Starred repositories
SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents through trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts.
MA-EgoQA: Question Answering over Egocentric Videos from Multiple Embodied Agents
General plug-and-play inference library for Recursive Language Models (RLMs), supporting various sandboxes.
Biologically-inspired memory for AI agents. Decay, retrieval strengthening, consolidation. Zero dependencies.
[ACL 2026 Oral] UniversalRAG: Retrieval-Augmented Generation over Corpora of Diverse Modalities and Granularities
[CVPR 2026 Highlight] WorldMM: Dynamic Multimodal Memory Agent for Long Video Reasoning
Official implementation of Generative Diffusion Models for Antibody Design, Docking, and Optimization.
Score-Control for Hallucination Reduction in Diffusion Models.
TRACE: Evidence Grounding-Guided Multi-Video Event Understanding and Claim Generation
[CVPR 2026] Official implementation of Fairness-aware Parameter Efficient Finetuning For Large Vision Language Assistants
[ACL 2026] CRAFT: Critic-Refined Adaptive Key-Frame Targeting for Multimodal Video Question Answering
Train transformer language models with reinforcement learning.
[CVPR 2026 Highlight] VideoITG: Multimodal Video Understanding with Instructed Temporal Grounding
[ICML 2026] The offical code of Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis
Cosmos-Predict2 is a collection of general-purpose world foundation models for Physical AI that can be fine-tuned into customized world models for downstream applications.
rCM & Causal-rCM: Leading and Unified Algorithms/Infrastructures for Bidirectional/Autoregressive Video Diffusion Distillation at Scale
Official PyTorch Implementation of "SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers"
An efficient distillation method for flow matching models
[CVPR 2025] Adaptive Keyframe Sampling for Long Video Understanding
(NeurIPS 2024 Oral đŸ”¥) Improved Distribution Matching Distillation for Fast Image Synthesis
PyTorch code and model checkpoints for Score identity Distillation (SiD) and its adversarial version (SiDA)
A minimal implementation of Drifting Models for 2D toy data. Unlike diffusion/flow models that iterate at inference, drifting models evolve the pushforward distribution during training and generat…
Personal PyTorch implementation of "Generative Modeling via Drifting" with Claude