johnlockejrr/hayyim-deepseek-ocr

Document OCR using DeepSeek-OCR

This dataset contains markdown-formatted OCR results from images in johnlockejrr/hayyim using DeepSeek-OCR via vLLM API.

Processing Details

• Source Dataset: johnlockejrr/hayyim
• Model: deepseek-ai/DeepSeek-OCR
• Number of Samples: 140
• Processing Time: 122.1 min
• Processing Date: 2025-12-01 11:53 UTC
• API Endpoint: http://127.0.0.1:8000/v1/chat/completions

Configuration

• Image Column: image
• Output Column: markdown
• Dataset Split: train
• Batch Size: 8
• Resolution Mode: base
• Base Size: 1024
• Image Size: 1024
• Crop Mode: False
• Max Output Tokens: 4,096

Model Information

DeepSeek-OCR is a state-of-the-art document OCR model that excels at:

• 📐 LaTeX equations - Mathematical formulas preserved in LaTeX format
• 📊 Tables - Extracted and formatted as HTML/markdown
• 📝 Document structure - Headers, lists, and formatting maintained
• 🖼️ Image grounding - Spatial layout and bounding box information
• 🔍 Complex layouts - Multi-column and hierarchical structures
• 🌍 Multilingual - Supports multiple languages

Resolution Modes

• Tiny (512×512): Fast processing, 64 vision tokens
• Small (640×640): Balanced speed/quality, 100 vision tokens
• Base (1024×1024): High quality, 256 vision tokens
• Large (1280×1280): Maximum quality, 400 vision tokens
• Gundam (dynamic): Adaptive multi-tile processing for large documents

Dataset Structure

The dataset contains all original columns plus:

• markdown: The extracted text in markdown format with preserved structure
• inference_info: JSON list tracking all OCR models applied to this dataset

Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("{{output_dataset_id}}", split="train")

# Access the markdown text
for example in dataset:
    print(example["markdown"])
    break

# View all OCR models applied to this dataset
inference_info = json.loads(dataset[0]["inference_info"])
for info in inference_info:
    print(f"Column: {{info['column_name']}} - Model: {{info['model_id']}}")

Performance

• Processing Speed: ~0.0 images/second
• Processing Method: vLLM API client

Generated with 🤖 DeepSeek-OCR vLLM API Client