global embeddings & qa chain

app.py

@@ -10,10 +10,19 @@ from langchain.llms import HuggingFaceHub
 from langchain.chains import ConversationalRetrievalChain
 
 
+embeddings = None
+qa_chain = None
+
+
 def load_embeddings():
-    print("loading embeddings...")
-    model_name = os.environ['HUGGINGFACEHUB_EMBEDDINGS_MODEL_NAME']
-    return HuggingFaceInstructEmbeddings(model_name=model_name)
+    global embeddings
+
+    if not embeddings:
+        print("loading embeddings...")
+        model_name = os.environ['HUGGINGFACEHUB_EMBEDDINGS_MODEL_NAME']
+        embeddings = HuggingFaceInstructEmbeddings(model_name=model_name)
+
+    return embeddings
 
 
 def split_file(file, chunk_size, chunk_overlap):
@@ -73,6 +82,10 @@ def load_vectordb(file_name):
 
 def create_qa_chain(collection_name, temperature, max_length):
     print('creating qa chain...', collection_name, temperature, max_length)
+    if not collection_name:
+        return
+
+    global qa_chain
     memory = ConversationBufferMemory(
         memory_key="chat_history", return_messages=True)
     llm = HuggingFaceHub(
@@ -80,7 +93,8 @@ def create_qa_chain(collection_name, temperature, max_length):
         model_kwargs={"temperature": temperature, "max_length": max_length}
     )
     vectordb = load_vectordb(collection_name)
-    return ConversationalRetrievalChain.from_llm(llm=llm, retriever=vectordb.as_retriever(), memory=memory)
+    qa_chain = ConversationalRetrievalChain.from_llm(
+        llm=llm, retriever=vectordb.as_retriever(), memory=memory)
 
 
 def submit_message(bot_history, text):
@@ -88,12 +102,12 @@ def submit_message(bot_history, text):
     return bot_history, ""
 
 
-def bot(bot_history, collection_name, temperature, max_length):
-    qa = create_qa_chain(collection_name, temperature, max_length)
-    print(qa, bot_history[-1][1])
-    qa.run(bot_history[-1][0])
-
-    bot_history[-1][1] = 'so cool!'
+def bot(bot_history):
+    global qa_chain
+    print(qa_chain, bot_history[-1][1])
+    result = qa_chain.run(bot_history[-1][0])
+    print(result)
+    bot_history[-1][1] = result
     return bot_history
 
 
@@ -122,7 +136,8 @@ with gr.Blocks() as demo:
                     choices, value=choices[0] if choices else None, label="Document")
                 temperature = gr.Slider(
                     0.0, 1.0, value=0.5, step=0.05, label="Temperature")
-                max_length = gr.Slider(20, 1000, value=64, label="Max Length")
+                max_length = gr.Slider(
+                    20, 1000, value=100, step=10, label="Max Length")
 
             with gr.Column(scale=0.5):
                 chatbot = gr.Chatbot([], elem_id="chatbot").style(height=550)
@@ -130,11 +145,19 @@ with gr.Blocks() as demo:
                     show_label=False, placeholder="Ask me anything!")
                 clear = gr.Button("Clear")
 
-    process.click(process_file, [upload, chunk_size,
-                  chunk_overlap], [result, collection])
+    process.click(
+        process_file,
+        [upload, chunk_size, chunk_overlap],
+        [result, collection]
+    )
+
+    create_qa_chain(collection.value, temperature.value, max_length.value)
+    collection.change(create_qa_chain, [collection, temperature, max_length])
+    temperature.change(create_qa_chain, [collection, temperature, max_length])
+    max_length.change(create_qa_chain, [collection, temperature, max_length])
 
     message.submit(submit_message, [chatbot, message], [chatbot, message]).then(
-        bot, [chatbot, collection, temperature, max_length], chatbot
+        bot, chatbot, chatbot
     )
     clear.click(clear_bot, None, chatbot)