olmOCR-2-7B-1025 — CrispEmbed GGUF

(Apache-2.0, a Qwen2.5-VL-7B document-OCR fine-tune) for the

CrispEmbed native runtime.

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

| olmocr-2-7b-q8_0.gguf | Q8_0 | 8.4 GiB | norms/bias/embeddings F32 |

| olmocr-2-7b-q4_k.gguf | Q4_K | 5.3 GiB | vision kept at Q8_0 floor |

Converted with models/convert-qwen2vl-to-gguf.py (F16 intermediate) and

crispembed-quantize.

Usage

crispembed --ocr-pipeline page.png --ocr-engine olmocr

The runtime auto-detects the olmOCR contract from the model name: the exact

no-anchoring fine-tune prompt, text-before-image user turn, longest-side-1288

input scaling, and 8000-token output budget. The model answers in markdown

with a YAML front-matter block (language / rotation / table / diagram flags);

CrispEmbed strips it by default (CRISPEMBED_OLMOCR_RAW=1 keeps the raw

output).

All model weights derive from allenai/olmOCR-2-7B-1025 (Ai2), Apache-2.0.

See the upstream model card for training details and the olmOCR toolkit.

Question 2

What license applies to cstr/olmOCR-2-7B-1025-GGUF?

Accepted Answer

License: apache-2.0. Verify terms on Hugging Face before commercial use.

Question 3

How do I run cstr/olmOCR-2-7B-1025-GGUF locally?

Accepted Answer

Download a GGUF file from this page and load it in guIDE or llama.cpp. Pipeline task: text-generation.

Question 4

How much VRAM or disk space does cstr/olmOCR-2-7B-1025-GGUF need?

Accepted Answer

Runs locally from ~5.34 GB disk (8 GB VRAM class GPUs with llama.cpp / guIDE).

File	Type	Quantization	Size	Link
olmocr-2-7b-q4_k.gguf	GGUF	Q4_K	5.34 GB	Download
olmocr-2-7b-q8_0.gguf	GGUF	Q8_0	8.38 GB	Download