push

Dockerfile

@@ -55,11 +55,8 @@ RUN mkdir -p /models/huggingface && \
     mkdir -p /code/models && \
     chmod -R 777 /models/huggingface
 
-# Pre-download model at build time (YarnGPT2 model)
-RUN python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='saheedniyi/YarnGPT2')"
-
-# Preload tokenizer (avoid runtime delays)
-RUN python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('saheedniyi/YarnGPT2', use_fast=True)"
+# Note: Models will be downloaded lazily at runtime to save build storage
+# Pre-downloading is disabled to avoid storage limit issues
 
 # Download wavtokenizer configuration file (to both root and models directory)
 RUN wget -O /code/wavtokenizer_mediumdata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml \

main.py

@@ -172,15 +172,10 @@ def load_model():
 
 @app.on_event("startup")
 async def startup_event():
-    """Initialize model and tokenizer on startup."""
-    try:
-        logger.info("Initializing YarnGPT2 TTS model...")
-        load_model()
-        load_audio_tokenizer()
-        logger.info("Model initialization complete")
-    except Exception as e:
-        logger.error(f"Failed to initialize model: {e}")
-        logger.warning("Server will start but TTS functionality will be unavailable")
+    """Initialize model and tokenizer lazily."""
+    logger.info("Server started. Models will be loaded on first request.")
+    # Don't load models at startup to save storage and startup time
+    # Models will be loaded lazily when first API call is made
 
 @app.get("/")
 async def root():
@@ -224,10 +219,19 @@ async def text_to_speech(request: TTSRequest):
     Returns:
     - Audio file in WAV format
     """
+    # Lazy load models on first request
+    global model, audio_tokenizer
+    if model is None:
+        logger.info("Loading YarnGPT2 model (lazy loading)...")
+        load_model()
+    if audio_tokenizer is None:
+        logger.info("Loading audio tokenizer (lazy loading)...")
+        load_audio_tokenizer()
+    
     if model is None or audio_tokenizer is None:
         raise HTTPException(
             status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
-            detail="Model not loaded. Please wait or restart the server."
+            detail="Model loading failed. Please check logs."
         )
     
     try:
@@ -280,10 +284,19 @@ async def text_to_speech_stream(request: TTSRequest):
     
     Same parameters as /tts endpoint.
     """
+    # Lazy load models on first request
+    global model, audio_tokenizer
+    if model is None:
+        logger.info("Loading YarnGPT2 model (lazy loading)...")
+        load_model()
+    if audio_tokenizer is None:
+        logger.info("Loading audio tokenizer (lazy loading)...")
+        load_audio_tokenizer()
+    
     if model is None or audio_tokenizer is None:
         raise HTTPException(
             status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
-            detail="Model not loaded. Please wait or restart the server."
+            detail="Model loading failed. Please check logs."
         )
     
     try: