cstr/german-ocr-3.1-crispembed-GGUF

german-ocr-3.1 — CrispEmbed GGUF

German-OCR-3.1 (Qwen2.5-VL fine-tune, specialised for German business documents: invoices, forms, receipts) converted to the single-file CrispEmbed GGUF layout, for the qwen2vl_ocr engine.

Built from the upstream llama.cpp split GGUFs (german-ocr-3.1-F16.gguf + mmproj-german-ocr-3.1-F16.gguf) via models/merge-llamacpp-qwen2vl-gguf.py, then quantized with crispembed-quantize. CrispEmbed uses one combined file with its own tensor naming (llm.layers.N.attn.q, vis.blocks.N), not llama.cpp's LLM+mmproj split — these files are not interchangeable with llama.cpp.

File	Size	Notes
german-ocr-3.1-q4_k.gguf	1684 MB	391 tensors quantized, 466 kept

Usage

crispembed -m german-ocr-3.1 --ocr document.png     # auto-downloads
crispembed -m german-ocr-3.1-q4_k.gguf --ocr document.png

Attribution & licence

Upstream model © its authors, Apache-2.0 — see keyvan-ai/german-ocr-3.1 (also mirrored at Keyven/german-ocr-3.1; the GGUFs in both are byte-identical). Quantization does not relicense the model. See CrispEmbed and its POLICY.md for intended purpose and acceptable use — in particular, OCR output is a probabilistic reconstruction, not a faithful copy.

Provenance and EU AI Act Art. 53 note

• Upstream model: keyvan-ai/german-ocr-3.1 — published by keyvan-ai.
• 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.