Create app.py

app.py

@@ -1,185 +1,47 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import gradio as gr
 import torch
-import spaces
-from diffusers import FluxInpaintPipeline
-from PIL import Image, ImageFile
 
-#ImageFile.LOAD_TRUNCATED_IMAGES = True
 
-# Initialize the pipeline
-pipe = FluxInpaintPipeline.from_pretrained(
-    "black-forest-labs/FLUX.1-dev", 
-    torch_dtype=torch.bfloat16
-)
-pipe.to("cuda")
-pipe.load_lora_weights(
-    "ali-vilab/In-Context-LoRA", 
-    weight_name="visual-identity-design.safetensors"
-)
+title = "????AI ChatBot"
+description = "A State-of-the-Art Large-scale Pretrained Response generation model (DialoGPT)"
+examples = [["How are you?"]]
 
-def square_center_crop(img, target_size=768):
-    if img.mode in ('RGBA', 'P'):
-        img = img.convert('RGB')
 
-    width, height = img.size
-    crop_size = min(width, height)
+tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-large")
+model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-large")
 
-    left = (width - crop_size) // 2
-    top = (height - crop_size) // 2
-    right = left + crop_size
-    bottom = top + crop_size
 
-    img_cropped = img.crop((left, top, right, bottom))
-    return img_cropped.resize((target_size, target_size), Image.Resampling.LANCZOS)
-
-def duplicate_horizontally(img):
-    width, height = img.size
-    if width != height:
-        raise ValueError(f"Input image must be square, got {width}x{height}")
-
-    new_image = Image.new('RGB', (width * 2, height))
-    new_image.paste(img, (0, 0))
-    new_image.paste(img, (width, 0))
-    return new_image
-
-
-@spaces.GPU
-def generate(image, prompt_description, prompt_user, progress=gr.Progress(track_tqdm=True)):
-    prompt_structure = "The two-panel image showcases the logo on the left and the application on the right, [LEFT] the left panel is showing "+prompt_description+" [RIGHT] this logo is applied to "
-    prompt = prompt_structure + prompt_user 
-    
-    mask = Image.open("mask_square.png")
-    cropped_image = square_center_crop(image)
-    logo_dupli = duplicate_horizontally(cropped_image)
-
-    out = pipe(
-        prompt=prompt,
-        image=logo_dupli,
-        mask_image=mask,
-        guidance_scale=3.5,
-        height=768,
-        width=1536,
-        num_inference_steps=28,
-        max_sequence_length=256,
-        strength=1
-    ).images[0]
-
-    width, height = out.size
-    half_width = width // 2
-    image_2 = out.crop((half_width, 0, width, height))
-    return image_2, out
-
-with gr.Blocks() as demo:
-    gr.Markdown("# Logo in Context")
-    gr.Markdown("### [In-Context LoRA](https://huggingface.co/ali-vilab/In-Context-LoRA) + Image-to-Image + Inpainting, apply your logo to anything. diffusers implementation based on the [workflow by WizardWhitebeard/klinter](https://civitai.com/articles/8779)")
-
-    with gr.Tab("Demo"):
-        with gr.Row():
-            with gr.Column():
-                input_image = gr.Image(
-                    label="Upload Logo Image",
-                    type="pil",
-                    height=384
-                )
-                prompt_description = gr.Textbox(
-                    label="Describe your logo",
-                    placeholder="A Hugging Face emoji logo",
-                )
-                prompt_input = gr.Textbox(
-                    label="Where should the logo be applied?",
-                    placeholder="e.g., a coffee cup on a wooden table"
-                )
-                generate_btn = gr.Button("Generate Application", variant="primary")
-    
-            with gr.Column():
-                output_image = gr.Image(label="Generated Application")
-                output_side = gr.Image(label="Side by side")
-        
-        gr.Examples(
-            examples=[
-                ["huggingface.png", "A Hugging Face emoji logo", "An embroidered hat"],
-                ["awesome.png", "An awesome face logo", "A tattoo on a leg"],
-                ["dvd_logo.png", "A DVD logo", "a coconut, engraved logo on a green coconut"]
-            ],
-            inputs=[input_image, prompt_description, prompt_input],
-            outputs=[output_image, output_side],
-            fn=generate,
-            cache_examples="lazy"
-        )
-        
-        with gr.Row():
-            gr.Markdown("""
-            ### Instructions:
-            1. Upload a logo image (preferably square)
-            2. Describe where you'd like to see the logo applied
-            3. Click 'Generate Application' and wait for the result
-    
-            Note: The generation process might take a few moments.
-            """)
-    
-    with gr.Tab("🧨 diffusers implementation"):
-        gr.Markdown("The way this works is combining the [IC LoRA](https://github.com/ali-vilab/In-Context-LoRA) with image-to-image + inpainting. Where the image on the left (the logo) is uploaded by the user, and the image on the right is masked and applied on the product by the LoRA. Based on the [ComfyUI workflow by WizardWhitebeard/klinter](https://civitai.com/articles/8779). Below is a diffusers implementation of the idea")
-        
-        gr.Code(language="python", value="""# Support functions
-def square_center_crop(img, target_size=768):
-    if img.mode in ('RGBA', 'P'):
-        img = img.convert('RGB')
-    
-    width, height = img.size
-    crop_size = min(width, height)
-    
-    left = (width - crop_size) // 2
-    top = (height - crop_size) // 2
-    right = left + crop_size
-    bottom = top + crop_size
-    
-    img_cropped = img.crop((left, top, right, bottom))
-    return img_cropped.resize((target_size, target_size), Image.Resampling.LANCZOS)
-
-def duplicate_horizontally(img):
-    width, height = img.size
-    if width != height:
-        raise ValueError(f"Input image must be square, got {width}x{height}")
-    
-    new_image = Image.new('RGB', (width * 2, height))
-    new_image.paste(img, (0, 0))
-    new_image.paste(img, (width, 0))
-    return new_image"""
-        )
-        
-        gr.Code(language="python", value="""import torch
-from diffusers import FluxInpaintPipeline
-from PIL import Image
-
-pipe = FluxInpaintPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
-pipe.to("cuda")
-pipe.load_lora_weights("ali-vilab/In-Context-LoRA", weight_name="visual-identity-design.safetensors")
-
-mask = load_image("mask_square.png")
-image = load_image("the_logo.png")
-cropped_image = square_center_crop(image) #crop the image you upload to square
-logo_dupli = duplicate_horizontally(cropped_image) #duplicate it so the right side can be masked
-
-prompt_structure = "The two-panel image showcases the logo of a brand, [LEFT] the left panel is showing the logo [RIGHT] the right panel has this logo applied to "
-prompt = prompt_structure + "an coconut, engraved logo on a green coconut"
-out = pipe(
-    prompt=prompt,
-    image=logo_dupli,
-    mask_image=mask,
-    guidance_scale=6,
-    height=768,
-    width=1536,
-    num_inference_steps=28,
-    max_sequence_length=256,
-    strength=1
-).images[0]"""
-        )
-        
-    # Set up the click event
-    generate_btn.click(
-        fn=generate,
-        inputs=[input_image, prompt_description, prompt_input],
-        outputs=[output_image, output_side]
+def predict(input, history=[]):
+    # tokenize the new input sentence
+    new_user_input_ids = tokenizer.encode(
+        input + tokenizer.eos_token, return_tensors="pt"
     )
 
-demo.launch()
+    # append the new user input tokens to the chat history
+    bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)
+
+    # generate a response
+    history = model.generate(
+        bot_input_ids, max_length=4000, pad_token_id=tokenizer.eos_token_id
+    ).tolist()
+
+    # convert the tokens to text, and then split the responses into lines
+    response = tokenizer.decode(history[0]).split("<|endoftext|>")
+    # print('decoded_response-->>'+str(response))
+    response = [
+        (response[i], response[i + 1]) for i in range(0, len(response) - 1, 2)
+    ]  # convert to tuples of list
+    # print('response-->>'+str(response))
+    return response, history
+
+
+gr.Interface(
+    fn=predict,
+    title=title,
+    description=description,
+    examples=examples,
+    inputs=["text", "state"],
+    outputs=["chatbot", "state"],
+    theme="finlaymacklon/boxy_violet",
+).launch()