Tesseract LSTM OCR — GGUF models for CrispEmbed

Apache-2.0 Tesseract LSTM line-recognition models converted to GGUF for CrispEmbed. The repository contains English, German, French, Spanish, Italian, Portuguese, Dutch, Russian, Arabic, Simplified Chinese, Japanese, and Korean variants.

These are line recognizers: normalize and crop individual text lines before recognition, or pair them with a detector such as DBNet for full-page OCR.

Precision variants

Each language has an F32 reference, an F16 deployment model, and Q8_0/Q4_K variants. The quantized models preserve output.weight and output.bias at the source precision (F16 for the multilingual F16-derived models; F32 for German regenerated from its original .traineddata). Only recurrent matrices are quantized. This preserves the CTC decision boundary and keeps critical character logits stable.

Variant	Purpose
`*-f32.gguf`	Canonical reference and parity baseline
`*-f16.gguf`	High-fidelity deployment
`*-q8_0.gguf`	Recommended compact deployment
`*-q4_k.gguf`	Smallest deployment; validate on the target corpus

Usage

crispembed -m tesseract-eng-q8_0.gguf --ocr line.png

The native runtime performs height normalization and CTC greedy decoding. Word spacing and page reading order must be supplied by the surrounding OCR pipeline; the Tesseract DAWG language models are not part of this GGUF graph.

Conversion

python models/convert-tesseract-to-gguf.py \
  --model eng.traineddata --output tesseract-eng-f32.gguf
crispembed-quantize tesseract-eng-f32.gguf tesseract-eng-q8_0.gguf q8_0
crispembed-quantize tesseract-eng-f32.gguf tesseract-eng-q4_k.gguf q4_k

The exact source URL/revision and SHA-256 are stored in each GGUF's metadata.

License

Apache-2.0, following the upstream Tesseract language data. Preserve the upstream attribution and source checksum when redistributing derivatives.

Provenance and EU AI Act Art. 53 note