Add convenience function to load models from Hugging Face.

[favyen2]

Note

Adds a utility to build OlmoEarth models from Hugging Face config and optionally load weights.

• Python / Pretrain:
  • New module olmoearth_pretrain/model_loader.py:
    • Introduces ModelID enum with OLMOEARTH_V1_NANO, OLMOEARTH_V1_TINY, OLMOEARTH_V1_BASE.
    • Adds load_model(model_id, load_weights=True) to download config.json, build the model via Config.from_dict(...).build(), and optionally fetch weights.pth and load_state_dict from Hugging Face.

^{Written by Cursor Bugbot for commit d53cbc3. This will update automatically on new commits. Configure here.}

[cursor]

Bug: Model Loading Ignores load_weights Parameter

The load_model function unconditionally downloads the .pth weights file from Hugging Face, even when load_weights=False. This contradicts the docstring, which states that setting load_weights=False should skip downloading the weights, leading to unnecessary bandwidth usage and increased load times.

 

[Hgherzog]

LGTM