---

language:

- en

- zh

- de

- multilingual

license: apache-2.0

tags:

- gguf

- vision-language

- ocr

- document-understanding

- qwen2.5-vl

- crispembed

base_model: Qwen/Qwen2.5-VL-3B-Instruct

pipeline_tag: image-text-to-text

---

Qwen2.5-VL-3B — GGUF (CrispEmbed)

GGUF conversion of Qwen/Qwen2.5-VL-3B-Instruct for the CrispEmbed qwen2vl-ocr backend.

3B vision-language model used primarily for document OCR and visual question answering. Processes images at dynamic resolution with a ViT encoder and generates text with a Qwen2.5 LLM decoder.

Architecture

• Vision encoder (ViT): 32 layers (1280-dim, 16 heads, patch_size=14), windowed attention (window=112) with full attention at layers {7, 15, 23, 31}, SwiGLU FFN, mRoPE
• Spatial merger: 4:1 merge to 2048-dim
• LLM decoder: 36-layer Qwen2.5 (2048-dim, GQA, mRoPE), up to 32K context
• Input: image(s) + text prompt
• Output: autoregressive text (OCR transcripts, descriptions, answers)

Files

| File | Size | Description |

|------|------|-------------|

| qwen2.5-vl-3b-q4_k.gguf | 2.6 GB | Q4_K quantized (recommended) |

| qwen2.5-vl-3b-q8_0.gguf | 3.9 GB | Q8_0 quantized |

| qwen2.5-vl-3b-f16.gguf | 7.8 GB | F16 full precision |

Usage

crispembed --backend qwen2vl-ocr -m qwen2.5-vl-3b-q4_k.gguf \
    --image document.png

License

Apache 2.0.

Credits

• Original model: Qwen / Alibaba
• GGUF conversion & runtime: CrispEmbed

Provenance and EU AI Act Art. 53 note

• Upstream model: Qwen/Qwen2.5-VL-3B-Instruct — published by Qwen.
• Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository. No training-content summary was found on the upstream model card at the time of writing; that documentation gap is upstream's and is not filled here.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.