updated to claude-3

app.py

@@ -6,7 +6,7 @@ import os
 import arxiv
 import gradio as gr
 import requests
-from anthropic import AI_PROMPT, HUMAN_PROMPT, Anthropic
+from anthropic import Anthropic
 from arxiv_latex_extractor import get_paper_content
 from huggingface_hub import HfApi
 
@@ -70,43 +70,54 @@ def get_paper_from_huggingface(paper_id):
 
 
 class ContextualQA:
-    def __init__(self, client, model="claude-2.1"):
+    def __init__(self, client, model="claude-3-opus-20240229", initial_context=""):
         self.client = client
         self.model = model
-        self.context = ""
+        self.context = initial_context  # Set the initial context here
         self.questions = []
         self.responses = []
 
     def load_text(self, text):
-        self.context = text
+        self.context = text  # Update the context with new text
 
     def ask_question(self, question):
-        if self.questions:
-            # For the first question-answer pair, don't add HUMAN_PROMPT before the question
-            first_pair = f"Question: {self.questions[0]}\n{AI_PROMPT} Answer: {self.responses[0]}"
-            # For subsequent questions, include both HUMAN_PROMPT and AI_PROMPT
-            subsequent_pairs = "\n".join(
-                [
-                    f"{HUMAN_PROMPT} Question: {q}\n{AI_PROMPT} Answer: {a}"
-                    for q, a in zip(self.questions[1:], self.responses[1:])
-                ]
-            )
-            history_context = f"{first_pair}\n{subsequent_pairs}"
-        else:
-            history_context = ""
-
-        full_context = f"{self.context}\n\n{history_context}\n"
-
-        prompt = f"{HUMAN_PROMPT}  {full_context} {HUMAN_PROMPT} {question} {AI_PROMPT}"
-
-        response = self.client.completions.create(
-            prompt=prompt,
-            stop_sequences=[HUMAN_PROMPT],
-            max_tokens_to_sample=6000,
+        # Prepare the messages list with previous Q&A pairs and the current context
+        messages = [{
+            "role": "user",
+            "content": [{"type": "text", "text":  "Read the document bleow and answer the questions\n"+self.context}]
+        }]
+        messages.append({
+                "role": "assistant",
+                "content": [{"type": "text", "text": "The document is loaded. You can now ask questions."}]
+        })
+        
+        for q, a in zip(self.questions, self.responses):
+            messages.append({
+                "role": "user",
+                "content": [{"type": "text", "text": q}]
+            })
+            messages.append({
+                "role": "assistant",
+                "content": [{"type": "text", "text": a}]
+            })
+        
+        # Add the new question
+        messages.append({
+            "role": "user",
+            "content": [{"type": "text", "text": question}]
+        })
+
+        # Create the message with the system context and the list of messages
+        response = self.client.messages.create(
             model=self.model,
-            stream=False,
+            max_tokens=1024,
+            system=self.context,  # Pass the context directly as a string
+            messages=messages,
+            temperature=0  # Assuming you want deterministic responses
         )
-        answer = response.completion
+        
+        # Assuming the response object has a 'content' attribute that contains the answer
+        answer = response.content[0].text
         self.questions.append(question)
         self.responses.append(answer)
         return answer
@@ -186,7 +197,7 @@ def load_context(paper_id):
 
     # Initialize the Anthropic client and QA model
     client = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
-    qa_model = ContextualQA(client, model="claude-2.1")
+    qa_model = ContextualQA(client, model="claude-3-opus-20240229")
     context = f"{LEADING_PROMPT}\n{latex_source}"
     qa_model.load_text(context)
 
@@ -235,17 +246,10 @@ with gr.Blocks(
     gr.HTML(
         """
     <h1 style='text-align: center; font-size: 24px;'>
-        Explore ArXiv Papers in Depth with 🔥 <code>claude-2.1</code> 🔥- Ask Questions and Get Answers Instantly
+        Explore ArXiv Papers in Depth with 🔥 <code>claude-3-opus-20240229</code> 🔥- Ask Questions and Get Answers Instantly
         </h1>
         """
     )
-    # gr.HTML(
-    #     """
-    # <p style='text-align: justify; font-size: 18px; margin: 10px;'>
-    #     Explore the depths of ArXiv papers with our interactive app, powered by the advanced <code>claude-2.1</code> model. Ask detailed questions and get immediate, context-rich answers from academic papers.
-    # </p>
-    # """
-    # )
 
     gr.HTML(
         """
@@ -287,7 +291,7 @@ with gr.Blocks(
     gr.Markdown(
         "## Acknowledgements\n"
         "This project is made possible through the generous support of "
-        "[Anthropic](https://www.anthropic.com/), who provided free access to the `claude-2.1` API."
+        "[Anthropic](https://www.anthropic.com/), who provided free access to the `claude-3` API."
     )
 
     btn_load.click(load_context, inputs=[paper_id_input], outputs=[qa_model, chatbot])
@@ -307,6 +311,3 @@ with gr.Blocks(
     btn_clear.click(clear_context, outputs=[chatbot])
 
     demo.launch()
-
-# app.mount("/js", StaticFiles(directory="js"), name="js")
-# gr.mount_gradio_app(app, demo, path="/")