Mehdi comments

Repositories
Issues
Comments

Results 3 comments of


                                            Mehdi

requirements.txt refers to both nltk==3.5 and nltk==3.8.

Added PR https://github.com/amazon-science/mm-cot/pull/12

Inference with "FULL_STATE_DICT" checkpoint from FSDP fine tuning

You may also check [this](https://github.com/meta-llama/llama-recipes/issues/711) issue.

Merging lora adapter with Llama 3.2 vision

I was able to merge it without any issue. Here is my code that might help: ```python from peft import PeftModel base_model = MllamaForConditionalGeneration.from_pretrained( model_id, torch_dtype=torch.bfloat16, device_map="auto", ) processor =...