# StarsfieldAI/R1-V

Witness the aha moment of VLM with less than $3.

Repository: https://github.com/StarsfieldAI/R1-V
Canonical: https://ross.abutalabs.com/products/r1-v
Language: Python
License Family: other
Last push: 2025-05-19T16:32:41+00:00

## Health v2 (maintenance only)
Score: 21/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 22, release rhythm 8, longevity 41
- inputs: {"age_days": 577, "days_push": 471, "days_rel": 565, "gap_med": null, "n_releases_24m": 1}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4063, forks 281 (observed 2026-08-28T04:08:34.050034+00:00)

## What it is
R1-V is an open-source research codebase for training vision-language models with reinforcement learning (RLVR/GRPO), demonstrating strong generalization for under $3 of compute. It includes training scripts, datasets, and evaluation tooling for models like Qwen2-VL and Qwen2.5-VL.

## Use cases
- train a vision language model with reinforcement learning
- reproduce the R1 aha moment on VLMs cheaply
- run GRPO training on Qwen2.5-VL
- fine-tune VLMs with verifiable rewards
- distill DeepSeek-R1 visual reasoning traces
- experiment with RLVR for multimodal reasoning

## When to choose
- you want to RL-train open VLMs on a small budget
- you're doing research on RLVR for multimodal models
- you need CLEVR/GEOQA training datasets and scripts

## When to avoid
- you need a production-ready training framework with support guarantees
- you need a licensed library for commercial use (no license is specified)
- you only need inference, not training

## Facets
- artifact type: library
- maturity: active
- function: llm-training, reinforcement-learning, machine-learning, deep-learning
- domain: large-language-models, machine-learning, computer-vision, artificial-intelligence, deep-learning
- platform: python
- tags: vision-language-models, rlvr, grpo, vlm-training, research-code, qwen2-vl, vllm, multimodal, gpu, linux

## Member repositories
- StarsfieldAI/R1-V (main) score 21

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:34.050034+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 2026-08-29T18:23:34.197949+00:00, confidence not recorded.
  - readme: https://github.com/StarsfieldAI/R1-V (fetched 2026-08-28T04:08:34.050034+00:00, sha 477f22e03fdb)
- Data as of 2026-08-30T08:39:29.467469+00:00.