Update app.py

app.py

@@ -4,80 +4,72 @@ from transformers import AutoModelForCausalLM, AutoProcessor
 import torch
 from PIL import Image
 import subprocess
+
+# Install flash-attn with no CUDA build isolation
 subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
+# Load model and processor
 models = {
     "microsoft/Phi-3.5-vision-instruct": AutoModelForCausalLM.from_pretrained("microsoft/Phi-3.5-vision-instruct", trust_remote_code=True, torch_dtype="auto", _attn_implementation="flash_attention_2").cuda().eval()
 }
-
 processors = {
     "microsoft/Phi-3.5-vision-instruct": AutoProcessor.from_pretrained("microsoft/Phi-3.5-vision-instruct", trust_remote_code=True)
 }
 
-DESCRIPTION = " "
-
-kwargs = {}
-kwargs['torch_dtype'] = torch.bfloat16
-
-user_prompt = '<|user|>\n'
-assistant_prompt = '<|assistant|>\n'
-prompt_suffix = "<|end|>\n"
-
-default_question = "You are an image to prompt converter. Your work is to observe each and every detail of the image and craft a detailed prompt under 100 words in this format: [image content/subject, description of action, state, and mood], [art form, style], [artist/photographer reference if needed], [additional settings such as camera and lens settings, lighting, colors, effects, texture, background, rendering]."
+# Default description and prompt
+DESCRIPTION = ""
+default_question = "You are an image to prompt converter. Your work is to observe each and every detail of the image and craft a detailed prompt under 100 words."
 
+# Gradio function for generating output from image input
 @spaces.GPU
 def run_example(image, text_input=default_question, model_id="microsoft/Phi-3.5-vision-instruct"):
     model = models[model_id]
     processor = processors[model_id]
-
+    user_prompt = '<|user|>\n'
+    assistant_prompt = '<|assistant|>\n'
+    prompt_suffix = "<|end|>\n"
+    
     prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
     image = Image.fromarray(image).convert("RGB")
 
     inputs = processor(prompt, image, return_tensors="pt").to("cuda:0")
-    generate_ids = model.generate(**inputs, 
-                                max_new_tokens=1000,
-                                eos_token_id=processor.tokenizer.eos_token_id,
-                                )
+    generate_ids = model.generate(**inputs, max_new_tokens=1000, eos_token_id=processor.tokenizer.eos_token_id)
     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]
+    response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
     return response
 
+# Custom CSS for styling
 css = """
-  #output {
-    margin-top: 15px;
-    border: 2px solid #333; /* Darker outline */
-    border-radius: 8px;
-    height: 180px; /* Fixed height */
-    object-fit: contain; /* Ensure image fits within the fixed height */ 
+  #output_text {
+    height: 500px;
+    overflow: auto;
+    border: 1px solid #333;
   }
-
-   #input_img {
-    margin-top: 15px;
-    border: 2px solid #333; /* Darker outline */
-    border-radius: 8px;
-    height: 180px; /* Fixed height */
-    object-fit: contain; /* Ensure image fits within the fixed height */
+  #model_selector, #text_input {
+    display: none !important;
   }
-  #model_selector, #text_input { 
-    display: none !important; 
+  #main_container {
+    border: 2px solid black;
+    padding: 20px;
+    border-radius: 10px;
   }
 """
 
+# Gradio interface with styling and layout improvements
 with gr.Blocks(css=css) as demo:
     gr.Markdown(DESCRIPTION)
-    with gr.Tab(label="Phi-3.5 Input"):
-        with gr.Row():
-            with gr.Column():
-                input_img = gr.Image(label="Input Picture")
-                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="microsoft/Phi-3.5-vision-instruct", visible=False)
-                text_input = gr.Textbox(label="Question", value=default_question, visible=False)
-                submit_btn = gr.Button(value="Submit")
-            with gr.Column():
-                output_text = gr.Textbox(label="Output Text")
+    with gr.Row(id="main_container"):
+        with gr.Column():
+            input_img = gr.Image(label="Input Image", interactive=True)
+            model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="microsoft/Phi-3.5-vision-instruct", visible=False)
+            text_input = gr.Textbox(label="Question", value=default_question, visible=False)
+            submit_btn = gr.Button(value="Generate Prompt")
+        
+        output_text = gr.Textbox(label="Output", id="output_text", interactive=False)
 
-        submit_btn.click(run_example, [input_img, text_input, model_selector], [output_text])
+    # Link button action to function
+    submit_btn.click(run_example, [input_img, text_input, model_selector], output_text)
 
+# Launch Gradio interface
 demo.queue(api_open=False)
 demo.launch(debug=True, show_api=False)