Create download_model.py

download_model.py

@@ -0,0 +1,49 @@
+"""
+Pre-download the model to ensure it's available during app startup.
+This script should be run before launching the app.
+"""
+
+import os
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
+
+# Set cache directories
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
+os.environ["HF_HOME"] = "/tmp/hf_home"
+os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/hf_hub_cache"
+
+# Create cache directories
+for cache_dir in ["/tmp/transformers_cache", "/tmp/hf_home", "/tmp/hf_hub_cache"]:
+    os.makedirs(cache_dir, exist_ok=True)
+
+# Model configuration
+MODEL_NAME = "openai/whisper-large-v3-turbo"
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+
+print("Pre-downloading model to cache...")
+print(f"Using device: {device}")
+print(f"Using dtype: {torch_dtype}")
+
+try:
+    # Download model
+    model = AutoModelForSpeechSeq2Seq.from_pretrained(
+        MODEL_NAME, 
+        torch_dtype=torch_dtype, 
+        low_cpu_mem_usage=True, 
+        use_safetensors=True,
+        cache_dir="/tmp/transformers_cache",
+    )
+    print("Model downloaded successfully!")
+    
+    # Download processor
+    processor = AutoProcessor.from_pretrained(
+        MODEL_NAME,
+        cache_dir="/tmp/transformers_cache",
+    )
+    print("Processor downloaded successfully!")
+    
+    print("Pre-download complete! Model is ready for use.")
+except Exception as e:
+    print(f"Error during pre-download: {e}")
+    raise