cstr/trocr-base-printed-GGUF

TrOCR-base Printed Text — GGUF

Text recognition model for CrispEmbed. Larger and more capable than trocr-small-printed. Recognizes printed text from cropped text-line images. Uses GPT-2 BPE tokenizer (50,265 tokens).

Architecture: BEiT encoder (12L, 768d, 12 heads) + TrOCR decoder (12L, 1024d, 16 heads). 333M parameters. Tied embeddings (lm_head = embed_tokens).

Source: microsoft/trocr-base-printed (MIT).

Model Variants

Variant	Size	Recognition quality
F32	1.3 GB	exact token match vs HuggingFace (greedy)
F16	639 MB	identical to F32
Q8_0	340 MB	identical to F32

Q4_K not tested — d_model=1024 should handle it better than small (256d), but Q8_0 is recommended for this model size.

Usage

Pair with cstr/dbnet-ic15-GGUF for end-to-end OCR.

crispembed --det dbnet-ic15-q4_k.gguf -m trocr-base-printed-q8_0.gguf --ocr document.png

License

MIT (same as microsoft/trocr-base-printed).

Provenance and EU AI Act Art. 53 note

• Upstream model: microsoft/trocr-base-printed — published by microsoft.
• Upstream licence: mit. 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.