A SOTA quantization algorithm for high-accuracy low-bit LLM inference, seamlessly optimized for CPU/XPU/CUDA, with multi-datatype support and full compatibility with vLLM, SGLang, and Transformers.

A framework for few-shot evaluation of language models.

Fastest MoE/LLM inference runtime for consumer and edge Blackwell GPUs. SN74 on Gittensor.

SDK for developing enclaves

Go ahead and axolotl questions

CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation

An optimized quantization and inference library for running LLMs locally on modern consumer-class GPUs

LMCache: Supercharge Your LLM with the Fastest KV Cache Layer

SGLang is a high-performance serving framework for large language models and multimodal models.

NanoCluster: Compact & Affordable Cluster for Everyone

compiler learning resources collect.

[MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

Real-Time VLAs via Future-state-aware Asynchronous Inference.

DFlash: Block Diffusion for Flash Speculative Decoding

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

An Optimizer for Nvidia Compilers.

Daily ArXiv Papers.

A retargetable MLIR-based machine learning compiler and runtime toolkit.

Development repository for the Triton language and compiler

TokenSpeed is a speed-of-light LLM inference engine.

FlyDSL is the Python front‑end of the project: Flexible LaYout DSL.

High-performance, light-weight C++ LLM and VLM Inference Software for Physical AI

Medusa: Simple Framework for Accelerating LLM Generation with Multiple Decoding Heads

Open Machine Learning Compiler Framework

All information and news with respect to Falcon-H1 series

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

Interactive 3D visualization of dense decoder-only LLM inference. Companion to the AI Inference Engineer 2026 course.