Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

Frequency-based KV cache pruning for llama.cpp — 25% cache reduction, improved PPL at long context. GPU compaction kernel for HIP/ROCm.

Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding

TriAttention — Efficient long reasoning with trigonometric KV cache compression. Enables OpenClaw local deployment on memory-constrained GPUs.

JoyAI-Image is the unified multimodal foundation model for image understanding, text-to-image generation, and instruction-guided image editing.

[Official Repo] SpatialEdit: Benchmarking Fine-Grained Image Spatial Editing

[CVPRW oral 2022] MANIQA: Multi-dimension Attention Network for No-Reference Image Quality Assessment

The repo is finally unlocked. enjoy the party! The fastest repo in history to surpass 100K stars ⭐. Join Discord: https://discord.gg/5TUQKqFWd Built in Rust using oh-my-codex.

ShotStream: Streaming Multi-Shot Video Generation for Interactive Storytelling

AutoGaze automatically removes redundant patches in a video, reducing #tokens in ViT/MLLM by 4x-100x.

Official code of Motus: A Unified Latent Action World Model

Code for the paper “Stable Asynchrony: Variance-Controlled Off-Policy RL for LLMs”

Minute-long video generation at 24FPS.

A unified library of SOTA model optimization techniques like quantization, pruning, distillation, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks …

Official repository of Utonia: Toward One Encoder for All Point Clouds

CVPR 2026 | Official Implementation of "MultiShotMaster: A Controllable Multi-Shot Video Generation Framework" 🔥

Spa3R: Predictive Spatial Field Modeling for 3D Visual Reasoning

A rejection-sampling based distribution alignment method for extreme actor-policy mismatch RL Training

[ICLR 2026 Oral] Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation

NVIDIA FastGen: Fast Generation from Diffusion Models

Qwen3-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

Consistent Autoregressive Video Generation with Long Context

Code for "LIVE: Long-horizon Interactive Video World ModEling"

[NeurIPS 2025] Radial Attention: O(nlogn) Sparse Attention with Energy Decay for Long Video Generation

A unified inference and post-training framework for accelerated video generation.