Tesseract Lstm GGUF
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,...
Source du modèle
- Type
- Other
- Taille du fichier
- 279 KB–7,38 MB
- Version
- vRevision e5ec906
- Mis à jour
- 23 oct. 2026
- Téléchargements
- 8,4 k
- Source
- Hugging Face
Description de la source
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
- Upstream model: tesseract-ocr/tessdata.
- 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.
- 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.
Sources
1 source
Sources
1 sourceVérifié 23 oct.
Artefacts du modèle
48 artefacts
Artefacts du modèle
48 artefactsExtraits de sources
3 extraits
Extraits de sources
3 extraits--- license: apache-2.0 language: - en - de - fr - es - it - pt - nl - ru - ar - zh - ja - ko tags: - gguf - ocr - tesseract - lstm - crispembed library_name: crispembed pipeline_tag: image-to-text --- # 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 ```bash 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 ```bash 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 - **Upstream model:** tesseract-ocr/tessdata. - **Upstream licence:** `apache-2.0`. This repository redistributes under the same terms; it grants no rights the upstream licence does not. - **What was done...
Source context: 8418 downloads · 0 likes · Pipeline image-to-text · Library crispembed · Repo cstr/tesseract-lstm-GGUF
Source context: 2205 downloads · 0 likes · Pipeline image-to-text · Library crispembed · Repo cstr/tesseract-lstm-GGUF
