# 2U1/Qwen-VL-Series-Finetune

An open-source implementaion for fine-tuning Qwen-VL series by Alibaba Cloud.

Repository: https://github.com/2U1/Qwen-VL-Series-Finetune
Canonical: https://ross.abutalabs.com/products/qwen-vl-series-finetune
Language: Python
License: Apache-2.0
License Family: permissive
Topics: multimodal, qwen2-vl, vision-language, vision-language-model, qwen3-vl, qwen2-5-vl, qwen3-5, vlm
Last push: 2026-08-22T20:39:07+00:00

## Health v2 (maintenance only)
Score: 67/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 51
- inputs: {"age_days": 722, "days_push": 11, "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 1960, forks 222 (observed 2026-08-28T04:05:59.381156+00:00)

## What it is
An open-source Python repository providing training scripts for fine-tuning Alibaba's Qwen-VL series of vision-language models (Qwen2-VL, Qwen2.5-VL, Qwen3-VL, Qwen3.5) using HuggingFace Transformers and Liger-Kernel. It supports SFT, DPO, GRPO, LoRA/DoRA, classification, and multi-image/video training with memory optimizations.

## Use cases
- fine-tune qwen2-vl on custom image datasets
- train qwen3-vl with lora on my own data
- run dpo training on a vision language model
- grpo training for multimodal models
- fine-tune qwen2.5-vl for video understanding
- train a vlm classifier on custom categories
- reduce gpu memory when fine-tuning vision language models

## When to choose
- you need to fine-tune any Qwen-VL series model with SFT, DPO, or GRPO
- you want memory-efficient training via Liger-Kernel and attention optimizations
- you need LoRA/DoRA, partial layer freezing, or mixed-modality (image/video) training support

## When to avoid
- you want to fine-tune non-Qwen vision-language models (use the author's sibling repos or generic frameworks like LLaMA-Factory)
- you need inference/serving rather than training
- you prefer a GUI or no-code fine-tuning workflow

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, deep-learning
- domain: machine-learning, deep-learning, large-language-models, computer-vision
- platform: python
- tags: fine-tuning, vision-language-model, qwen-vl, multimodal, lora, dpo, grpo, liger-kernel, huggingface, gpu

## Member repositories
- 2U1/Qwen-VL-Series-Finetune (main) score 67

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:59.381156+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 2026-08-30T03:05:41.056688+00:00, confidence not recorded.
  - readme: https://github.com/2U1/Qwen-VL-Series-Finetune (fetched 2026-08-28T04:05:59.381156+00:00, sha d373ca69eeba)
- Data as of 2026-08-30T08:39:29.467469+00:00.