Protenix v0.5.0 base checkpoint (unmodified mirror)

The base checkpoint of ByteDance Protenix v0.5.0, byte for byte as upstream published it. Not retrained, not quantised, not converted, not modified in any way.

File model_v0.5.0.pt
Size 1,474,265,486 bytes
sha256 9ea20b0aba42f2256711da1d0cd081510a4b291e64375bff6b70ced70b87a5f1
Upstream URL https://af3-dev.tos-cn-beijing.volces.com/release_model/model_v0.5.0.pt
Upstream code bytedance/Protenix, tag v0.5.0
Licence Apache 2.0, ByteDance. Full text in LICENSE.

What it is

Weights for Protenix, an open reproduction of AlphaFold 3 that predicts the 3D structure of proteins, nucleic acids and their complexes with ligands. The upstream URL above is the one protenix/web_service/dependency_url.py names at tag v0.5.0.

Why this mirror exists

Upstream serves this file from a single Volcengine bucket in Beijing. A network that cannot route there gets a TCP connection that is accepted and then sends nothing, so an install hangs rather than failing. This copy is the same bytes from a host that answers from anywhere. Upstream stays the fallback, not the only door.

Published so tt-bio can fetch it, which runs Protenix on Tenstorrent accelerators. Nothing about the file is tt-bio specific: any Protenix install can use it.

Verify before you trust it

sha256sum model_v0.5.0.pt
# 9ea20b0aba42f2256711da1d0cd081510a4b291e64375bff6b70ced70b87a5f1

That hash was taken from a fresh upstream download and compared against this copy.

Use it

With tt-bio, nothing to do: tt-bio predict --model protenix-v1 <fasta> fetches from here.

By hand, for an upstream Protenix install:

huggingface-cli download moritztng/protenix-v0.5.0 model_v0.5.0.pt --local-dir .

Attribution

Copyright 2024 ByteDance and/or its affiliates, licensed under Apache 2.0. Redistributed under section 4 of that licence: the licence text travels with the file, the notices in it are unchanged, and this file is unmodified.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support