Ongoing research training transformer models at scale

Official Repository for "Glyph: Scaling Context Windows via Visual-Text Compression"

🌿 DeepPrune: Parallel Scaling without Inter-trace Redundancy

Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

[SIGGRAPH Asia 2025] CHARM: Control-point-based 3D Anime Hairstyle Auto-Regressive Modeling

🚀🚀 Efficient implementations of Native Sparse Attention

Renderer for the harmony response format to be used with gpt-oss

GLM-SIMPLE-EVALS: The evaluation repository for the GLM-4.5 series of models by Z.ai.

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

[COLM 2024] A Survey on Deep Learning for Theorem Proving

An efficient implementation of the NSA (Native Sparse Attention) kernel

TradingAgents: Multi-Agents LLM Financial Trading Framework

Scaling RL on advanced reasoning models

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

[SIGGRAPH 2025] PrimitiveAnything: Human-Crafted 3D Primitive Assembly Generation with Auto-Regressive Transformer

ReLE评测：中文AI大模型能力评测（持续更新）：目前已囊括335个大模型，覆盖chatgpt、gpt-5.2、o4-mini、谷歌gemini-3-pro、Claude-4.5、文心ERNIE-X1.1、ERNIE-5.0-Thinking、qwen3-max、百川、讯飞星火、商汤senseChat等商用模型， 以及kimi-k2、ernie4.5、minimax-M2、deepseek-…

[ICML 2025] XAttention: Block Sparse Attention with Antidiagonal Scoring

[ACL 2025] Agentic Reward Modeling: Integrating Human Preferences with Verifiable Correctness Signals for Reliable Reward Systems

Official codebase for "STAIR: Improving Safety Alignment with Introspective Reasoning"

[AAAI 2026] VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation

[ICML'24] Data and code for our paper "Training-Free Long-Context Scaling of Large Language Models"

[NeurIPS 2024] AlphaTablets: A Generic Plane Representation for 3D Planar Reconstruction from Monocular Videos