SalamBadak/fufufafa-chat-bot

Clean up model list to natively supported Qwen 7B and 72B

Use native Qwen models to bypass Hugging Face provider restrictions

Fix timeout parameter location in hf_client initialization

Optimize ZeroGPU quota usage by separating API calls from GPU decorator

Implement sequential multi-stage reranker queue with 5s timeouts

Fix response_format type to json_object for HF API compatibility

Implement HF Inference API with local Cross-Encoder fallback

Decorate predict function with @spaces.GPU for ZeroGPU compliance

Import spaces first to prevent ZeroGPU CUDA initialization error

Add robust path resolution for nested data folder

Remove sdk_version, gradio and huggingface_hub dependencies to use HF pre-installed defaults

Rename requirement.txt to requirements.txt for HF auto-installation

Switch SDK to Gradio and remove Docker files

Prep codebase deployment with datasets integration