dawood/Chat_With_Blip2-test

1import requests2from PIL import Image3import gradio as gr4from transformers import AutoProcessor, Blip2ForConditionalGeneration5import torch6 7 8css = """9#column_container {10  position: relative;11  height: 800px;12  max-width: 700px;13  display: flex;14  flex-direction: column;15  background-color: lightgray;16  border: 1px solid gray;17  border-radius: 5px;18  padding: 10px;19  box-shadow: 2px 2px 5px gray;20  margin-left: auto; 21  margin-right: auto;22}23#input_prompt {24  position: fixed;25  bottom: 0;26  max-width: 680px;27}28#chatbot-component {29  overflow: auto;30}31"""32 33processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")34model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16) 35 36device = "cuda" if torch.cuda.is_available() else "cpu"37model.to(device)38 39def upload_button_config():40    return gr.update(visible=False)41 42def update_textbox_config(text_in):43    return gr.update(visible=True)44 45#takes input and generates the Response46def predict(btn_upload, counter,image_hid, input, history):47    48    if counter == 0:49      image_in = Image.open(btn_upload)50      #Resizing the image51      basewidth = 51252      wpercent = (basewidth/float(image_in.size[0]))53      hsize = int((float(image_in.size[1])*float(wpercent)))54      image_in = image_in.resize((basewidth,hsize)) #, Image.Resampling.LANCZOS)55      # Save the image to the file-like object56      #seed = random.randint(0, 1000000)57      img_name = "uploaded_image.png" #f"./edited_image_{seed}.png"58      image_in.save(img_name)59      #add state60      history = history or []61      response = '<img src="/file=' + img_name + '">'62      history.append((input, response))63      counter += 164      return history, history, img_name, counter, image_in65 66    #process the prompt67    print(f"prompt is :{input}") 68    #Getting prompt in the format - Question: Is this photo unusual? Answer:69    prompt = f"Question: {input} Answer: "70    inputs = processor(image_hid, text=prompt, return_tensors="pt").to(device, torch.float16)71    72    #generate the response73    generated_ids = model.generate(**inputs, max_new_tokens=10)74    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()75    print(f"generated_text is : {generated_text}")76 77    #add state78    history = history or []79    response = generated_text 80    history.append((input, response))81    counter += 182    return history, history, "uploaded_image.png", counter, image_hid83 84#Blocks Layout - leaving this here for moment - "#chatbot-component .overflow-y-auto{height:800px}"85with gr.Blocks(css="#chatbot-component {height: 900px}") as demo:  86  with gr.Row():87    with gr.Column(scale=1):88        #with gr.Accordion("See details"):89        gr.HTML("""<div style="text-align: center; max-width: 700px; margin: 0 auto;">90                    <div91                style="92                    display: inline-flex;93                    align-items: center;94                    gap: 0.8rem;95                    font-size: 1.75rem;96                "97                >98                <h1 style="font-weight: 900; margin-bottom: 7px; margin-top: 5px;">99                    Bringing Visual Conversations to Life with BLIP2100                </h1>101                </div>102                <p style="margin-bottom: 10px; font-size: 94%">103                Blip2 is functioning as an <b>instructed zero-shot image-to-text generation</b> model using OPT-2.7B in this Space. 104                It shows a wide range of capabilities including visual conversation, visual knowledge reasoning, visual commensense reasoning, storytelling, 105                personalized image-to-text generation etc.<br>106                BLIP-2 by <a href="https://huggingface.co/Salesforce" target="_blank">Salesforce</a> is now available in🤗Transformers! 107                This model was contributed by <a href="https://twitter.com/NielsRogge" target="_blank">nielsr</a>. 108                The BLIP-2 model was proposed in <a href="https://arxiv.org/abs/2301.12597" target="_blank">BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models</a> 109                by Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi.<br><br>110                </p></div>""")111        gr.HTML("""<a href="https://huggingface.co/spaces/ysharma/InstructPix2Pix_Chatbot?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>Duplicate Space with GPU Upgrade for fast Inference & no queue<br>""")112    113    with gr.Column(elem_id = "column_container", scale=2):114        #text_in = gr.Textbox(value='', placeholder="Type your questions here and press enter", elem_id = "input_prompt", visible=False, label='Great! Now you can ask questions to get more information about the image')115        btn_upload = gr.UploadButton("Upload image!", file_types=["image"], file_count="single", elem_id="upload_button")116        chatbot = gr.Chatbot(elem_id = 'chatbot-component', label='Converse with Images')117        text_in = gr.Textbox(value='', placeholder="Type your questions here and press enter", elem_id = "input_prompt", visible=False, label='Great! Now you can ask questions to get more information about the image')118        state_in = gr.State()119        counter_out = gr.Number(visible=False, value=0, precision=0)120        text_out = gr.Textbox(visible=False)  #getting image name out121        image_hid = gr.Image(visible=False) #, type='pil')122 123  #Using Event Listeners124  btn_upload.upload(predict, [btn_upload, counter_out, image_hid, text_in, state_in], [chatbot, state_in, text_out, counter_out, image_hid])125  btn_upload.upload(fn = update_textbox_config, inputs=text_in, outputs = text_in)126 127  text_in.submit(predict, [btn_upload, counter_out, image_hid, text_in, state_in], [chatbot, state_in, text_out, counter_out, image_hid])128  129  chatbot.change(fn = upload_button_config, outputs=btn_upload) #, scroll_to_output = True)130    131demo.queue(concurrency_count=10)132demo.launch(debug=True) #, width="80%", height=2000)