torchcomms: a modern PyTorch communications API

Official code repo for our work "Native Visual Understanding: Resolving Resolution Dilemmas in Vision-Language Models"

Qwen3-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.

SGLang is a fast serving framework for large language models and vision language models.

The TinyLlama project is an open endeavor to pretrain a 1.1B Llama model on 3 trillion tokens.

Train a 1B LLM with 1T tokens from scratch by personal

个人构建MoE大模型：从预训练到DPO的完整实践

训练一个对中文支持更好的LLaVA模型，并开源训练代码和数据。

Phi2-Chinese-0.2B 从0开始训练自己的Phi2中文小模型，支持接入langchain加载本地知识库做检索增强生成RAG。Training your own Phi2 small chat model from scratch.

中文对话0.2B小模型（ChatLM-Chinese-0.2B），开源所有数据集来源、数据清洗、tokenizer训练、模型预训练、SFT指令微调、RLHF优化等流程的全部代码。支持下游任务sft微调，给出三元组信息抽取微调示例。

Reference PyTorch implementation and models for DINOv3

Nano vLLM

A single-file educational implementation for understanding vLLM's core concepts and running LLM inference.

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

将SmolVLM2的视觉头与Qwen3-0.6B模型进行了拼接微调

[ICCV2025] PyTorch implementation of "Perceive, Understand and Restore: Real-World Image Super-Resolution with Autoregressive Multimodal Generative Models"

High-performance inference framework for large language models, focusing on efficiency, flexibility, and availability.

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

[CVPR 2025] Official implementation for "Empowering LLMs to Understand and Generate Complex Vector Graphics" https://arxiv.org/abs/2412.11102

从无名小卒到大模型（LLM）大英雄~ 欢迎关注后续！！！

🚀 「大模型」1小时从0训练26M参数的视觉多模态VLM！🌏 Train a 26M-parameter VLM from scratch in just 1 hours!

整理开源的中文大语言模型，以规模较小、可私有化部署、训练成本较低的模型为主，包括底座模型，垂直领域微调及应用，数据集与教程等。

Solve Visual Understanding with Reinforced VLMs

Omegance: A Single Parameter for Various Granularities in Diffusion-Based Synthesis (ICCV, 2025)

[ICLR 2025] Animate-X: Universal Character Image Animation with Enhanced Motion Representation

[CVPR 2025] MatAnyone: Stable Video Matting with Consistent Memory Propagation

Exploiting unlabeled data with vision and language models for object detection, ECCV 2022

🚀🚀 「大模型」2小时完全从0训练26M的小参数GPT！🌏 Train a 26M-parameter GPT from scratch in just 2h!

DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding