Gramscii-IT/dots.mocr-GGUF

dots.mocr GGUF

GGUF conversion of dots-studio/dots.mocr for llama.cpp. Built with dots.mocr.

Nothing about the model was changed: these are the published weights in a different container, plus quantized variants of that same conversion.

Provenance

• —Source: dots-studio/dots.mocr, revision `e539fbb52280393adc081b289ec597430a0f9031`, every file verified by sha256 before conversion.
• —Converter: llama.cpp convert_hf_to_gguf.py (master, 2026-08-21, commit 1719747), whose DotsOCRForCausalLM support landed with PR #17575. The language model converts as a qwen2-architecture GGUF; the vision tower converts separately as a dots_ocr multimodal projector.
• —Quantization: llama-quantize Q8_0 from the f16 conversion, in the same llama.cpp build.

Files

Usage

llama-server -m dots.mocr-q8_0.gguf --mmproj mmproj-dots.mocr-q8_0.gguf \
  --host 127.0.0.1 --port 8080

Requires a llama.cpp build that ships PROJECTOR_TYPE_DOTS_OCR (b8731 or later). If you are serving with vLLM or Transformers, use the original repository: these files buy you nothing there.

License and attribution

The model weights remain under the dots.mocr License Agreement, unchanged from the model these files were converted from; a copy travels with this repository, as the agreement requires, together with the upstream NOTICE. Built with dots.mocr.