Download repochat/chain.py from rcs3/Codebot: direct link, hf CLI and curl.
- Browser
- Download file 1.84 kB
-
https://huggingface.co/spaces/rcs3/Codebot/resolve/main/repochat/chain.py
- Command line
-
hf download hf://spaces/rcs3/Codebot/repochat/chain.py
-
curl -L -o chain.py https://huggingface.co/spaces/rcs3/Codebot/resolve/main/repochat/chain.py
1.84 kB
| from langchain.prompts import PromptTemplate, ChatPromptTemplate | |
| from langchain.memory import ConversationBufferWindowMemory | |
| from langchain.chains import ConversationalRetrievalChain | |
| def response_chain(db, llm): | |
| """ | |
| Build a conversational RAG chain over the code vector store. | |
| Uses a sliding window memory (last 6 turns) to keep context fresh. | |
| """ | |
| retriever = db.as_retriever( | |
| search_type="similarity", | |
| search_kwargs={"k": 4}, | |
| ) | |
| memory = ConversationBufferWindowMemory( | |
| k=6, | |
| memory_key="chat_history", | |
| return_messages=True, | |
| output_key="answer", | |
| ) | |
| system_prompt = ( | |
| "You are an expert software engineer assistant. " | |
| "You have been given relevant code snippets from a GitHub repository as context. " | |
| "Answer the user's question accurately and concisely, referencing the code where useful. " | |
| "If the answer is not in the provided context, say so clearly. " | |
| "Always format code with proper markdown code blocks.\n\n" | |
| "Context:\n{context}" | |
| ) | |
| qa_prompt = ChatPromptTemplate.from_messages( | |
| [ | |
| ("system", system_prompt), | |
| ("human", "{question}"), | |
| ] | |
| ) | |
| condense_prompt = PromptTemplate.from_template( | |
| "Given the chat history and a follow-up question, " | |
| "rewrite the follow-up as a standalone question.\n\n" | |
| "Chat History:\n{chat_history}\n\n" | |
| "Follow-up: {question}\n\n" | |
| "Standalone question:" | |
| ) | |
| qa_chain = ConversationalRetrievalChain.from_llm( | |
| llm=llm, | |
| retriever=retriever, | |
| memory=memory, | |
| return_source_documents=False, | |
| combine_docs_chain_kwargs={"prompt": qa_prompt}, | |
| condense_question_prompt=condense_prompt, | |
| verbose=False, | |
| ) | |
| return qa_chain |