SkillZero
[EMNLP 2026] Official code for "SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization"
- 391
- Python
- Apache-2.0
- 3 mo ago
[EMNLP 2026] Official code for "SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization"
Code & data for the EMNLP 2024 paper: Is Child-Directed Speech Effective Training Data for Language Models?
Guarded recursive self-improvement (RSI) for code agents with persistent e-LOND False Discovery Control (FDR) across 4 comprehensive studies.
GPU-optimized UL2 mixture-of-denoisers data collator for T5/FLAN encoder-decoder pretraining. Supports span corruption, prefix LM, infilling, curriculum learning, Flash Attention unpadding, and HuggingFace Trainer integration.
LLM learns to fix data, not models — GRPO RL env.
Runs in the browser on Hugging Face