Check deployed models and upload local model

[uelkerd]

Enable custom model deployment by adding a script to upload trained models to HuggingFace Hub and updating deployment configurations to use them.

Previously, the model-as-a-service deployment was configured to fetch models externally but was falling back to generic base models (distilroberta-base, bert-base-uncased) because the user's custom-trained models were not available on HuggingFace Hub. This PR provides a complete solution to upload these custom models and ensures the deployment uses them for improved accuracy and specialized emotion detection, prioritizing a user-specified local directory for model discovery.

---

Summary by Sourcery

Enable end-to-end custom model deployment by adding a script and documentation to upload and integrate trained emotion detection models with the existing deployment pipeline

New Features:

• Add script to find, convert, and upload custom-trained emotion detection models to HuggingFace Hub
• Introduce automated deployment config updates to point to uploaded custom models

Bug Fixes:

• Fix model-as-a-service fallback to generic base models by ensuring custom models are uploaded and used

Enhancements:

• Support loading of 12 specialized emotion labels in deployment
• Generate model cards, requirements, and handle PyTorch to HuggingFace format conversion automatically

Deployment:

• Switch deployment infrastructure to pull custom models from HuggingFace Hub instead of local base models

Documentation:

• Add comprehensive custom model deployment guide
• Add README for deployment/models directory outlining expected model structure

Summary by CodeRabbit

• New Features

  • Introduced an automated script for uploading custom-trained emotion detection models to the HuggingFace Hub, streamlining production deployment.
  • Deployment pipeline now supports direct loading of custom models from HuggingFace Hub and automates model conversion, configuration updates, and model card generation.
  • Added a flexible API server supporting serverless, endpoint, and self-hosted deployment modes with unified emotion detection responses.

• Documentation

  • Added a comprehensive deployment guide detailing steps for uploading, configuring, and deploying custom emotion detection models.
  • Included a new README for the models directory, clarifying model management and deployment procedures.
  • Added a detailed deployment checklist for HuggingFace models covering preparation, testing, security, and troubleshooting.
  • Provided example environment configuration files and improvement summaries for deployment scripts.
  • Added multiple detailed code review and security fix documents addressing authentication, error handling, secure server binding, and best practices.
  • Updated changelog with details about the new deployment solution, model improvements, and production changes.

• Bug Fixes

  • Fixed an issue where deployment defaulted to untrained base models instead of intended custom-trained models.
  • Resolved a critical bug preventing safe handling of empty model parameters in the API server.

• Chores

  • Updated ignore rules to ensure the models directory README is tracked.
  • Applied code style fixes to eliminate unused variable warnings across deployment scripts.