RedAI Infra ORG

An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale

Implementation of an efficient LLM architecture: the Pair-In / Pair-Out Model (PIPO)

A research project that speeds up AI text generation by compressing two input tokens into one and predicting an extra output token per step, cutting inference time without sacrificing accuracy.

[ACL 2026] HiSVD: Principled Low-Rank Approximation of LLMs via Hierarchical Modeling of Information Capacity and Spectral Structure

Official code, data, and models for "Hint Tuning: Less Data Makes Better Reasoners"

A research project that trains AI reasoning models more efficiently by giving each problem only as much step-by-step reasoning as it actually needs, short answers for easy problems, long chains for hard ones.

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

Allow torch tensor memory to be released and resumed later

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

Training library for Megatron-based models with bidirectional Hugging Face conversion capability

Xiaohongshu AI Platform Team

This is the official **live** mirror of https://gitcode.com/Ascend/TransferQueue. Feel free to contribute!

A unified library of SOTA model optimization techniques like quantization, pruning, distillation, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.

A Datacenter Scale Distributed Inference Serving Framework

A high-throughput and memory-efficient inference and serving engine for LLMs