AI Content Detection Service (E5 Small LoRA)

Flask microservice exposing MayZhou/e5-small-lora-ai-generated-detector via a simple HTTP API.

Files

• e5-small-lora.py: the Flask app exposing /health and /classify.
• test_e5_client.py: small client to send text and print AI/Human scores.

Environment

Python packages

pip install flask torch transformers huggingface_hub requests

Optional environment variables

• HF_TOKEN — if the model requires authentication.
• AI_SCORE_API_URL — not used directly here; consumed by the Similarity service.

Model cache

• First run downloads to models/e5_small_lora_ai_detector/ at repository root.

API

• GET /health

  • Response: { "status": "ok", "device": "cpu|cuda", "model_dir": "..." }

• POST /classify

  • Request: { "text": "..." }
  • Response: { "ai_score": 0..1 } where 1.0 = very likely AI-generated.

Run

python e5-small-lora.py
# Service listens on 0.0.0.0:5001

Test

python test_e5_client.py --url http://localhost:5001/classify --text "Sample text"