MaziyarPanahi commited on
Commit
8f558df
1 Parent(s): bc53c96

Create app.py (#3)

Browse files
Files changed (1) hide show
  1. app.py +48 -0
app.py ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ from transformers import AutoModelForCausalLM, AutoProcessor
3
+ from PIL import Image
4
+
5
+ # Define constants
6
+ MODEL_NAME = "microsoft/Phi-3.5-vision-instruct"
7
+ DESCRIPTION = "# [Phi-3.5-vision Demo](https://huggingface.co/microsoft/Phi-3.5-vision-instruct)"
8
+ DEVICE = "cuda"
9
+
10
+ # Load model and processor
11
+ model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, trust_remote_code=True, torch_dtype="auto", _attn_implementation="flash_attention_2").to(DEVICE).eval()
12
+ processor = AutoProcessor.from_pretrained(MODEL_NAME, trust_remote_code=True)
13
+
14
+ def run_example(image, text_input, model_id):
15
+ # Prepare prompt and image for processing
16
+ prompt = f"{text_input}\n"
17
+ image = Image.fromarray(image).convert("RGB")
18
+
19
+ # Process input
20
+ inputs = processor(prompt, image, return_tensors="pt").to(DEVICE)
21
+ generate_ids = model.generate(**inputs, max_new_tokens=1000, eos_token_id=processor.tokenizer.eos_token_id)
22
+ generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
23
+ response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
24
+
25
+ return response
26
+
27
+ css = """
28
+ #output {
29
+ height: 500px;
30
+ overflow: auto;
31
+ border: 1px solid #ccc;
32
+ }
33
+ """
34
+
35
+ # Set up the Gradio interface
36
+ with gr.Blocks(css=css) as demo:
37
+ gr.Markdown(DESCRIPTION)
38
+ with gr.Tab(label="Phi-3.5 Input"):
39
+ with gr.Row():
40
+ with gr.Column():
41
+ input_img = gr.Image(label="Input Picture")
42
+ text_input = gr.Textbox(label="Question")
43
+ submit_btn = gr.Button(value="Submit")
44
+ with gr.Column():
45
+ output_text = gr.Textbox(label="Output Text")
46
+ submit_btn.click(run_example, inputs=[input_img, text_input, MODEL_NAME], outputs=output_text)
47
+
48
+ demo.launch(debug=True)