# merveenoyan/smol-vision

Recipes for shrinking, optimizing, customizing cutting edge vision models. 💜

Repository: https://github.com/merveenoyan/smol-vision
Canonical: https://ross.abutalabs.com/products/smol-vision
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Last push: 2026-08-12T11:23:25+00:00

## Health v2 (maintenance only)
Score: 69/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 64
- inputs: {"age_days": 903, "days_push": 21, "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 1973, forks 151 (observed 2026-08-28T04:06:01.075699+00:00)

## What it is
A collection of Jupyter Notebook recipes for shrinking, optimizing, and fine-tuning cutting-edge vision and multimodal AI models. It covers quantization with ONNX/Optimum, knowledge distillation, and fine-tuning models like PaliGemma, Florence-2, and DINOv3.

## Use cases
- quantize vision models to run on smaller hardware
- fine-tune a vision language model like PaliGemma
- optimize object detection models with ONNX Runtime
- learn knowledge distillation for image classification
- fine-tune Florence-2 for document question answering
- fine-tune Kosmos2.5 for OCR with bounding boxes
- export vision models to ONNX and reduce them to fp16

## When to choose
- you want hands-on notebooks for optimizing or shrinking vision models
- you need recipes for fine-tuning multimodal or vision-language models
- you want to learn quantization and ONNX export for computer vision models

## When to avoid
- you need a production-ready library or CLI rather than example notebooks
- you work with non-vision modalities like audio or text-only LLMs
- you need guaranteed long-term API stability since content is tutorial notebooks

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, image-processing, ocr, llm-training
- domain: computer-vision, machine-learning, deep-learning, tutorials
- platform: python
- tags: notebooks, vision-models, quantization, fine-tuning, multimodal, knowledge-distillation, onnx, huggingface

## Member repositories
- merveenoyan/smol-vision (main) score 69

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:01.075699+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 2026-08-30T03:04:59.110189+00:00, confidence not recorded.
  - readme: https://github.com/merveenoyan/smol-vision (fetched 2026-08-28T04:06:01.075699+00:00, sha 1b8152b91e86)
- Data as of 2026-08-30T08:39:29.467469+00:00.