Switch to handler.py deployment

Dockerfile

@@ -1,28 +0,0 @@
-FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
-
-# Prevent interactive tzdata prompts and accept Coqui TOS
-ENV DEBIAN_FRONTEND=noninteractive
-ENV TZ=Etc/UTC
-ENV COQUI_TOS_AGREED=1
-
-# Install system dependencies
-RUN apt-get update && \
-    apt-get install -y git ffmpeg libsndfile1 tzdata && \
-    rm -rf /var/lib/apt/lists/*
-
-# Create working directory
-WORKDIR /app
-
-# Install Python dependencies
-COPY requirements.txt .
-
-RUN pip install --ignore-installed --no-cache-dir -r requirements.txt
-
-# Copy app code
-COPY app.py .
-
-# Expose port
-EXPOSE 7860
-
-# Start FastAPI server
-CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -1,24 +0,0 @@
-from fastapi import FastAPI, UploadFile, Form
-from fastapi.responses import FileResponse
-import uuid, os, soundfile as sf
-from TTS.api import TTS
-
-app = FastAPI()
-
-# Load XTTS model (multi-lingual, multi-speaker)
-# This will be slow on cold start but stays cached in memory
-tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to("cuda")
-
-@app.post("/generate")
-async def generate(text: str = Form(...), speaker_wav: UploadFile = None):
-    output_path = f"/tmp/{uuid.uuid4().hex}.wav"
-    
-    if speaker_wav:
-        wav_path = f"/tmp/{speaker_wav.filename}"
-        with open(wav_path, "wb") as f:
-            f.write(await speaker_wav.read())
-        tts.tts_to_file(text=text, speaker_wav=wav_path, file_path=output_path)
-    else:
-        tts.tts_to_file(text=text, speaker="random", file_path=output_path)
-
-    return FileResponse(output_path, media_type="audio/wav")

handler.py

@@ -0,0 +1,27 @@
+from typing import Dict
+from TTS.api import TTS
+import uuid
+import os
+
+
+class EndpointHandler:
+    def __init__(self, path=""):
+        # Load XTTS model
+        self.tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2")
+
+    def __call__(self, data: Dict[str, str]) -> Dict[str, str]:
+        # Input text
+        text = data.get("text", "")
+        speaker_wav = data.get("speaker_wav", None)
+
+        # Output path
+        output_path = f"/tmp/{uuid.uuid4().hex}.wav"
+
+        if speaker_wav:
+            self.tts.tts_to_file(
+                text=text, speaker_wav=speaker_wav, file_path=output_path)
+        else:
+            self.tts.tts_to_file(
+                text=text, speaker="random", file_path=output_path)
+
+        return {"audio_file": output_path}

huggingface.yaml

@@ -1,15 +0,0 @@
-# Hugging Face Inference Endpoint config
-build:
-  dockerfile: Dockerfile
-
-resources:
-  # Use CPU instance for free/low-cost deployment
-  cpu: "2"
-  memory: "8Gi"
-
-scaling:
-  # Scale-to-zero when idle → no charges while unused
-  min_replicas: 0
-  max_replicas: 1
-
-entrypoint: uvicorn app:app --host 0.0.0.0 --port 7860

requirements.txt

@@ -1,8 +1,6 @@
-fastapi==0.110.0
-uvicorn==0.29.0
-soundfile==0.12.1
-numpy==1.26.4
 torch==2.1.0
 torchaudio==2.1.0
 transformers==4.38.1
 TTS==0.21.1
+soundfile==0.12.1
+numpy==1.26.4