feat: switch demo to lighter GGUF (llama.cpp) — 1.06GB vs 3.3GB fp16; add PyPI link

remove device_map=auto, remove accelerate, fix repetition_penalty=1.1

use fused model (eulogik/Bharat-Tiny-LLM-fused) instead of base+peft; add eos_token_id=<|im_end|>; drop peft/bitsandbytes

pin starlette==0.51.0 (pre-api-break), keep gradio 4.44.1 + gradio_client monkey-patch

Switch to Python 3.12 - fixes all compat issues

Patch gradio_client source for Python 3.13 compat

Fix gradio_client JSON schema crash on Python 3.13

Add requests dependency

Pin gradio 4.44.1 + huggingface_hub<1.22

Add requirements.txt with Dockerfile for pre-cached model

Delete files * with huggingface_hub

Fix: add HfFolder shim for gradio 4.44 compat

Fix: pin huggingface_hub for gradio compat

Fix Python 3.13 audioop compat

Fix: add pyaudioop for Python 3.13 compatibility

Initial space: Bharat-Tiny-LLM Gradio demo