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

This repository catalogs cutting-edge research papers, practical tools, datasets, and learning materials for AI-powered SVG generation, processing, and manipulation.

Framework for building enterprise-level assistant agents.

A universal sandbox platform for AI application scenarios, providing multi-language SDKs, unified sandbox protocols, and sandbox runtimes for LLM-related capabilities.

Demystifying Reinforcement Learning in Agentic Reasoning

A construction kit for reinforcement learning environment management.

An interface library for RL post training with environments.

QuTLASS: CUTLASS-Powered Quantized BLAS for Deep Learning

Superposition Yields Robust Neural Scaling

Developer Asset Hub for NVIDIA Nemotron — A one-stop resource for training recipes, usage cookbooks, and full end-to-end reference examples to build with Nemotron models

Open-source release accompanying Gao et al. 2025

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

SkyRL: A Modular Full-stack RL Library for LLMs

FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/

APOLLO: SGD-like Memory, AdamW-level Performance; MLSys'25 Oustanding Paper Honorable Mention

The official PyTorch implementation of the paper "Conda: Column-Normalized Adam for Training Large Language Models Faster"

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

Complete simulation of IEEE 754 fixed and floating point specification to any precision

Miles is an enterprise-facing reinforcement learning framework for large-scale MoE post-training and production workloads, forked from and co-evolving with slime.

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

A Survey of Efficient Attention Methods: Hardware-efficient, Sparse, Compact, and Linear Attention

A framework to compare low-bit integer and float-point formats

Agentic Learning Powered by AWorld

Tongyi Deep Research, the Leading Open-source Deep Research Agent

[ACL 2025] Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.