Revert: remove COPY model files, use huggingface_hub download instead

Fix: COPY model files into image instead of downloading at runtime

Fix Space boot and wire up TTS UI

v4: move model downloads to runtime, simplify build

Fix v2: proper build-essential, minimal deps, CPU config

Fix: CPU-only Dockerfile with pretrained models, proper start.sh

wget tarball instead of git clone

pure CPU build, no CUDA deps

Fix: force reinstall torchaudio for CUDA 12.6 at build time using conda pip

Fix: reinstall torchaudio for CUDA 12.6 to match PyTorch

Use xxxxrt666/gpt-sovits:latest-cu126-lite official image

use official gpt-sovits docker image

fix: add gradio dep

connect api_v2.py inference

auto download models on startup

switch to FastAPI

fix Dockerfile

fix: simplify build, download models on runtime

fix: pin huggingface_hub==0.26.5 for Gradio compatibility

add Dockerfile