ShlokArora2709/jester

FastAPI inference server for CLIP image/text embeddings plus OCR.

Required model files

Place these under HF_server/model (or mount at runtime):

• —clip_image_encoder_int8.onnx
• —clip_text_encoder.onnx (you will add)

Environment variables

• —IMAGE_MODEL_PATH (default: /model/clip_image_encoder_int8.onnx)
• —TEXT_MODEL_PATH (default: /model/clip_text_encoder_int8.onnx)
• —TOKENIZER_NAME (default: openai/clip-vit-base-patch32)
• —MAX_IMAGE_BYTES (default: 8388608)
• —OCR_LANG (default: eng)
• —ORT_PROVIDERS (default: CPUExecutionProvider)
• —API_KEY (default: empty, disables auth)
• —API_HEADER (default: X-API-Key)

Local run

python -m venv .venv
source .venv/bin/activate
pip install -r HF_server/requirements.txt
uvicorn HF_server.app:app --host 0.0.0.0 --port 7860

Docker (Hugging Face Spaces)

From the repo root:

docker build -f HF_server/Dockerfile -t hf-server .
docker run -p 7860:7860 -v $(pwd)/HF_server/model:/model hf-server

API

GET /

Health check.

POST /process

Request:

{ "image_b64": "..." }

Auth: set X-API-Key: <API_KEY> header (or Authorization: Bearer <API_KEY>). Response:

{ "embedding": [0.0, 0.1], "ocr_text": "example" }

POST /embed_text

Request:

{ "text": "hello world" }

Auth: set X-API-Key: <API_KEY> header (or Authorization: Bearer <API_KEY>). Response:

{ "embedding": [0.0, 0.1] }