chat interface

app.py

@@ -1,73 +1,50 @@
 import spaces
 import torch
-import re
 import gradio as gr
 from threading import Thread
-from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# Install the necessary package for the model
 import subprocess
 
-# Install flash-attn for faster inference
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
+               shell=True)
 
-# Model and tokenizer setup
+# Initialize the tokenizer and model
 model_id = "vikhyatk/moondream2"
 revision = "2024-04-02"
 tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
 moondream = AutoModelForCausalLM.from_pretrained(
-    model_id, trust_remote_code=True, revision=revision,
+    model_id, revision=revision, trust_remote_code=True,
     torch_dtype=torch.bfloat16, device_map={"": "cuda"},
-    attn_implementation="flash_attention_2")
+    attn_implementation="flash_attention_2"
+)
 moondream.eval()
 
-# Function to generate responses
+
 @spaces.GPU(duration=10)
-def answer_question(img, prompt):
+def chatbot_response(img, text_input):
+    # Here we assume an encoded image processing if needed
     image_embeds = moondream.encode_image(img)
-    streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)
-    thread = Thread(
-        target=moondream.answer_question,
-        kwargs={
-            "image_embeds": image_embeds,
-            "question": prompt,
-            "tokenizer": tokenizer,
-            "streamer": streamer,
-        },
-    )
-    thread.start()
-    buffer = ""
-    for new_text in streamer:
-        buffer += new_text
-        yield buffer.strip()
-
-# Create the Gradio interface
-with gr.Blocks(theme="Monochrome") as demo:
-    gr.Markdown(
-        """
-        # AskMoondream: Moondream 2 Demonstration Space
-        Moondream2 is a 1.86B parameter model initialized with weights from SigLIP and Phi 1.5.
-        Modularity AI presents this open source huggingface space for running fast experimental inferences on Moondream2.
-        """
-    )
+    inputs = tokenizer.encode(text_input, return_tensors="pt")
+    outputs = moondream.generate(inputs, max_length=200)
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return response
 
-    # Chatbot layout
-    chatbot = gr.Chatbot()
 
-    # Image upload and prompt input
+# Setting up Gradio Interface
+with gr.Blocks(theme="Monochrome") as demo:
+    gr.Markdown("# AskMoondream Chatbot")
     with gr.Row():
         img = gr.Image(type="pil", label="Upload an Image")
-        prompt = gr.Textbox(label="Your Question", placeholder="Ask something about the image...", show_label=False)
-
-    # Send message button
-    send_btn = gr.Button("Send")
-
-    # Function to send message and get response
-    def send_message(history, prompt):
-        history.append((prompt, None))
-        response = answer_question(img.value, prompt)
-        history.append((None, response))
-        return history, ""  # Clear the input box
+        text_input = gr.Textbox(label="Ask a question or describe an image", placeholder="Type here...")
+    with gr.Row():
+        submit = gr.Button("Submit")
+        response = gr.TextArea(label="Response", placeholder="Moondream's response will appear here...")
 
-    send_btn.click(send_message, [chatbot, prompt], [chatbot, prompt])
-    prompt.submit(send_message, [chatbot, prompt], [chatbot, prompt])
+    # Define what happens when the user interacts with the interface
+    submit.click(chatbot_response, inputs=[img, text_input], outputs=response)
+    text_input.submit(chatbot_response, inputs=[img, text_input], outputs=response)
 
-demo.queue().launch()
+# Launch the demo
+demo.queue().launch()