chat interface

app.py

@@ -4,21 +4,22 @@ import re
 import gradio as gr
 from threading import Thread
 from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
-
 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)
 
+# Model and tokenizer setup
 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,
     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):
     image_embeds = moondream.encode_image(img)
@@ -33,14 +34,13 @@ def answer_question(img, prompt):
         },
     )
     thread.start()
-
     buffer = ""
     for new_text in streamer:
         buffer += new_text
         yield buffer.strip()
 
-
-with gr.Blocks(theme="Glass") as demo:
+# Create the Gradio interface
+with gr.Blocks(theme="Monochrome") as demo:
     gr.Markdown(
         """
         # AskMoondream: Moondream 2 Demonstration Space
@@ -48,13 +48,26 @@ with gr.Blocks(theme="Glass") as demo:
         Modularity AI presents this open source huggingface space for running fast experimental inferences on Moondream2.
         """
     )
-    with gr.Row():
-        prompt = gr.Textbox(label="Input", value="Describe this image.", scale=4)
-        submit = gr.Button("Submit")
+
+    # Chatbot layout
+    chatbot = gr.Chatbot()
+
+    # Image upload and prompt input
     with gr.Row():
         img = gr.Image(type="pil", label="Upload an Image")
-        output = gr.TextArea(label="Response")
-    submit.click(answer_question, [img, prompt], output)
-    prompt.submit(answer_question, [img, prompt], output)
+        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
+
+    send_btn.click(send_message, [chatbot, prompt], [chatbot, prompt])
+    prompt.submit(send_message, [chatbot, prompt], [chatbot, prompt])
 
-demo.queue().launch()
+demo.queue().launch()