Update app.py

app.py

@@ -3,67 +3,78 @@ import spaces
 from transformers import AutoModelForCausalLM, AutoProcessor
 import torch
 from PIL import Image
-import subprocess
-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 = "[Phi-3.5-vision Demo](https://huggingface.co/microsoft/Phi-3.5-vision-instruct)"
-
-kwargs = {}
-kwargs['torch_dtype'] = torch.bfloat16
-
-user_prompt = '<|user|>\n'
-assistant_prompt = '<|assistant|>\n'
-prompt_suffix = "<|end|>\n"
+# Set the default query
+DEFAULT_QUERY = "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]."
 
 @spaces.GPU
-def run_example(image, text_input=None, model_id="microsoft/Phi-3.5-vision-instruct"):
+def run_example(image, model_id="microsoft/Phi-3.5-vision-instruct"):
     model = models[model_id]
     processor = processors[model_id]
 
-    prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
+    # Use the default query directly without a user input text field
+    prompt = f"<|user|>\n<|image_1|>\n{DEFAULT_QUERY}<|end|>\n<|assistant|>"
     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
 
 css = """
-  #output {
-    height: 500px; 
-    overflow: auto; 
-    border: 1px solid #ccc; 
+  #container {
+    background-color: #f9f9f9;
+    padding: 20px;
+    border-radius: 15px;
+    border: 2px solid #333;
+    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2);
+    max-width: 450px;
+    margin: auto;
+  }
+  #input_image {
+    margin-top: 15px;
+    border: 2px solid #333;
+    border-radius: 8px;
+    height: 180px;
+    object-fit: contain;
+  }
+  #output_text {
+    margin-top: 15px;
+    border: 2px solid #333;
+    border-radius: 8px;
+    height: 180px;
+    overflow-y: auto;
+  }
+  #submit_btn {
+    background-color: #fff;
+    color: black;
+    border-radius: 10px;
+    padding: 10px;
+    cursor: pointer;
+    transition: background-color 0.3s ease;
+    margin-top: 15px;
+  }
+  #submit_btn:hover {
+    background-color: #333;
   }
 """
 
 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")
-                text_input = gr.Textbox(label="Question")
-                submit_btn = gr.Button(value="Submit")
-            with gr.Column():
-                output_text = gr.Textbox(label="Output Text")
+    with gr.Column(elem_id="container"):
+        input_image = gr.Image(type="pil", label="Upload Image", elem_id="input_image")
+        submit_btn = gr.Button(value="Generate Prompt", elem_id="submit_btn")
+        output_text = gr.Textbox(label="Prompt Output", elem_id="output_text", show_copy_button=True, lines=6)
 
-        submit_btn.click(run_example, [input_img, text_input, model_selector], [output_text])
+    submit_btn.click(run_example, [input_image], output_text)
 
-demo.queue(api_open=False)
-demo.launch(debug=True, show_api=False)
+demo.launch(share=False)