on this library btw!

Actually this metrics uses the IDF so it requires computing across the whole dataset at once

Thank you @jordiclive! `peft==0.3.0` fixes the issue with `lora-llama-13b` and `openlm-research/open_llama_13b`. I still have the issue with `llama-7b` though. `decapoda-research/llama-7b-hf` gives `ValueError: Tokenizer class LLaMATokenizer does not exist or is...

I've tried following the updated environment you provided ``` bitsandbytes==0.41.0 deepspeed==0.10.0 peft==0.4.0 transformers==4.31.0 flash-attn==2.0.0.post1 ``` but still hit the same issue

With the following deepspeed command ``` deepspeed trainer_sft.py --configs llama-7b webgpt_dataset_only --deepspeed ``` I get an OOM error on a 40GB A100 (even with batch size 1 and sequence length...

@aliberts I still get ``` FileNotFoundError: train_config.json not found on the HuggingFace Hub in lerobot/pi0 ``` Do you have a working inference script for Pi0 in sim? cc @Cadene this...

@panjiacheng @FerryHuang @xsank Have you found a fix? * I'm using `vllm==0.8.2` with `V1` and still see this issue * I'm sure the inputs don't contain any extra `""` *...