Downloads · 30 days
10
43% of all-time downloads
OLMo-2-1B-Midtrain-50B-AdamW-10percent
WattsIshaan/OLMo-2-1B-Midtrain-50B-AdamW-10percent
OLMo-2-1B-Midtrain-50B-AdamW-10percent is a machine learning model from WattsIshaan. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as cc-by-4.0.
OLMo-2-1B checkpoint midtrained with AdamW on 50B tokens from Dolmino. This is a sharded FSDP checkpoint (rank0.pt–rank7.pt) at 90% of midtraining (step 21000 at 1024 bsz), and can be annealed for the remaining 10% st…
All-time downloads
23
Public
Repo size
17.8 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt17.8 GB · 100%
At a glance
- License
- cc-by-4.0
- Access
- Public
- Created
- Sep 12, 2026
- Updated
- Sep 12, 2026
- SHA
- 6bb6d3f6
Base models
Datasets
Model card
From the Hugging Face model README
OLMo-2-1B checkpoint midtrained with AdamW on 50B tokens from Dolmino. This is a sharded FSDP checkpoint (rank0.pt–rank7.pt) at 90% of midtraining (step 21000 at 1024 bsz), and can be annealed for the remaining 10% steps.
Unshard with python scripts/unshard.py ./OLMo-2-1B-Midtrain-50B-AdamW-10percent ./unsharded from the OLMo repo before loading as a single model.pt.
Paper:"Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting" Ishaan Watts*, Catherine Li*, Sachin Goyal, Jacob Mitchell Springer, Aditi Raghunathan ICML2026 https://arxiv.org/abs/2605.02105
- License
- cc-by-4.0
- Created
- Sep 12, 2026
- Updated
- Sep 12, 2026