Marketplace · Text & chat
richardyoung/olmocr2:7b-q8
Call richardyoung/olmocr2:7b-q8 through one OpenAI-compatible API. It's served by 1 host on the Inpherio grid, each at a price it sets — your request goes to the cheapest healthy one, and fails over to the next if it's down.
At a glance.
vision ocr
from openai import OpenAI
client = OpenAI(base_url="https://inpherio.co.uk/v1",
api_key=INPHERIO_API_KEY)
reply = client.chat.completions.create(
model="richardyoung/olmocr2:7b-q8",
messages=[{"role": "user", "content": "Hello"}],
)richardyoung/olmocr2:7b-q8 in detail.
State-of-the-art OCR (Optical Character Recognition) vision language model based on [allenai/olmOCR-2-7B-1025](https://huggingface.co/allenai/olmOCR-2-7B-1025).
- Published by
- allenai
- Parameters
- 8.3B
- Download size
- 9.5 GB (as packaged for Ollama)
- Quantisation
- Q8_0
- Format
- gguf
- Architecture
- qwen2_5_vl
- Licence
- Apache 2.0
- Based on
- Qwen/Qwen2.5-VL-7B-Instruct
- Task
- image text to text
- Library
- transformers
- Languages
- en
- First published
- 6 Oct 2025
- Last updated
- 22 Oct 2025
- Downloads (30 days)
- 78,970 on Hugging Face
- Likes
- 161
qwen2_5_vl image-text-to-text en
The model card
Published by the model's authors on Hugging Face — shown as they wrote it, unedited by Inpherio.
olmOCR-2-7B-1025
Full BF16 version of olmOCR-2-7B-1025-FP8. We recommend using the FP8 version for all practical purposes except further fine tuning.
This is a release of the olmOCR model that's fine tuned from Qwen2.5-VL-7B-Instruct using the olmOCR-mix-1025 dataset. It has been additionally fine tuned using GRPO RL training to boost its performance at math equations, tables, and other tricky OCR cases.
Quick links:
- 📃 Paper
- 🤗 SFT Dataset
- 🤗 RL Dataset
- 🛠️ Code
- 🎮 Demo
The best way to use this model is via the olmOCR toolkit. The toolkit comes with an efficient inference setup via VLLM that can handle millions of documents at scale.
olmOCR-Bench Scores
This model scores the following scores on olmOCR-bench when used with the olmOCR toolkit toolkit which automatically renders, rotates, and retries pages as needed.
Model ArXiv Old Scans Math Tables Old Scans Headers and Footers Multi column Long tiny text Base Overall
olmOCR pipeline v0.4.0 with olmOCR-2-7B-1025 82.9 82.1 84.3 48.3 95.7 84.3 81.4 99.7 82.3 ± 1.1
olmOCR pipeline v0.4.0 with olmOCR-2-7B-1025-FP8 83.0 82.3 84.9 47.7 96.1 83.7 81.9 99.7 82.4 ± 1.1
Usage
This model expects as input a single document image, rendered such that the longest dimension is 1288 pixels.
The prompt must then contain the additional metadata from the document, and the easiest way to generate this is to use the methods provided by the olmOCR toolkit.
Manual Prompting
If you want to prompt this model manually instead of using the olmOCR toolkit, please see the code below.
In normal usage, the olmOCR toolkit builds the prompt by rendering the PDF page, and extracting relevant text blocks and image metadata. To duplicate that you will need to
pip install olmocr>=0.4.0
and then run the following sample code.
import torch
import base64
import urllib.request
from io import BytesIO
from PIL import Image
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
from olmocr.data.renderpdf import render_pdf_to_base64png
from olmocr.prompts import build_no_anchoring_v4_yaml_prompt
### Initialize the model
model = Qwen2_5_VLForConditionalGeneration.from_pretrained("allenai/olmOCR-2-7B-1025", torch_dtype=torch.bfloat16).eval()
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
### Grab a sample PDF
urllib.request.urlretrieve("https://olmocr.allenai.org/papers/olmocr.pdf", "./paper.pdf")
### Render page 1 to an image
image_base64 = render_pdf_to_base64png("./paper.pdf", 1, target_longest_image_dim=1288)
### Build the full prompt
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": build_no_anchoring_v4_yaml_prompt()},
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{image_base64}"}},
],
}
]
Read the rest of the model card
Apply the chat template and processor
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) main_image = Image.open(BytesIO(base64.b64decode(image_base64)))
inputs = processor( text=[text], images=[main_image], padding=True, return_tensors="pt", ) inputs = {key: value.to(device) for (key, value) in inputs.items()}
Generate the output
output = model.generate( **inputs, temperature=0.1, max_new_tokens=50, num_return_sequences=1, do_sample=True, )
Decode the output
prompt_length = inputs["input_ids"].shape[1] new_tokens = output[:, prompt_length:] text_output = processor.tokenizer.batch_decode( new_tokens, skip_special_tokens=True )
print(text_output)
['---\nprimary_language: en\nis_rotation_valid: True\nrotation_correction: 0\nis_table: False\nis_diagram: False\n---\nolmOCR: Unlocking Trillions of Tokens in PDFs with Vision Language Models\n\nJake Poz']
### License and use
This model is licensed under Apache 2.0. It is intended for research and educational use in accordance with Ai2's [Responsible Use Guidelines](https://allenai.org/responsible-use).Sources: allenai/olmOCR-2-7B-1025 on Hugging Face · richardyoung/olmocr2:7b-q8 on Ollama. Collected from public sources on 8 Nov 2026; the host serving it may run a different build or quantisation — the marketplace shows each host's own measurements.
Who's serving it.
Every host benchmarked this model on its own hardware before serving it. Prices are set by each host; health is scored by the platform from what it observes.
| Host | Tier | Price | Measured speed | Health | State | Rent |
|---|---|---|---|---|---|---|
| GTX 1080 Ti (11G) | L1 | £0.072 in · £0.15 out | 3 tok/s · 24452 ms first token | 78 | loads on request | £0.60/h |
Live data, refreshed on every visit. How billing works.
Room on the grid for your machine.
Right now one machine serves richardyoung/olmocr2:7b-q8. If it drops, so does every request for it — a second host is what turns it into failover, and the grid needs that more than another copy. The router weighs price, measured speed, health and whether the model is already loaded — so a host with a sharper price or quicker hardware wins the next call.
Set your own price — you keep 85% of what it earns, and new hosts start with 12 months fee-free (up to £250 of earnings). A payout fee of £2 + 1% per payout to a UK bank account, £2 + 2% to any other country applies when you're paid. Install the connector, load richardyoung/olmocr2:7b-q8, and the platform benchmarks it on your hardware before your first request.