Scale agents (model+harness) and go beyond the imagination.
- Hugging Face Transformers (149K+ ⭐) - Pioneered multiple choice training pipeline (PR #1004) and RoBERTa SQuAD support with 20x preprocessing speedup (PR #2173). Top 10 contributor in 2019.
- NVIDIA TensorRT-LLM (12K+ ⭐) - Supported RoBERTa for high-performance inference
- verl-tool (980+ ⭐) - Core design contributor | Unified RL framework for tool-calling agents
- OpenResearcher (750+ ⭐) - Core contributor | Fully open pipeline for long-horizon deep research
- ClawBench (320+ ⭐) - Project Lead/Advisor | Browser AI agent benchmark across 144 live websites
- BrowseComp-Plus (270+ ⭐) - Core contributor | Fair evaluation for deep-research agents | ACL 2026
- DCI-Agent-Lite (270+ ⭐) - Project Lead/Advisor | Agentic search via direct corpus interaction
- BrowserAgent (33 ⭐) - Core contributor | Web agents with human-inspired browsing | TMLR 2025
- SWE-Next - Core contributor | Scalable real-world software engineering tasks for agents
- AceCoder (101 ⭐) - Core contributor | Coder RL via automated test-case synthesis | ACL 2025
- ScholarCopilot (250+ ⭐) - Core contributor | Academic writing with accurate citations | COLM 2025
- One-Shot-CFT (34 ⭐) - Core contributor | Unleashing reasoning via critique fine-tuning | EMNLP 2025
- StructEval (23 ⭐) - Core contributor | Benchmarking LLMs' structured output | TMLR 2025
- SWE-QA-Pro - Core contributor | Repository-level code understanding | ACL 2026
- EvolveCoder - Core contributor | Adversarial test-case evolution for code RL
- EditReward (146 ⭐) - Core contributor | Reward model for image editing | ICLR 2026
- Context-Forcing (85 ⭐) - Core contributor | Consistent autoregressive video generation | ICML 2026
- VideoScore2 (46 ⭐) - Core contributor | Think before you score in video evaluation
- VisualWebInstruct (40 ⭐) - Core contributor | Scaling multimodal instruction data | EMNLP 2025
- ImagenWorld (33 ⭐) - Core contributor | Stress-testing image generation | ICLR 2026
- VisCoder (20 ⭐) - Core contributor | LLMs for Python visualization code | EMNLP 2025
- learn-nlp-with-transformers (3.2K ⭐) - Owner & lead author
- deeplearningbasics (107 ⭐) - Owner & lead author
49 publications | 1,040+ citations | h-index 16 | Google Scholar
NeurIPS (Spotlight) · ACL · EMNLP · SIGIR · ICLR · ICML · COLM · TMLR
Research supported by OpenAI, Google Cloud, HuggingFace, Lambda AI
Making state-of-the-art NLP and LLM accessible
