feat: DeBERTa emotion detection API deployment

[uelkerd]

🚀 DeBERTa Emotion Detection API Deployment

SCOPE DECLARATION

ALLOWED: DeBERTa model integration and Cloud Run deployment
FORBIDDEN: Other model architectures, data preprocessing changes, training script modifications
FILES TOUCHED: 4 files (deployment scripts, Dockerfile, API server)
TIME ESTIMATE: 4 hours

🎯 What This PR Does

Deploys a production-ready DeBERTa emotion detection API to Google Cloud Run with 28 emotion classes and comprehensive security features.

✅ Key Features

• 28 Emotion Classes: vs 6 in production (excitement, joy, sadness, anger, fear, etc.)
• High Accuracy: 51.8% F1 Macro score, 93%+ confidence on test cases
• Production Ready: Security headers, rate limiting, input validation
• Cloud Run Optimized: AMD64 platform, 4GB memory, auto-scaling
• API Documentation: RESTful endpoints with proper error handling

🌐 Live API Endpoints

Service URL: https://samo-emotion-deberta-71517823771.us-central1.run.app

• GET /api/health - Health check and model status
• POST /api/predict - Single text emotion prediction
• GET /api/emotions - List available emotion classes
• GET /admin/model/status - Model performance metrics

🧪 Testing Results

✅ Basic Emotions: Happy (83% excitement), Sad (96% sadness), Angry (90% anger), Fearful (95% fear)
✅ Error Handling: Proper validation for empty text, invalid inputs
✅ Performance: Sub-2 second response times
✅ Security: API key authentication, rate limiting enabled

🚀 Deployment Status

LIVE AND OPERATIONAL - Ready for production use!

Summary by Sourcery

Integrate a production-grade 28-class DeBERTa emotion detection model into the API ecosystem, extend and deploy both Flask-RESTX and FastAPI servers with new endpoints and configuration, and provide Docker-based Cloud Run deployment tooling alongside comprehensive tests and documentation.

New Features:

• Add SAMOBERTEmotionClassifier module for multi-label emotion detection with temperature scaling and utility functions.
• Introduce unified FastAPI server offering T5 summarization, Whisper transcription, BERT emotion detection, health checks, and combined audio processing pipelines.
• Create emotion_labels module defining the full 28-class GoEmotions taxonomy and lookup utilities.
• Add configuration files for emotion detection and unified API server settings.

Enhancements:

• Update Flask-RESTX secure API server to register DeBERTa endpoints and configure model environment variables.
• Include startup script for launching the unified API server with CLI options.

Build:

• Add Dockerfile.deberta for building the DeBERTa emotion API image.

Deployment:

• Provide deploy_deberta.sh script to automate Google Cloud Run deployment with security headers, rate limiting, and autoscaling settings.

Documentation:

• Add DEBERTA_DEPLOYMENT_README.md outlining deployment steps, performance metrics, and API compatibility.

Tests:

• Add comprehensive tests for unified API server endpoints and standalone emotion detection model functionality.

Summary by CodeRabbit

• New Features

  • Unified API with health, single/batch prediction, summarize, transcribe, and combined audio→summary+emotions; emotion detection now multi-label, calibrated, and returns 28 emotions.

• Documentation

  • New DeBERTa deployment guide with setup, pinned dependency notes, Cloud Run deployment steps, and validation examples.

• Bug Fixes

  • Compatibility and security fixes (protobuf/safetensors handling) to stabilize model loading.

• Chores

  • Centralized configs and modernized ML-focused dependencies.

• Tests

  • Expanded end-to-end, integration, benchmarking, and deployment test/monitoring tools.