MC-Sparse: Deconstructing and Closing the Dense–Sparse Attention Gap in Diffusion Transformers.

Skills for running, porting, profiling, and tuning code on Intel CPUs, GPUs, and the libraries and runtimes behind them.

An end-to-end agent project for GPU kernel implementation, analysis, profiling, and iterative optimization. It helps an agent turn PyTorch logic or an existing kernel into a high-performance GPU ke…

A benchmark of real-world DL kernel problems

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

REAP: Router-weighted Expert Activation Pruning for SMoE compression

Archived: b12x development has moved to FlashInfer. See README for migration and PR-porting guidance.

Common recipes to run vLLM

DeepSeek Harness: Everything is a Plugin.

A modern, high customizable, responsive Jekyll theme for documentation with built-in search.

A unified, torch-native training framework for MTP and block-parallel speculative decoding.

Reference implementation and examples of the CuTe Layout representation and algebra.

vLLM plugin for attention-ffn disaggregation support

A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training

[CVPR 2026 Highlight] Official implementation of Log-linear Sparse Attention (LLSA).

Nsight Python is a Python kernel profiling interface based on NVIDIA Nsight Tools

Model compression toolkit engineered for enhanced usability, comprehensiveness, and efficiency.

DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms

[ICML2025] SpargeAttention: A training-free sparse attention that accelerates any model inference.

[ICML2025, NeurIPS2025 Spotlight] Sparse VideoGen 1 & 2: Accelerating Video Diffusion Transformers with Sparse Attention

view triton linear layouts

🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talking like a caveman.

The 100 line AI agent that solves GitHub issues or helps you in your command line. Radically simple, no huge configs, no giant monorepo—but scores >74% on SWE-bench verified!

A simple and effective post training quantization toolkit for high-accuracy low-bit LLM inference|简洁且高效的后训练量化工具包

Wrapper around tool using LLMs for agentic workflows

A Python DSL to write Nvidia PTX for Hopper and Blackwell in JAX and PyTorch

Agent skills for vLLM