---

license: other

license_name: glm-4

tags:

- ocr

- vlm

- document-understanding

- gguf

- crispembed

library_name: crispembed

base_model: THUDM/glm-edge-v-2b

---

GLM-OCR CrispEmbed GGUF

GLM-Edge-V 2B vision-language model converted to GGUF for OCR with CrispEmbed.

Models

| File | Quant | Size |

|------|-------|------|

| glm-ocr-f16.gguf | F16 | ~3.8 GB |

| glm-ocr-q8_0.gguf | Q8_0 | ~2.0 GB |

| glm-ocr-q4_k.gguf | Q4_K | ~1.1 GB |

Architecture

• Base: GLM-Edge-V 2B (THUDM, Apache-2.0)
• Vision: SigLIP vision encoder
• LLM: GLM-4 decoder (2B params)
• Task: Document OCR, scene text, handwriting

Usage

from crispembed import CrispOcrPipeline

ocr = CrispOcrPipeline(vlm_model="glm-ocr-q8_0.gguf")
text = ocr.recognize("document.png")

Original Model

THUDM/glm-edge-v-2b — GLM-Edge-V 2B, CogViT + GLM-0.5B, 8 languages.

License

Apache-2.0

Licence — GLM-4 Model Licence, not Apache-2.0 (corrected 2026-08-02)

This repository declared license: apache-2.0. The upstream checkpoint

THUDM/glm-edge-v-2b is released

under the GLM-4 Model Licence, which is not Apache-2.0 and carries use

restrictions Apache-2.0 does not. Converting the weights to GGUF changes their

numeric representation, not their licence, so declaring Apache-2.0 here granted

rights the upstream never granted.

Consult the upstream licence before using these weights; the terms that govern

them are THUDM's, not this repository's.