uni-mumer-qwen2.5-vl-3b-GGUF

<!-- crispembed-license-notice -->

License: Qwen RESEARCH LICENSE AGREEMENT — non-commercial use only. phxember/Uni-MuMER-Qwen2.5-VL-3B is a fine-tune of Qwen/Qwen2.5-VL-3B-Instruct; its apache-2.0 tag cannot relax the base's Qwen Research License (LICENSE, verbatim, section 3d), which governs these weights; commercial use needs a license from Alibaba Cloud. An earlier version of this card said apache-2.0; that was incorrect. Built with Qwen. CrispEmbed loads them only with CRISPEMBED_ACCEPT_QWEN_RESEARCH_LICENSE=1. Evidence: license audit.

<!-- /crispembed-license-notice -->

Uni-MuMER-Qwen2.5-VL-3B GGUF

GGUF conversions of phxember/Uni-MuMER-Qwen2.5-VL-3B for CrispEmbed inference.

Handwritten Mathematical Expression Recognition (HMER) model. Converts images of handwritten math into LaTeX. Fine-tuned from Qwen2.5-VL-3B-Instruct using multi-task training (recognition + symbol counting + position identification).

Based on the Uni-MuMER paper (NeurIPS 2025 Spotlight). This is the highest-accuracy variant at 82.25% CROHME.

Model variants

Architecture

• Vision: Qwen2.5-VL ViT (32L, 1280d, 16 heads, patch=14, windowed attn)
• Merger: RMSNorm + SwiGLU MLP (4:1 spatial merge)
• LLM: Qwen2.5 decoder (36L, 2048d, GQA 16/2, SwiGLU, mRoPE)
• Parameters: 3.4B
• mRoPE sections: [16, 24, 24]

Usage

# CLI — auto-detects math OCR prompt from model name
./crispembed -m uni-mumer-qwen2.5-vl-3b-q4_k.gguf --ocr equation.png

# Server
./crispembed-server --ocr uni-mumer-qwen2.5-vl-3b-q4_k.gguf --port 8080
curl -X POST http://localhost:8080/math/ocr -F "[email protected]"

from crispembed import CrispMathOcr

ocr = CrispMathOcr("uni-mumer-qwen2.5-vl-3b-q4_k.gguf")
latex = ocr.recognize("equation.png")
print(latex)  # x ^ { 2 } + 2 x y + y ^ { 2 } = 0

Accuracy

• 82.25% ExpRate on CROHME (handwritten math benchmark)
• Highest accuracy among the Uni-MuMER variants
• Replaces CC BY-NC-SA models (PosFormer/BTTR/HMER at 57%) with Apache-2.0 licensed alternative

License

Apache-2.0 — same as the base model.

Credits

Original model by BFlameSwift/Uni-MuMER (phxember on HuggingFace). GGUF conversion and inference engine by CrispEmbed.

Provenance and EU AI Act Art. 53 note

• Upstream model: phxember/Uni-MuMER-Qwen2.5-VL-3B — published by phxember.
• 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/GGML). 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.