Refactored to use official huggingface_hub methods: chat_completion, translation, and feature_extraction

Changed .post to ._inner_post to fix InferenceClient attribute error on new huggingface_hub version

Bumped huggingface_hub to >=0.25.0 to ensure InferenceClient.post() is available

Refactored API calls to use huggingface_hub.InferenceClient to bypass HF Space DNS sandbox blocking

Switched to Qwen 2.5 7B Instruct (best free model) and ensured requests is used

Refactored all Hugging Face Inference API calls from urllib to requests to fix DNS Errno -5 bug on Docker

Switched default LLM to Zephyr to bypass HF Pro restrictions

Wrapped MongoDB connection in try/except block to allow startup without database credentials

Completely neutralised check_imports to fix flash_attn bug

Patched check_imports to bypass flash_attn requirement for Florence-2

Removed flash_attn to fix build

Added einops, timm, flash_attn to requirements.txt

Add auto-retry logic for 503 Model Loading errors

Migrate heavy ML models to HF Inference API

Deploy FastAPI Backend to HF Space