Advancing Open-source World Models

Official codebase for "Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation"

Official implementation of "Focus-dLLM: Accelerating Long-Context Diffusion LLM Inference via Confidence-Guided Context Focusing".

Official Implementation of DART (DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference).

[ICLR 2026] This is the official PyTorch implementation of "QVGen: Pushing the Limit of Quantized Video Generative Models".

Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models

Block Diffusion for Ultra-Fast Speculative Decoding

Towards Scalable Pre-training of Visual Tokenizers for Generation

A Survey of Reinforcement Learning for Large Reasoning Models

The official implementation for [NeurIPS2025 Oral] Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free

Code for the paper “Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling”

HY-World 1.5: A Systematic Framework for Interactive World Modeling with Real-Time Latency and Geometric Consistency

Official Implementation of "MemFlow: Flowing Adaptive Memory for Consistent and Efficient Long Video Narratives"

TurboDiffusion: 100–200× Acceleration for Video Diffusion Models

dLLM: Simple Diffusion Language Modeling

[NeurIPS 2025] Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance

2026 AI/ML internship & new graduate job list updated daily

DiffusionNFT: Online Diffusion Reinforcement with Forward Process

Train transformer language models with reinforcement learning.

Official Code of "Distribution Matching Distillation Meets Reinforcement Learning"

A curated list of papers on reinforcement learning for video generation

Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation

[ICLR 26] TempFlow-GRPO (Temporal Flow GRPO), a principled GRPO framework that captures and exploits the temporal structure inherent in flow-based generation.

[ICCV2025 highlight]Rectifying Magnitude Neglect in Linear Attention

slime is an LLM post-training framework for RL Scaling.

Implementation for FP8/INT8 Rollout for RL training without performence drop.

QeRL enables RL for 32B LLMs on a single H100 GPU.

An official implementation of DanceGRPO: Unleashing GRPO on Visual Generation

[NeurIPS 2025] An official implementation of Flow-GRPO: Training Flow Matching Models via Online RL

[ATC'25] Katz is a high-performance serving system designed specifically for diffusion model workflows with multiple adapters.