praxis-briefing / app.py
jempf's picture
ZeroGPU attempt: boot-safe loader + @spaces.GPU realtime handler.
ca99e8a
Raw
History Blame Contribute Delete
1.02 kB
"""HF Spaces (ZeroGPU) entry point: Peitho 1.5B v6 German voice demo.
Background: the stock LFM2.5-Audio-1.5B base model is English-only for
speech-to-speech (see https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B —
"Supported languages: English"). Our v6 fine-tune is what teaches it German
audio output. A short A/B with the unmodified base model produced garbled
German + robotic audio — confirming the fine-tune is essential.
Model loading + the German v6 overlay live in `peitho_model.py`, which is
written to be boot-safe on ZeroGPU (no CUDA warmup at import). This entry
point just imports the model, then launches the chat demo.
"""
from __future__ import annotations
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent))
def main() -> None:
import peitho_model # noqa: F401 # triggers load + v6 overlay (boot-safe)
print("Importing chat demo and launching...")
import chat as chat_mod
chat_mod.demo.launch()
if __name__ == "__main__":
main()