Fine-tuning & Reinforcement Learning for LLMs. 🦥 Train OpenAI gpt-oss, DeepSeek-R1, Qwen3, Gemma 3, TTS 2x faster with 70% less VRAM.

Finetune SAM3 with LoRA — optimized for images. A simple setup for training SAM3 on image datasets. Video finetuning is not yet supported but planned for future releases.

Use DINOv3’s powerful, self-supervised visual features + YOLOv12’s blazing-fast detection, all in one repo. Whether you have only a few hundred labeled images or a medium-sized dataset, DINOV3-YOLO…

The official code of "VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning"

an method to make vlm think like r1

This is the official implementation of our paper "Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension"

Parsing-free RAG supported by VLMs

This repository contains a project I completed during my internship at meituan. Specifically, it performs SFT on Qwen2-vl, uses internal company data, and fine-tunes Qwen2-vl for downstream tasks (…

Repository for PrimeVul Vulnerability Detection Dataset

A Generalizable World Model for Autonomous Driving

AL-Ref-SAM 2: Unleashing the Temporal-Spatial Reasoning Capacity of GPT for Training-Free Audio and Language Referenced Video Object Segmentation

视频理解：千问视频多模态模型 & Dify

Code for 'LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders'

Official Implementation of SnAG (CVPR 2024)

A web UI Project In order to learn the large language model. This project includes features such as chat, quantization, fine-tuning, prompt engineering templates, and multimodality.

Inference code for LLaMA models

code for Ego4D Workshop@CVPR 2023 - 1st in MQ & 2nd in NLQ challenge

The code of the paper "NExT-Chat: An LMM for Chat, Detection and Segmentation".

MMICL, a state-of-the-art VLM with the in context learning ability from ICL, PKU

The official implementation of AdaTAD: End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames

A general representation model across vision, audio, language modalities. Paper: ONE-PEACE: Exploring One General Representation Model Toward Unlimited Modalities

Video-LLaVA: Learning United Visual Representation by Alignment Before Projection