# PKU-YuanGroup/LLaVA-CoT

[ICCV 2025] LLaVA-CoT, a visual language model capable of spontaneous, systematic reasoning

Repository: https://github.com/PKU-YuanGroup/LLaVA-CoT
Canonical: https://ross.abutalabs.com/products/llava-cot
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-12-12T02:30:57+00:00

## Health v2 (maintenance only)
Score: 47/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 56, release rhythm 35, longevity 46
- inputs: {"age_days": 654, "days_push": 265, "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 2132, forks 81 (observed 2026-08-28T04:06:17.278156+00:00)

## What it is
LLaVA-CoT is an 11B vision language model with training, inference, and dataset-generation code for spontaneous step-by-step multimodal reasoning, released with pretrained weights on Hugging Face. It outperforms several larger commercial VLMs on six multimodal benchmarks and was accepted to ICCV 2025.

## Use cases
- run a vision language model that reasons step by step
- fine-tune a multimodal model for chain-of-thought reasoning
- evaluate a VLM on multimodal benchmarks
- generate a reasoning dataset for vision language model training
- compare open VLM against GPT-4o-mini and Gemini
- deploy a gradio demo for image question answering

## When to choose
- you need an open 11B VLM with systematic reasoning and available weights
- you want reproducible training and dataset generation code for multimodal CoT
- you need a research baseline for multimodal reasoning benchmarks

## When to avoid
- you need a small model that runs without a GPU
- you want a production-ready commercial-grade VLM API
- you need non-English or non-reasoning-focused multimodal tasks

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-inference, nlp, computer-vision, rag
- domain: artificial-intelligence, large-language-models, computer-vision, deep-learning
- platform: python
- tags: vision-language-model, chain-of-thought, multimodal-reasoning, llava, model-weights, iccv-2025, natural-language-processing, gpu, linux

## Member repositories
- PKU-YuanGroup/LLaVA-CoT (main) score 47

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:17.278156+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 2026-08-30T02:52:01.053562+00:00, confidence not recorded.
  - readme: https://github.com/PKU-YuanGroup/LLaVA-CoT (fetched 2026-08-28T04:06:17.278156+00:00, sha 1069e9bee3f4)
- Data as of 2026-08-30T08:39:29.467469+00:00.