Update app.py

app.py

@@ -10,6 +10,8 @@ import os
 
 # Read the Hugging Face access token from the environment variable
 read_token = os.getenv('AccToken')
+if not read_token:
+    raise ValueError("Hugging Face access token not found. Please set the AccToken environment variable.")
 login(read_token)
 
 # Define a dictionary of conversational models
@@ -19,7 +21,7 @@ conversational_models = {
     "Perplexity (R1 Post-trained)": "perplexity-ai/r1-1776",
     "Llama-Instruct by Meta": "meta-llama/Llama-3.2-3B-Instruct",
     "Mistral": "mistralai/Mistral-7B-v0.1",
-    "Gemma": "google/gemma-3-27b-it",
+    "Gemma": "google/gemma-2-2b-it",
 }
 
 # Define a dictionary of Text-to-Image models
@@ -45,17 +47,26 @@ text_to_image_pipelines = {}
 text_to_speech_pipelines = {}
 
 # Initialize pipelines for other tasks
-visual_qa_pipeline = pipeline("visual-question-answering", model="dandelin/vilt-b32-finetuned-vqa")
-document_qa_pipeline = pipeline("question-answering", model="deepset/roberta-base-squad2")
-image_classification_pipeline = pipeline("image-classification", model="facebook/detr-resnet-50")  # This will be replaced
-object_detection_pipeline = pipeline("object-detection", model="facebook/detr-resnet-50")
-video_classification_pipeline = pipeline("video-classification", model="facebook/timesformer-base-finetuned-k400")
-summarization_pipeline = pipeline("summarization", model="facebook/bart-large-cnn")
-text_to_audio_pipeline = pipeline("text-to-audio", model="stabilityai/stable-audio-open-1.0")
-audio_classification_pipeline = pipeline("audio-classification", model="facebook/wav2vec2-base")
+visual_qa_pipeline = pipeline("visual-question-answering", model="dandelin/vilt-b32-finetuned-vqa", use_auth_token=read_token)
+document_qa_pipeline = pipeline("question-answering", model="deepset/roberta-base-squad2", use_auth_token=read_token)
+image_classification_pipeline = pipeline("image-classification", model="facebook/detr-resnet-50", use_auth_token=read_token)
+object_detection_pipeline = pipeline("object-detection", model="facebook/detr-resnet-50", use_auth_token=read_token)
+video_classification_pipeline = pipeline("video-classification", model="facebook/timesformer-base-finetuned-k400", use_auth_token=read_token)
+summarization_pipeline = pipeline("summarization", model="facebook/bart-large-cnn", use_auth_token=read_token)
+
+# Use a different model for text-to-audio if stabilityai/stable-audio-open-1.0 is not supported
+try:
+    text_to_audio_pipeline = pipeline("text-to-audio", model="stabilityai/stable-audio-open-1.0", use_auth_token=read_token)
+except ValueError as e:
+    print(f"Error loading stabilityai/stable-audio-open-1.0: {e}")
+    print("Falling back to a different text-to-audio model.")
+    text_to_audio_pipeline = pipeline("text-to-audio", model="microsoft/speecht5_tts", use_auth_token=read_token)
+
+audio_classification_pipeline = pipeline("audio-classification", model="facebook/wav2vec2-base", use_auth_token=read_token)
 
 def load_conversational_model(model_name):
     if model_name not in conversational_models_loaded:
+        print(f"Loading conversational model: {model_name}")
         tokenizer = AutoTokenizer.from_pretrained(conversational_models[model_name], use_auth_token=read_token)
         model = AutoModelForCausalLM.from_pretrained(conversational_models[model_name], use_auth_token=read_token)
         conversational_tokenizers[model_name] = tokenizer
@@ -84,6 +95,7 @@ def chat(model_name, user_input, history=[]):
 
 def generate_image(model_name, prompt):
     if model_name not in text_to_image_pipelines:
+        print(f"Loading text-to-image model: {model_name}")
         text_to_image_pipelines[model_name] = StableDiffusionPipeline.from_pretrained(
             text_to_image_models[model_name], use_auth_token=read_token
         )
@@ -93,6 +105,7 @@ def generate_image(model_name, prompt):
 
 def generate_speech(model_name, text):
     if model_name not in text_to_speech_pipelines:
+        print(f"Loading text-to-speech model: {model_name}")
         text_to_speech_pipelines[model_name] = pipeline(
             "text-to-speech", model=text_to_speech_models[model_name], use_auth_token=read_token
         )