Added GPT-4(untested)

app.py

@@ -5,19 +5,35 @@ import gradio as gr
 from threading import Lock
 
 from langchain.llms import OpenAI
-from langchain.chains import ChatVectorDBChain
+from langchain.chat_models import ChatOpenAI
+from langchain.chains import ChatVectorDBChain, ConversationalRetrievalChain
 from template import QA_PROMPT, CONDENSE_QUESTION_PROMPT
 from pdf2vectorstore import convert_to_vectorstore
 
 def get_chain(api_key, vectorstore, model_name):
-    llm = OpenAI(model_name = model_name, temperature=0,  openai_api_key=api_key)
-    qa_chain = ChatVectorDBChain.from_llm(
-        llm,
-        vectorstore,
-        qa_prompt=QA_PROMPT,
-        condense_question_prompt=CONDENSE_QUESTION_PROMPT,
-    )
-    return qa_chain
+    if model_name == "gpt-4":
+        llm = ChatOpenAI(model_name = model_name, temperature=0,  openai_api_key=api_key)
+        retriever = vectorstore.as_retriever()
+        retriever.search_kwargs['distance_metric'] = 'cos'
+        retriever.search_kwargs['fetch_k'] = 100
+        retriever.search_kwargs['maximal_marginal_relevance'] = True
+        retriever.search_kwargs['k'] = 10
+        qa_chain = ConversationalRetrievalChain.from_llm(
+            llm,
+            retriever,
+            qa_prompt=QA_PROMPT,
+            condense_question_prompt=CONDENSE_QUESTION_PROMPT,
+        )
+        return qa_chain
+    else:
+        llm = OpenAI(model_name = model_name, temperature=0,  openai_api_key=api_key)
+        qa_chain = ChatVectorDBChain.from_llm(
+            llm,
+            vectorstore,
+            qa_prompt=QA_PROMPT,
+            condense_question_prompt=CONDENSE_QUESTION_PROMPT,
+        )
+        return qa_chain
 
 def set_openai_api_key(api_key: str, vectorstore, model_name: str):
     if api_key:
@@ -108,8 +124,8 @@ with block:
             )
         with gr.Column(width="auto"):
             model_dropdown = gr.Dropdown(
-                label="Choose a model (GPT-4 coming soon!)",
-                choices=["gpt-3.5-turbo"],
+                label="Choose a model",
+                choices=["gpt-3.5-turbo", "gpt-4"],
             )
 
     chatbot = gr.Chatbot()
@@ -138,8 +154,9 @@ with block:
                 <p>ArxivGPT is a chatbot that answers questions about research papers. It uses a pretrained GPT-3.5 model to generate answers.</p>
                 <p>Currently, it can answer questions about the paper you just linked.</p>
                 <p>It's still in development, so please report any bugs you find. It can take up to a minute to start a conversation for every new paper as there is a parsing delay.</p>
-                <p>The answers can be quite limited as there is a 4096 token limit for GPT-3.5, hence waiting for GPT-4 access to upgrade.</p>
-                <p>Possible upgrades coming up: GPT-4, faster parsing, status messages, other research paper hubs.</p>
+                <p>The answers can be quite limited as there is a 4096 token limit for GPT-3.5, hence wait for GPT-4 access for better quality.</p>
+                <p>If you don't get a response for GPT-4, it is likely that you don't have API access, try 3.5</p>
+                <p>Possible upgrades coming up: faster parsing, status messages, other research paper hubs.</p>
             </div>
             <style>
                 p {