cstr/trocr-small-handwritten-GGUF

TrOCR-small Handwritten Text — GGUF

Handwritten text recognition model for CrispEmbed. Recognizes handwritten text from cropped text-line images. Outputs mixed-case text (unlike trocr-small-printed which uppercases).

Architecture: DeiT-small encoder (12L, 384d, 6 heads) + TrOCR decoder (6L, 256d, 8 heads). XLM-R vocabulary (64,044 tokens). 61M parameters.

Source: microsoft/trocr-small-handwritten (MIT).

Model Variants

Variant	Size	Recognition quality
F32	235 MB	exact token match vs HuggingFace (greedy)
F16	118 MB	identical to F32
Q8_0	63 MB	identical to F32

Q4_K not provided — 256d decoder bottleneck is too narrow for 4-bit quantization.

Usage

# Single crop recognition
crispembed -m trocr-small-handwritten-q8_0.gguf --ocr crop.png

# Full pipeline with DBNet detection
crispembed --det dbnet-ic15-q4_k.gguf -m trocr-small-handwritten-q8_0.gguf --ocr scan.png

License

MIT (same as microsoft/trocr-small-handwritten).

Provenance and EU AI Act Art. 53 note

• Upstream model: microsoft/trocr-small-handwritten — 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.