Does the original model have to be in HF format?
#3
by
adivekar
- opened
Hi @akoksal , thanks for your work on LongForm! I really liked the models you've created, and the paper.
Quick question: to apply the weights and re-use the model, do we first have to convert LLaMa-7B using this HuggingFace script? https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/convert_llama_weights_to_hf.py
I feel like the answer is yes, but I would like to be sure.
I converted them with that script if I remember correctly.
Just FYI: You can also use huggyllama/llama-7b as the original_llama_path. It creates the same model in my case.