---
license: mit
library_name: tt-bio
tags:
  - biology
  - protein-structure-prediction
  - protein-ligand
  - binding-affinity
  - boltz
  - tenstorrent
---

# Boltz-2 weights (tt-bio mirror)

Mirror of the [Boltz-2](https://github.com/jwohlwend/boltz) structure- and
affinity-prediction weights, packaged for use with
[tt-bio](https://github.com/moritztng/tt-bio) on Tenstorrent hardware. The
files are byte-for-byte identical to the upstream Boltz-2 release; this repo
simply hosts them on the Hugging Face Hub so tt-bio can fetch them with
`huggingface_hub` like every other model it runs.

## Files

| File | Description |
|------|-------------|
| `boltz2_conf.ckpt` | Boltz-2 structure / confidence model |
| `boltz2_aff.ckpt`  | Boltz-2 binding-affinity model |
| `mols.tar`         | CCD molecule / component library used during featurization |

## Usage

tt-bio downloads these automatically:

```bash
tt-bio predict examples/prot.yaml --model boltz2 --use_msa_server --override
```

Or fetch a single file directly:

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download("moritztng/boltz-2", "boltz2_conf.ckpt")
```

## Credit & license

Boltz-2 is developed by Jeremy Wohlwend, Gabriele Corso, Saro Passaro and the
Boltz authors (MIT Jameel Clinic) and released under the MIT License. The
weights and the accompanying `LICENSE` are redistributed here unmodified under
those same terms. See the upstream project at
<https://github.com/jwohlwend/boltz> and please cite the Boltz-2 work if you
use these weights.