LightRFT
LightRFT: Light, Efficient, Omni-modal & Reward-model Driven Reinforcement Fine-Tuning Framework
- 381
- Python
- Apache-2.0
- 3 mo ago
LightRFT: Light, Efficient, Omni-modal & Reward-model Driven Reinforcement Fine-Tuning Framework
OpenJudge: A Unified Framework for Holistic Evaluation and Quality Rewards
Pairwise LLM judges (A/B/tie): budget-aware multi-turn packing, position-bias correction, pseudo-label distillation. Generalized from the 4th-place (gold) solution to Kaggle LMSYS Chatbot Arena.
[ICLR 2024] SemiReward: A General Reward Model for Semi-supervised Learning
Self-evolving agentic reward framework for image-editing evaluation — 47.4% on EditReward-Bench from only 100 preference demos, no reward-model training. arXiv 2605.08703.
[NeurIPS 2026] Official Implementation of "Visual-ERM: Reward Modeling for Visual Equivalence"
An official implementation of "SPARK: Synergistic Policy And Reward Co-Evolving Framework"
Text classification model, used with transformers.
Athena-PRM: Enhancing Large Reasoning Model with Data-efficient Process Reward Models
A world model as a judge: early, auditable verdicts on partial robot-manipulation rollouts
Image text to text model, used with transformers.
Error bounds and targeted falsification for Triton kernels
Runs in the browser on Hugging Face