Franc v1.0

app.py

@@ -6,6 +6,7 @@ from langchain import PromptTemplate
 from langchain.chains import RetrievalQA
 from langchain.embeddings.huggingface import HuggingFaceEmbeddings
 from langchain.llms import HuggingFaceEndpoint
+from langchain.memory import ConversationBufferWindowMemory
 from langchain.vectorstores import Pinecone
 from torch import cuda
 
@@ -47,18 +48,20 @@ def get_prompt_template(instruction, system_prompt):
 
 
 template = get_prompt_template(
-    """Use the following pieces of context to answer the question at the end.
+    """Use the following scientific literature to answer the question at the end.
+If the scientific literature is not relevant, please answer using your own knowledge about the topic:
 
 {context}
 
-Question: {question}
-Helpful Answer:""",
-    """Your name is Franc.
-You are a running coach and exercise physiologist.
-You communicate in the style of Hal Higdon, but you never reference Hal Higdon in your answers.
-You never uses hashtags.
-Your answers are always 512-character long or less.
-If you don't know the answer to a question, please don't share false information."""
+Question: {question}""",
+    """Your name is Franc. You are a running coach.
+You are assisting a runner with some of their questions.
+
+Reply in 10 sentences or less. Reply as a friend would.
+
+Here is a record of the conversation so far:
+
+{history}"""
 )
 
 endpoint_url = (
@@ -71,21 +74,34 @@ llm = HuggingFaceEndpoint(
     task="text-generation",
     model_kwargs={
         "max_new_tokens": 512,
-        "temperature": 0.4,
+        "temperature": 0.3,
         "repetition_penalty": 1.3,
         "return_full_text": True,
-    }
+    },
 )
 
+prompt = PromptTemplate(
+    template=template,
+    input_variables=["history", "context", "question"],
+)
+
+memory = ConversationBufferWindowMemory(
+    k=3,
+    memory_key="history",
+    input_key="question",
+    ai_prefix="Franc",
+    human_prefix="Runner",
+)
+
+memory.chat_memory.add_user_message("Hello, my name is Laurent!")
+
 rag_chain = RetrievalQA.from_chain_type(
     llm=llm,
     chain_type='stuff',
     retriever=vector_store.as_retriever(search_kwargs={'k': 5}),
     chain_type_kwargs={
-        "prompt": PromptTemplate(
-            template=template,
-            input_variables=["context", "question"],
-        ),
+        "prompt": prompt,
+        "memory": memory,
     },
 )