Downloads · 30 days
0
open-athena/marinfold-exp199
marinfold-exp199 is a machine learning model from open-athena. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for transformers.
Checkpoint artifacts from the MarinFold contacts-v1 AFDB/ESM optimization experiment tracked in Open-Athena/MarinFold199.
Downloads · 30 days
0
Access
Public
Updated Aug 15, 2026
Repo size
118 GB
Likes
0
Public
Click a slice to open those files.
Other88.3 GB · 75%
From the Hugging Face model README
Checkpoint artifacts from the MarinFold contacts-v1 AFDB/ESM optimization experiment tracked in Open-Athena/MarinFold#199.
| Run | Final <code>eval/tokenized/contacts-v1-val/loss</code> | Levanter checkpoints | Hugging Face export |
|---|---|---|---|
| <code>prot-exp199-cv1-s01-m1-p03-aug-us-east1</code> | <code>3.0115306377</code> | step-72599 | step-72599 |
| <code>prot-exp199-cv1-s01-m1-p03-base-us-east5</code> | <code>3.0074224472</code> | step-72599 | step-72599 |
| <code>prot-exp199-cw-cv1-s02-m1-p06-aug</code> | <code>2.9712009430</code> | step-116160, step-145199 | step-145199 |
| <code>prot-exp199-cv1-cont-s03-m1-p03-srcbase-aug100-us-east1</code> | <code>2.9638261795</code> | step-145199 | step-145199 |
| <code>prot-exp199-cw-cv1-p06-cool-s01</code> | <code>2.9396727085</code> | — | step-290400 |
The <code>checkpoints/</code> paths contain the original Levanter OCDBT full state, including AdamW state for continued training. The <code>hf/</code> paths contain <code>Qwen3ForCausalLM</code> safetensors exports and the 2,845-token contacts-v1 tokenizer.
The two original <code>m1-p03</code> runs trained for 72,600 steps, making <code>step-72599</code> their zero-indexed final checkpoint. The CoreWeave <code>m1-p06</code> and continued <code>srcbase-aug100</code> runs trained through <code>step-145199</code>. Only final checkpoint formats are retained for the continuation run.
<code>p06-cool-s01</code> is a cooldown of the CoreWeave <code>m1-p06</code> line. It resumed from <code>step-261360</code> of <code>prot-exp199-cw-cv1-cont-s02-m1-p06-srcaug-aug100</code> and decayed the learning rate linearly from <code>1e-3</code> to zero over 29,040 steps, ending at <code>step-290400</code>. Only its Hugging Face export is retained.