Instructions to use HuggingFriends/robojudge-iclr2027-checkpoints with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use HuggingFriends/robojudge-iclr2027-checkpoints with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("HuggingFriends/robojudge-iclr2027-checkpoints", device_map="auto") - Notebooks
- Google Colab
- Kaggle
RoboJudge checkpoints
This repository stores the public checkpoints used by the RoboJudge release:
final_rl_200step/: final SFT+RL RoboJudge-9B checkpoint, sharded BF16.sft_reference_1epoch/: one-epoch SFT checkpoint used in the training-stage ablation, sharded BF16.
The training, inference, and evaluation code is available at
https://github.com/SiyuanMaCS/robojudge-iclr2027. File-level sizes and SHA256
hashes are recorded in CHECKPOINTS_HF_MANIFEST.json and in the GitHub release.
Download
hf download HuggingFriends/robojudge-iclr2027-checkpoints \
--repo-type model \
--local-dir checkpoints_hf
Download only the final model:
hf download HuggingFriends/robojudge-iclr2027-checkpoints \
--repo-type model \
--include 'final_rl_200step/*' \
--local-dir checkpoints_hf
Download only the SFT reference:
hf download HuggingFriends/robojudge-iclr2027-checkpoints \
--repo-type model \
--include 'sft_reference_1epoch/*' \
--local-dir checkpoints_hf
The original FP32 final weights are mirrored at https://huggingface.co/HuggingFriends/RoboJudge-9B.