esmfold_v1-safetensors

This repo exists solely for CI testing in huggingface/transformers. Do not use it for anything else.

Origin

This is a safetensors conversion of facebook/esmfold_v1. All model weights and non-model files are identical to the original; only the weight format has changed from pytorch .bin to safetensors.

Why safetensors?

The original repo only has pytorch_model.bin. On Nvidia bucket CI runners, /mnt/cache is mounted via Xet FUSE. When torch.load() accesses large .bin blobs from Xet FUSE, it fails with OSError: [Errno 5] Input/output error on multi-gpu runners. Safetensors weights do not trigger this issue.

Used in

• Test: EsmModelIntegrationTest::test_inference_protein_folding
• File: tests/models/esm/test_modeling_esmfold.py
• PR: https://github.com/huggingface/transformers/pull/49194