# ESM-2 650M protein language model weights (esm2_t33_650M_UR50D)

## Facts

- Provider: Meta AI (FAIR)
- License: MIT (https://github.com/facebookresearch/esm/blob/main/LICENSE)
- Size: 9 files
- Format: JSON, HDF5 (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/facebook/esm2_t33_650M_UR50D
- Access: public, readable without an account
- Category: Biology
- Source: https://www.science.org/doi/10.1126/science.ade2574
- Page: https://scigantic.com/archive/5c5569aa-65ba-5e25-9da3-d07fb16c9664
- Keywords: Biology, protein structure, protein language model, JSON, HDF5, Hugging Face

## About

Open weights for ESM-2 650M, the 33-layer, 650 million parameter protein language model from Meta's ESM project (Lin et al., Science 2023), trained with masked language modelling on UniRef50 protein sequences. It produces per-residue and per-sequence embeddings used for fitness prediction, contact prediction, function annotation and as the base model for many fine-tuned protein predictors. The mounted repo holds model.safetensors, pytorch_model.bin and tf_model.h5 (about 2.6 GB each), config.json, vocab.txt and tokenizer files, and loads with transformers (EsmModel, EsmForMaskedLM) from the mount path. Pair it with the UniRef, ProteinGym or PDB archives to embed sequences next to the data. MIT licence.

## Load the data

```python
import os, json
base = "/mnt/archive"   # this model repo, mounted read-only
files = sorted(os.listdir(base)); print(files)
print(json.dumps(json.load(open(os.path.join(base, "config.json"))), indent=1)[:800])
import struct
with open(os.path.join(base, "model.safetensors"), "rb") as f:
    hdr = json.loads(f.read(struct.unpack("<Q", f.read(8))[0]))
tensors = {k: v for k, v in hdr.items() if k != "__metadata__"}
print(len(tensors), "tensors, e.g.", list(tensors.items())[:3])
# To load them: pip install safetensors torch, then safetensors.torch.load_file(path)
```

Written for a Scigantic notebook, where /mnt/archive (SCIGANTIC_MOUNT_PATH) is the mounted dataset.

## Data structure

File types in sampled files: JSON, HDF5

Example file: `model.safetensors`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/5c5569aa-65ba-5e25-9da3-d07fb16c9664
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/5c5569aa-65ba-5e25-9da3-d07fb16c9664
- Over MCP: `get_schema_card` with id `5c5569aa-65ba-5e25-9da3-d07fb16c9664` at https://api.scigantic.com/mcp

## Related topics

- [JSON datasets](https://scigantic.com/topics/json)
- [HDF5 datasets](https://scigantic.com/topics/hdf5)
- [Protein structure datasets](https://scigantic.com/topics/protein-structure)
- [Biology datasets](https://scigantic.com/topics/biology)
- [Protein language model datasets](https://scigantic.com/topics/protein-language-model)
