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 -f https://download.pytorch.org/whl/cpu/torch_stable.html
-
-# Copy app code
-COPY app.py .
-
-# Expose port
-EXPOSE 7860
-
-CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
-

handler.py

@@ -0,0 +1,36 @@
+from typing import Dict, Any
+import torch
+from TTS.api import TTS
+
+
+class EndpointHandler:
+    def __init__(self, path=""):
+        # Load Coqui XTTS model (weights are downloaded automatically)
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        self.tts = TTS(
+            "tts_models/multilingual/multi-dataset/xtts_v2").to(device)
+        self.device = device
+
+    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        Expected input:
+        {
+            "text": "Hello world",
+            "speaker_wav": "path/to/voice.wav" (optional, for cloning),
+            "language": "en"
+        }
+        """
+        text = data.get("text", "")
+        speaker_wav = data.get("speaker_wav", None)
+        language = data.get("language", "en")
+
+        # Generate speech
+        output_path = "output.wav"
+        self.tts.tts_to_file(
+            text=text,
+            speaker_wav=speaker_wav,
+            language=language,
+            file_path=output_path
+        )
+
+        return {"audio": output_path}

huggingface.yaml

@@ -1,6 +1,4 @@
-build:
-  dockerfile: Dockerfile
-
+# huggingface.yaml
 resources:
   cpu: "2"
   memory: "8Gi"
@@ -9,4 +7,5 @@ scaling:
   min_replicas: 0
   max_replicas: 1
 
-entrypoint: uvicorn app:app --host 0.0.0.0 --port 7860
+# Tell HF to use handler.py
+entrypoint: handler.py

requirements.txt

@@ -2,7 +2,5 @@ fastapi==0.110.0
 uvicorn==0.29.0
 soundfile==0.12.1
 numpy==1.26.4
-torch==2.1.0+cpu
-torchaudio==2.1.0+cpu
 transformers==4.38.1
 TTS==0.21.1