using llama instead

app.py

@@ -1,23 +1,69 @@
 import streamlit as st
-from PIL import Image
-import torch
+import base64
 
-from transformers import Blip2Processor, Blip2ForConditionalGeneration
+from huggingface_hub import InferenceClient
 
-processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
-model = Blip2ForConditionalGeneration.from_pretrained(
-    "Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16
+client = InferenceClient(api_key=st.secrets["HF_TOKEN"])
+
+def classify_picture(image_url):
+	messages = [
+		{
+			"role": "user",
+			"content": [
+				{
+					"type": "image_url",
+					"image_url": {
+						"url": image_url
+					}
+				},
+				{
+					"type": "text",
+					"text": """Based upon the image, at what sort of location or in where is this participant most likely attending their online meeting? Be more specific rather than general. Answer in one sentence in the format of: "This participant is most likely attending their online meeting [location]."""
+				},
+			]
+		}
+	]
+
+	stream = client.chat.completions.create(
+		model="meta-llama/Llama-3.2-11B-Vision-Instruct", 
+		messages=messages, 
+		max_tokens=100,
+		stream=True
+	)
+
+	result = ""
+	for chunk in stream:
+		result += chunk.choices[0].delta.content
+	return result
+
+st.title("Online Meeting Location Classifier")
+
+option = st.selectbox(
+    "Target Image",
+    ("Camera", "Demo 1", "Demo 2", "Demo 3", "Demo 4"),
+	index=None,
+    placeholder="Select Image...",
 )
 
+if option == "Camera":
+	enable = st.checkbox("Enable camera")
+	picture_buffer = st.camera_input("Take a picture")
 
-enable = st.checkbox("Enable camera")
-picture = st.camera_input("Take a picture", disabled=not enable)
+	if picture_buffer is not None:
+		image_url = f"data:image/jpeg;base64,{base64.b64encode(picture_buffer.getvalue()).decode("utf-8")}"
 
-if picture:
-	image = Image.open(picture)
-	prompt = "Question: At what location is this person most likely attending this online meeting? Answer:"
-	inputs = processor(images=image, text=prompt, return_tensors="pt")
+		if st.button("Generate Location"):
+			st.write(classify_picture(image_url))
+elif option is not None:
+	if option == "Demo 1":
+		image_url = "https://media.istockphoto.com/id/1353209475/video/happy-indian-business-man-talking-to-webcam-having-virtual-meeting-in-office.jpg?s=640x640&k=20&c=DUEzmtpErdZxDavYMOmLuKExEWIVWUusRUh9wYd20j0="
+	elif option == "Demo 2":
+		image_url = "https://cdn.prod.website-files.com/62180286278929909e43b116/63c216577ab752f188142733_Using%20a%20Custom%20Zoom%20Background.jpg"
+	elif option == "Demo 3":
+		image_url = "https://media.istockphoto.com/id/1319103417/photo/inspired-female-teacher-in-headphones-meet-students-online-give-class.jpg?s=612x612&w=0&k=20&c=cSGIQIwPFpVi7m4MR2_SLlwDcpBELdnNnLNgdtH6Bak="
+	elif option == "Demo 4":
+		image_url = "https://www.rivertribe.co.uk/wp-content/uploads/2020/04/EV1Twz9XgAExEJ5-e1587484451269.jpg"
+	st.image(image_url, use_column_width=True)
 
-	generated_ids = model.generate(**inputs)
-	generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
-	st.write(generated_text)
+	if st.button("Generate Location"):
+		st.write(classify_picture(image_url))