add

app.py

@@ -1,8 +1,10 @@
 import gradio as gr
-import soundfile as sf
 from PIL import Image
-import spaces
+import torch
+import soundfile as sf
 from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
+from urllib.request import urlopen
+import spaces
 
 # Define model path
 model_path = "microsoft/Phi-4-multimodal-instruct"
@@ -10,12 +12,14 @@ model_path = "microsoft/Phi-4-multimodal-instruct"
 # Load model and processor
 processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
-    model_path, 
-    device_map="cuda", 
-    torch_dtype="auto", 
-    trust_remote_code=True, 
+    model_path,
+    device_map="cuda",
+    torch_dtype="auto",
+    trust_remote_code=True,
+    attn_implementation="eager",  # Changed from 'flash_attention_2' to 'eager'
 ).cuda()
 
+# Load generation config
 generation_config = GenerationConfig.from_pretrained(model_path)
 
 # Define prompt structure
@@ -23,51 +27,104 @@ user_prompt = '<|user|>'
 assistant_prompt = '<|assistant|>'
 prompt_suffix = '<|end|>'
 
+# Define inference function
 @spaces.GPU
-def process_multimodal(input_file, query):
-    if input_file is None:
-        return "Please upload an image or an audio file."
-    
-    file_type = input_file.type
-    prompt = f"{user_prompt}<|media_1|>{query}{prompt_suffix}{assistant_prompt}"
-    
-    if "image" in file_type:
-        image = Image.open(input_file)
+def process_input(input_type, file, question):
+    if not file or not question:
+        return "Please upload a file and provide a question."
+
+    # Prepare the prompt
+    if input_type == "Image":
+        prompt = f'{user_prompt}<|image_1|>{question}{prompt_suffix}{assistant_prompt}'
+        # Open image from uploaded file
+        image = Image.open(file)
         inputs = processor(text=prompt, images=image, return_tensors='pt').to('cuda:0')
-    elif "audio" in file_type:
-        audio, samplerate = sf.read(input_file.name)
+    elif input_type == "Audio":
+        prompt = f'{user_prompt}<|audio_1|>{question}{prompt_suffix}{assistant_prompt}'
+        # Read audio from uploaded file
+        audio, samplerate = sf.read(file)
         inputs = processor(text=prompt, audios=[(audio, samplerate)], return_tensors='pt').to('cuda:0')
     else:
-        return "Unsupported file format. Please upload an image or audio file."
-    
-    generate_ids = model.generate(
-        **inputs,
-        max_new_tokens=1000,
-        generation_config=generation_config,
-        num_logits_to_keep=0,
-    )
+        return "Invalid input type selected."
+
+    # Generate response
+    with torch.no_grad():
+        generate_ids = model.generate(
+            **inputs,
+            max_new_tokens=1000,
+            generation_config=generation_config,
+            num_logits_to_keep=0,
+        )
     generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
     response = processor.batch_decode(
         generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
     )[0]
-    
+
     return response
 
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("""
-    # Phi-4 Multimodal Chat
-    Upload an image or an audio file and ask questions related to it!
-    """)
+# Gradio interface
+with gr.Blocks(
+    title="Phi-4 Multimodal Demo",
+    theme=gr.themes.Soft(
+        primary_hue="blue",
+        secondary_hue="gray",
+        radius_size="lg",
+    ),
+) as demo:
+    gr.Markdown(
+        """
+        # Phi-4 Multimodal Demo
+        Upload an **image** or **audio** file, ask a question, and get a response from the model!  
+        Built with the `microsoft/Phi-4-multimodal-instruct` model by xAI.
+        """
+    )
     
     with gr.Row():
-        with gr.Column():
-            input_file = gr.File(label="Upload Image or Audio")
-            query = gr.Textbox(label="Ask a question")
-            submit_btn = gr.Button("Submit")
+        with gr.Column(scale=1):
+            input_type = gr.Radio(
+                choices=["Image", "Audio"],
+                label="Select Input Type",
+                value="Image",
+            )
+            file_input = gr.File(
+                label="Upload Your File",
+                file_types=["image", "audio"],
+            )
+            question_input = gr.Textbox(
+                label="Your Question",
+                placeholder="e.g., 'What is shown in this image?' or 'Transcribe this audio.'",
+                lines=2,
+            )
+            submit_btn = gr.Button("Submit", variant="primary")
         
-        with gr.Column():
-            output = gr.Textbox(label="Response", interactive=False)
-    
-    submit_btn.click(process_multimodal, inputs=[input_file, query], outputs=output)
+        with gr.Column(scale=2):
+            output_text = gr.Textbox(
+                label="Model Response",
+                placeholder="Response will appear here...",
+                lines=10,
+                interactive=False,
+            )
+
+    # Example section
+    with gr.Accordion("Examples", open=False):
+        gr.Markdown("Try these examples:")
+        gr.Examples(
+            examples=[
+                ["Image", "https://www.ilankelman.org/stopsigns/australia.jpg", "What is shown in this image?"],
+                ["Audio", "https://upload.wikimedia.org/wikipedia/commons/b/b0/Barbara_Sahakian_BBC_Radio4_The_Life_Scientific_29_May_2012_b01j5j24.flac", "Transcribe the audio to text."],
+            ],
+            inputs=[input_type, file_input, question_input],
+            outputs=output_text,
+            fn=process_input,
+            cache_examples=False,
+        )
+
+    # Connect the submit button
+    submit_btn.click(
+        fn=process_input,
+        inputs=[input_type, file_input, question_input],
+        outputs=output_text,
+    )
 
-demo.launch()
+# Launch the demo
+demo.launch()