fix: update model directory paths and improve logging in TTS services

api/src/core/config.py

@@ -13,7 +13,7 @@ class Settings(BaseSettings):
     output_dir: str = "output"
     output_dir_size_limit_mb: float = 500.0  # Maximum size of output directory in MB
     default_voice: str = "af"
-    model_dir: str = "/app/api/model_files"  # Base directory for model files
+    model_dir: str = "/app/models"  # Base directory for model files
     pytorch_model_path: str = "kokoro-v0_19.pth"
     onnx_model_path: str = "kokoro-v0_19.onnx"
     voices_dir: str = "voices"

api/src/services/tts_base.py

@@ -39,6 +39,9 @@ class TTSBaseModel(ABC):
                 cls._device = "cpu"
                 model_path = os.path.join(settings.model_dir, settings.onnx_model_path)
             logger.info(f"Initializing model on {cls._device}")
+            logger.info(f"Model dir: {settings.model_dir}")
+            logger.info(f"Model path: {model_path}")
+            logger.info(f"Files in model dir: {os.listdir(settings.model_dir)}")
 
             # Initialize model first
             model = cls.initialize(settings.model_dir, model_path=model_path)

api/src/services/tts_gpu.py

@@ -3,7 +3,7 @@ import time
 
 import numpy as np
 import torch
-from builds.models import build_model
+from ..builds.models import build_model
 from loguru import logger
 
 from ..core.config import settings

docker/cpu/Dockerfile

@@ -16,17 +16,21 @@ COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
 RUN useradd -m -u 1000 appuser
 
 # Create directories and set ownership
-RUN mkdir -p /app/api/model_files && \
+RUN mkdir -p /app/models && \
     mkdir -p /app/api/src/voices && \
     chown -R appuser:appuser /app
 
 USER appuser
 
 # Download and extract models
-WORKDIR /app/api/model_files
-RUN curl -L -o model.tar.gz https://github.com/remsky/Kokoro-FastAPI/releases/download/v0.0.1/kokoro-82m-onnx.tar.gz && \
+WORKDIR /app/models
+RUN set -x && \
+    curl -L -o model.tar.gz https://github.com/remsky/Kokoro-FastAPI/releases/download/v0.0.1/kokoro-82m-onnx.tar.gz && \
+    echo "Downloaded model.tar.gz:" && ls -lh model.tar.gz && \
     tar xzf model.tar.gz && \
-    rm model.tar.gz
+    echo "Contents after extraction:" && ls -lhR && \
+    rm model.tar.gz && \
+    echo "Final contents:" && ls -lhR
 
 # Download and extract voice models
 WORKDIR /app/api/src/voices
@@ -54,7 +58,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 
 # Set environment variables
 ENV PYTHONUNBUFFERED=1
-ENV PYTHONPATH=/app:/app/Kokoro-82M
+ENV PYTHONPATH=/app:/app/models
 ENV PATH="/app/.venv/bin:$PATH"
 ENV UV_LINK_MODE=copy
 

docker/cpu/docker-compose.yml

@@ -1,17 +1,17 @@
 name: kokoro-tts
 services:
   kokoro-tts:
-    image: ghcr.io/remsky/kokoro-fastapi-cpu:latest
-    # Uncomment below (and comment out above) to build from source instead of using the released image
+    image: ghcr.io/remsky/kokoro-fastapi-cpu:v0.1.0
     # build:
-      # context: ../..
-      # dockerfile: docker/cpu/Dockerfile
+    #   context: ../..
+    #   dockerfile: docker/cpu/Dockerfile
     volumes:
       - ../../api/src:/app/api/src
+      - ../../api/src/voices:/app/api/src/voices
     ports:
       - "8880:8880"
     environment:
-      - PYTHONPATH=/app:/app/Kokoro-82M
+      - PYTHONPATH=/app:/app/models
       # ONNX Optimization Settings for vectorized operations
       - ONNX_NUM_THREADS=8  # Maximize core usage for vectorized ops
       - ONNX_INTER_OP_THREADS=4  # Higher inter-op for parallel matrix operations
@@ -22,10 +22,10 @@ services:
       
   # Gradio UI service [Comment out everything below if you don't need it]
   gradio-ui:
-    image: ghcr.io/remsky/kokoro-fastapi:latest-ui
+    image: ghcr.io/remsky/kokoro-fastapi-ui:v0.1.0
     # Uncomment below (and comment out above) to build from source instead of using the released image
-    # build:
-      # context: ../../ui
+    build:
+      context: ../../ui
     ports:
       - "7860:7860"
     volumes:
@@ -34,4 +34,4 @@ services:
     environment:
       - GRADIO_WATCH=True  # Enable hot reloading
       - PYTHONUNBUFFERED=1  # Ensure Python output is not buffered
-      - DISABLE_LOCAL_SAVING=false  # Set to 'true' to disable local saving and hide file view
+      - DISABLE_LOCAL_SAVING=false  # Set to 'true' to disable local saving and hide file view

docker/gpu/Dockerfile

@@ -18,14 +18,14 @@ COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
 RUN useradd -m -u 1000 appuser
 
 # Create directories and set ownership
-RUN mkdir -p /app/api/model_files && \
+RUN mkdir -p /app/models && \
     mkdir -p /app/api/src/voices && \
     chown -R appuser:appuser /app
 
 USER appuser
 
 # Download and extract models
-WORKDIR /app/api/model_files
+WORKDIR /app/models
 RUN curl -L -o model.tar.gz https://github.com/remsky/Kokoro-FastAPI/releases/download/v0.0.1/kokoro-82m-pytorch.tar.gz && \
     tar xzf model.tar.gz && \
     rm model.tar.gz
@@ -56,7 +56,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 
 # Set environment variables
 ENV PYTHONUNBUFFERED=1
-ENV PYTHONPATH=/app:/app/Kokoro-82M
+ENV PYTHONPATH=/app:/app/models
 ENV PATH="/app/.venv/bin:$PATH"
 ENV UV_LINK_MODE=copy
 

docker/gpu/docker-compose.yml

@@ -1,17 +1,17 @@
 name: kokoro-tts
 services:
   kokoro-tts:
-    image: ghcr.io/remsky/kokoro-fastapi-gpu:latest
-      # Uncomment below (and comment out above) to build from source instead of using the released image
-    # build:
-      # context: ../..
-      # dockerfile: docker/gpu/Dockerfile
+    # image: ghcr.io/remsky/kokoro-fastapi-gpu:latest
+    build:
+      context: ../..
+      dockerfile: docker/gpu/Dockerfile
     volumes:
       - ../../api/src:/app/api/src  # Mount src for development
+      - ../../api/src/voices:/app/api/src/voices  # Mount voices for persistence
     ports:
       - "8880:8880"
     environment:
-      - PYTHONPATH=/app:/app/Kokoro-82M
+      - PYTHONPATH=/app:/app/models
     deploy:
       resources:
         reservations: