Update app.py

app.py

@@ -1,4 +1,18 @@
+# Set cache directories first, before other imports
 import os
+
+# Set all cache directories to locations within /tmp
+os.environ["HF_HOME"] = "/tmp/hf_home"
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
+os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/huggingface_hub_cache"
+os.environ["TORCH_HOME"] = "/tmp/torch_home"
+os.environ["XDG_CACHE_HOME"] = "/tmp/xdg_cache"
+
+# Create necessary directories
+for path in ["/tmp/hf_home", "/tmp/transformers_cache", "/tmp/huggingface_hub_cache", "/tmp/torch_home", "/tmp/xdg_cache"]:
+    os.makedirs(path, exist_ok=True)
+
+# Now import the rest of the libraries
 import torch
 import torchaudio
 import soundfile as sf
@@ -6,19 +20,32 @@ from flask import Flask, request, jsonify, send_file
 from flask_cors import CORS
 from transformers import Wav2Vec2ForCTC, AutoProcessor, VitsModel, AutoTokenizer
 
-# Set cache directories
-os.environ["HF_HOME"] = "/tmp/hf_home"
-os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
-os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/huggingface_cache"
-os.environ["TORCH_HOME"] = "/tmp/torch_home"
-
 app = Flask(__name__)
 CORS(app)
 
-# ASR Model (facebook/mms-1b-all)
+# ASR Model
 ASR_MODEL_ID = "Coco-18/mms-asr-tgl-en-safetensor"
-asr_processor = AutoProcessor.from_pretrained(ASR_MODEL_ID)
-asr_model = Wav2Vec2ForCTC.from_pretrained(ASR_MODEL_ID)
+print(f"Loading ASR model: {ASR_MODEL_ID}")
+
+try:
+    asr_processor = AutoProcessor.from_pretrained(
+        ASR_MODEL_ID,
+        cache_dir="/tmp/transformers_cache"  # Explicitly set cache_dir
+    )
+    asr_model = Wav2Vec2ForCTC.from_pretrained(
+        ASR_MODEL_ID,
+        cache_dir="/tmp/transformers_cache"  # Explicitly set cache_dir
+    )
+    print("✅ ASR Model loaded successfully")
+except Exception as e:
+    print(f"❌ Error loading ASR model: {str(e)}")
+    # Provide more debugging information
+    import sys
+    print(f"Python version: {sys.version}")
+    print(f"Current working directory: {os.getcwd()}")
+    print(f"Temp directory exists: {os.path.exists('/tmp')}")
+    print(f"Temp directory writeable: {os.access('/tmp', os.W_OK)}")
+    # Let's continue anyway to see if we can at least start the API
 
 # Language-specific configurations
 LANGUAGE_CODES = {
@@ -38,8 +65,14 @@ tts_models = {}
 tts_processors = {}
 for lang, model_id in TTS_MODELS.items():
     try:
-        tts_models[lang] = VitsModel.from_pretrained(model_id, cache_dir="/tmp/huggingface_cache")
-        tts_processors[lang] = AutoTokenizer.from_pretrained(model_id, cache_dir="/tmp/huggingface_cache")
+        tts_models[lang] = VitsModel.from_pretrained(
+            model_id, 
+            cache_dir="/tmp/transformers_cache"  # Explicitly set cache_dir
+        )
+        tts_processors[lang] = AutoTokenizer.from_pretrained(
+            model_id, 
+            cache_dir="/tmp/transformers_cache"  # Explicitly set cache_dir
+        )
         print(f"✅ TTS Model loaded: {lang}")
     except Exception as e:
         print(f"❌ Error loading {lang} TTS model: {e}")
@@ -47,7 +80,7 @@ for lang, model_id in TTS_MODELS.items():
 
 # Constants
 SAMPLE_RATE = 16000
-OUTPUT_DIR = "/tmp/"
+OUTPUT_DIR = "/tmp/audio_outputs"
 os.makedirs(OUTPUT_DIR, exist_ok=True)
 
 
@@ -149,5 +182,4 @@ def download_audio(filename):
 
 
 if __name__ == "__main__":
-    app.run(host="0.0.0.0", port=7860, debug=True)
-
+    app.run(host="0.0.0.0", port=7860, debug=True)