Spaces:
Running
Running
| from langchain_core.prompts import PromptTemplate | |
| from typing import List | |
| import models | |
| from transformers import AutoTokenizer | |
| def format_prompt(prompt) -> PromptTemplate: | |
| # TODO: format the input prompt by using the model specific instruction template | |
| template = f""" | |
| <|begin_of_text|><|start_header_id|>system<|end_header_id|> | |
| You are a helpful assistant.<|eot_id|> | |
| <|start_header_id|>user<|end_header_id|> Before answering tell me if you are given an empty context or not then answer | |
| {prompt}<|eot_id|> | |
| <|start_header_id|>assistant<|end_header_id|> | |
| """ | |
| # raw_template = [ | |
| # {"role": "system", "content":"You are a helpful assistant." }, | |
| # {"role": "user", "content": "{{prompt}}"}, | |
| # ] | |
| # tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct") | |
| # formatted_template = tokenizer.apply_chat_template( | |
| # raw_template, | |
| # tokenize=False, | |
| # add_generation_prompt=True | |
| # ) | |
| prompt_template = PromptTemplate.from_template( | |
| # input_variables=["question"], the variables will be auto detected by langchain package | |
| template | |
| ) | |
| # TODO: return a langchain PromptTemplate | |
| return prompt_template | |
| def format_chat_history(messages: List[models.Message]): | |
| # TODO: implement format_chat_history to format | |
| # the list of Message into a text of chat history. | |
| chat_history = "" | |
| for msg in messages: | |
| chat_history += '{}:{}'.format(msg.type, msg.message) | |
| chat_history += "\n---\n" | |
| # combined all messages from the list for sending it to the model prompt. | |
| return chat_history | |
| # raise NotImplemented | |
| def format_context(docs: List[str]): | |
| # TODO: the output of the DataIndexer.search is a list of text, | |
| # so we need to concatenate that list into a text that can fit into | |
| # the rag_prompt_formatted. Implement format_context that takes a | |
| # like of strings and returns the context as one string. | |
| if not docs: | |
| return "" | |
| combined_text = "" | |
| combined_text = "\n\n---\n\n".join( | |
| doc.strip() for doc in docs if doc.strip() | |
| ) | |
| # raise NotImplemented | |
| return combined_text | |
| raw_prompt = "{question}" | |
| # TODO: Create the history_prompt prompt that will capture the question and the conversation history. | |
| # The history_prompt needs a {chat_history} placeholder and a {question} placeholder. | |
| history_prompt: str = """ | |
| Given the following conversation provide a helpful answer to the following up question. | |
| explain me the previous questions if I ask, | |
| Chat History: | |
| {chat_history} | |
| Follow Up Question: {question} | |
| helpful answer: | |
| """ | |
| # TODO: Create the standalone_prompt prompt that will capture the question and the chat history | |
| # to generate a standalone question. It needs a {chat_history} placeholder and a {question} placeholder, | |
| standalone_prompt: str = """ | |
| Given the following conversation and a follow up question, rephrase the | |
| follow up question to be a standalone question, in its original language. | |
| Chat History: | |
| {chat_history} | |
| Follow Up Input: {question} | |
| Standalone question: | |
| """ | |
| # TODO: Create the rag_prompt that will capture the context and the standalone question to generate | |
| # a final answer to the question. | |
| rag_prompt: str = """ | |
| Answer the question based only on the following context: | |
| {context} | |
| Question: {standalone_question} | |
| """ | |
| # TODO: create raw_prompt_formatted by using format_prompt | |
| raw_prompt_formatted = format_prompt(raw_prompt) | |
| raw_prompt = PromptTemplate.from_template(raw_prompt) | |
| # TODO: use format_prompt to create history_prompt_formatted | |
| history_prompt_formatted: PromptTemplate = format_prompt(history_prompt) | |
| # TODO: use format_prompt to create standalone_prompt_formatted | |
| standalone_prompt_formatted: PromptTemplate = format_prompt(standalone_prompt) | |
| # TODO: use format_prompt to create rag_prompt_formatted | |
| rag_prompt_formatted: PromptTemplate = format_prompt(rag_prompt) | |