Instructions to use Gsj49/verl_reasoning_checkpoints with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Gsj49/verl_reasoning_checkpoints with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Gsj49/verl_reasoning_checkpoints", device_map="auto") - Notebooks
- Google Colab
- Kaggle
VERL Reasoning โ JASA revision checkpoints
Curated merged Hugging Face checkpoints from the JASA revision experiments in
Gsj49/verl_reasoning.
- DAPO formal
global_step_500checkpoints: 34 - Assumption 6 checkpoints at steps 10, 90, and 170: 6
- Total checkpoint payload: 535.38 GiB
Each checkpoint directory contains a complete Transformers-format model and is
independently checksummed in checkpoint_manifest.tsv and
checkpoint_manifest.json.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support