Make download stats work

[NielsRogge]

Hi @mini-omni,

Thanks for this nice work! Niels here from HF.

I noticed you already use the 🤗 hub for loading the model, which is great!

This PR aims to improve the integration by:

• adding from_pretrained and push_to_hub capabilities to the GPT model
• making sure download numbers work for your model (similar to models in the Transformers library)

It leverages the PyTorchModelHubMixin class which allows to inherits these methods.

Usage is as follows:

from litgpt.model import GPT

# define model
model = GPT(...)

# equip with weights
model.load_state_dict(...)

# push to hub
model.push_to_hub("gpt-omni/mini-omni")

# reload
model = GPT.from_pretrained("gpt-omni/mini-omni")

This means people don't need to manually download a checkpoint first in their local environment, it just loads automatically from the hub.

Would you be interested in this integration?

Kind regards,

Niels