# Liuziyu77/Visual-RFT

Official repository of 'Visual-RFT: Visual Reinforcement Fine-Tuning' & 'Visual-ARFT: Visual Agentic Reinforcement Fine-Tuning'’

Repository: https://github.com/Liuziyu77/Visual-RFT
Canonical: https://ross.abutalabs.com/products/visual-rft
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Last push: 2025-10-29T07:28:02+00:00

## Health v2 (maintenance only)
Score: 42/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 49, release rhythm 35, longevity 39
- inputs: {"age_days": 555, "days_push": 308, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2271, forks 111 (observed 2026-08-28T04:06:33.164787+00:00)

## What it is
Official research code for Visual-RFT and Visual-ARFT, applying GRPO-based reinforcement fine-tuning with rule-based verifiable rewards to multimodal large vision-language models like Qwen2-VL. It extends Deepseek-R1's RL strategy to visual perception tasks such as open vocabulary detection, few-shot detection, reasoning grounding, and fine-grained image classification.

## Use cases
- reinforcement fine-tune a vision-language model with GRPO
- train Qwen2-VL for open vocabulary detection
- improve few-shot object detection with RL rewards
- apply R1-style verifiable rewards to multimodal tasks
- fine-tune LVLMs for reasoning grounding
- train agentic visual models with Visual-ARFT

## When to choose
- you want to reproduce or extend Visual-RFT/Visual-ARFT research
- you need GRPO-based RL fine-tuning for Qwen2-VL on perception tasks
- you want rule-based verifiable rewards for detection or classification fine-tuning

## When to avoid
- you need a production-ready training framework with broad model support
- you only want inference without fine-tuning
- you use models other than Qwen2-VL without adaptation work

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-training, computer-vision, agent-framework
- domain: machine-learning, computer-vision, large-language-models, deep-learning
- platform: python
- tags: reinforcement-fine-tuning, grpo, multimodal, qwen2-vl, visual-perception, research-code, verifiable-rewards, ai-agents, gpu, linux

## Member repositories
- Liuziyu77/Visual-RFT (main) score 42

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:33.164787+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:41:47.552135+00:00, confidence not recorded.
  - readme: https://github.com/Liuziyu77/Visual-RFT (fetched 2026-08-28T04:06:33.164787+00:00, sha 5c90b3d16082)
- Data as of 2026-08-30T08:39:29.467469+00:00.