Apppublic

ek-developers/my-ocr-api

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes

App README

🔍 DeepSeek OCR API

A self-hosted OCR API powered by DeepSeek-OCR via Ollama, deployed on Hugging Face Spaces using Docker. Upload an image and receive extracted text — with support for custom prompts.

🚀 Live Demo

https://ek-developers-my-ocr-api.hf.space

Interactive API docs available at:

https://ek-developers-my-ocr-api.hf.space/docs

📁 Project Structure

├── Dockerfile        # Container setup with Ollama + Python dependencies
├── start.sh          # Runtime script: installs Ollama, pulls model, starts API
├── app.py            # FastAPI application
└── README.md         # Documentation

🛠️ Tech Stack

Component	Technology
OCR Model	`deepseek-ocr` via Ollama
API Framework	FastAPI
Server	Uvicorn
Container	Docker (Ubuntu 22.04)
Hosting	Hugging Face Spaces (Free CPU)

📡 API Reference

`GET /`

Health check — confirms the API is running.

Response

json

{
  "message": "DeepSeek OCR API running. POST an image to /ocr"
}

`GET /health`

Lightweight health probe.

Response

json

{
  "status": "ok"
}

`POST /ocr`

Upload an image and extract text from it.

Request — multipart/form-data

Field	Type	Required	Default	Description
`file`	image file	✅	—	The image to process (jpg, png, etc.)
`prompt_text`	string	❌	`Free OCR.`	Custom instruction passed to the model

Example — default prompt

bash

curl -X POST https://ek-developers-my-ocr-api.hf.space/ocr \
  -F "file=@receipt.jpg"

Example — custom prompt

bash

curl -X POST https://ek-developers-my-ocr-api.hf.space/ocr \
  -F "file=@menu.jpg" \
  -F "prompt_text=Extract all menu items and prices."

Response

json

{
  "text": "Extracted text from the image..."
}

Error Responses

Status	Meaning
`400`	File is not an image
`504`	Model timed out (>120s)
`500`	Internal server error

⚙️ Local Development

Prerequisites

—Ollama installed locally
—Python 3.10+

Setup

bash

# 1. Pull the model
ollama pull deepseek-ocr

# 2. Install Python dependencies
pip install fastapi uvicorn requests pydantic python-multipart

# 3. Run the API
uvicorn app:app --host 0.0.0.0 --port 7860 --reload

Test locally

bash

curl -X POST http://localhost:7860/ocr \
  -F "file=@image.jpg" \
  -F "prompt_text=Free OCR."

🐳 Docker

Build and run locally

bash

docker build -t deepseek-ocr-api .
docker run -p 7860:7860 deepseek-ocr-api

⚠️ Performance Notes

This Space runs on free-tier CPU hardware (16GB RAM, no GPU).

Condition	Expected Response Time
First request (cold start)	2–3 minutes (model loading)
Subsequent requests (warm)	30–60 seconds
After inactivity (Space sleeps)	Cold start again

To reduce latency, upgrade the Space hardware to a GPU instance in Settings → Hardware.

📝 Example Use Cases

—Extract text from receipts or invoices
—Read menu items from restaurant photos
—Parse printed forms or documents
—Digitize handwritten or printed notes

📄 License

MIT License — free to use and modify.