search

openlm-research / open_llama

OpenLLaMA, a permissively licensed open source reproduction of Meta AI’s LLaMA 7B trained on the RedPajama dataset
Apache License 2.0
7.27k stars 370 forks source link

openlm-research/open_llama_13b_easylm cannot be downloaded #66

Closed leweex95 closed 1 year ago

leweex95 commented 1 year ago

I attempted to download the aforementioned model from: https://huggingface.co/openlm-research/open_llama_13b_easylm but it fails with the following error:

OSError: openlm-research/open_llama_13b_easylm does not appear to have a file named config.json. Checkout 'https://huggingface.co/openlm-research/open_llama_13b_easylm/main' for available files.

Code to reproduce:

model_id = "openlm-research/open_llama_13b_easylm"
model_name = model_id.split("/")[1]

model = pipeline(model=model_id)
model.save_pretrained(f"./models/{model_name}")

I expect the model to be downloadable locally for use in downstream NLP tasks. Noteworthy is that on the website it is indicated that 0 people downloaded this model over the past month.

With the above script, I can easily fetch other models such as "openlm-research/open_llama_13b":

model_id = "openlm-research/open_llama_13b"
model_name = model_id.split("/")[1]

model = pipeline(model=model_id)
model.save_pretrained(f"./models/{model_name}")

Looking at: https://huggingface.co/openlm-research/open_llama_13b_easylm/tree/main it seems that the file doesn't have a torch bin file. However, https://huggingface.co/openlm-research/open_llama_13b/tree/main has those .bin files.

Could you please tell me if this behavior is by design and if so, how I can fetch the model from HuggingFace?

young-geng commented 1 year ago

open_llama_13b_easylm is intended to be used with our own JAX based EasyLM framework. It is not compatible with the transformers pytorch implementation. To use the pytorch version, you need to use open_llama_13b. This is documented in the readme of this project.