From daa3e6dedbbd6bc57d939e4dd71f38d2e157a364 Mon Sep 17 00:00:00 2001 From: Harrison Chase Date: Sat, 13 May 2023 09:23:51 -0700 Subject: [PATCH] Harrison/prompt constructor methods (#4616) --- docs/modules/prompts.rst | 9 + docs/modules/prompts/getting_started.ipynb | 218 +++++++++++++++++++++ langchain/prompts/chat.py | 10 +- 3 files changed, 235 insertions(+), 2 deletions(-) create mode 100644 docs/modules/prompts/getting_started.ipynb diff --git a/docs/modules/prompts.rst b/docs/modules/prompts.rst index f72d09c7..bf5640a0 100644 --- a/docs/modules/prompts.rst +++ b/docs/modules/prompts.rst @@ -36,6 +36,15 @@ This is where output parsers come in. Output Parsers are responsible for (1) instructing the model how output should be formatted, (2) parsing output into the desired formatting (including retrying if necessary). +Getting Started +--------------- + +.. toctree:: + :maxdepth: 1 + + ./prompts/getting_started.ipynb + + Go Deeper --------- diff --git a/docs/modules/prompts/getting_started.ipynb b/docs/modules/prompts/getting_started.ipynb new file mode 100644 index 00000000..bd7758f1 --- /dev/null +++ b/docs/modules/prompts/getting_started.ipynb @@ -0,0 +1,218 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "3651e424", + "metadata": {}, + "source": [ + "# Getting Started\n", + "\n", + "This section contains everything related to prompts. A prompt is the value passed into the Language Model. This value can either be a string (for LLMs) or a list of messages (for Chat Models).\n", + "\n", + "The data types of these prompts are rather simple, but their construction is anything but. Value props of LangChain here include:\n", + "\n", + "- A standard interface for string prompts and message prompts\n", + "- A standard (to get started) interface for string prompt templates and message prompt templates\n", + "- Example Selectors: methods for inserting examples into the prompt for the language model to follow\n", + "- OutputParsers: methods for inserting instructions into the prompt as the format in which the language model should output information, as well as methods for then parsing that string output into a format.\n", + "\n", + "We have in depth documentation for specific types of string prompts, specific types of chat prompts, example selectors, and output parsers.\n", + "\n", + "Here, we cover a quick-start for a standard interface for getting started with simple prompts." + ] + }, + { + "cell_type": "markdown", + "id": "ff34414d", + "metadata": {}, + "source": [ + "## PromptTemplates\n", + "\n", + "PromptTemplates are responsible for constructing a prompt value. These PromptTemplates can do things like formatting, example selection, and more. At a high level, these are basically objects that expose a `format_prompt` method for constructing a prompt. Under the hood, ANYTHING can happen." + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "7ce42639", + "metadata": {}, + "outputs": [], + "source": [ + "from langchain.prompts import PromptTemplate, ChatPromptTemplate" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "5a178697", + "metadata": {}, + "outputs": [], + "source": [ + "string_prompt = PromptTemplate.from_template(\"tell me a joke about {subject}\")" + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "f4ef6d6b", + "metadata": {}, + "outputs": [], + "source": [ + "chat_prompt = ChatPromptTemplate.from_template(\"tell me a joke about {subject}\")" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "5f16c8f1", + "metadata": {}, + "outputs": [], + "source": [ + "string_prompt_value = string_prompt.format_prompt(subject=\"soccer\")" + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "863755ea", + "metadata": {}, + "outputs": [], + "source": [ + "chat_prompt_value = chat_prompt.format_prompt(subject=\"soccer\")" + ] + }, + { + "cell_type": "markdown", + "id": "8b3d8511", + "metadata": {}, + "source": [ + "## `to_string`\n", + "\n", + "This is what is called when passing to an LLM (which expects raw text)" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "1964a8a0", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "'tell me a joke about soccer'" + ] + }, + "execution_count": 18, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "string_prompt_value.to_string()" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "bf6c94e9", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "'Human: tell me a joke about soccer'" + ] + }, + "execution_count": 19, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "chat_prompt_value.to_string()" + ] + }, + { + "cell_type": "markdown", + "id": "c0825af8", + "metadata": {}, + "source": [ + "## `to_messages`\n", + "\n", + "This is what is called when passing to ChatModel (which expects a list of messages)" + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "e4da46f3", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "[HumanMessage(content='tell me a joke about soccer', additional_kwargs={}, example=False)]" + ] + }, + "execution_count": 20, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "string_prompt_value.to_messages()" + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "eae84b88", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "[HumanMessage(content='tell me a joke about soccer', additional_kwargs={}, example=False)]" + ] + }, + "execution_count": 21, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "chat_prompt_value.to_messages()" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "a34fa440", + "metadata": {}, + "outputs": [], + "source": [] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.1" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/langchain/prompts/chat.py b/langchain/prompts/chat.py index 9096b08d..af1fbef4 100644 --- a/langchain/prompts/chat.py +++ b/langchain/prompts/chat.py @@ -143,13 +143,19 @@ class ChatPromptTemplate(BaseChatPromptTemplate, ABC): input_variables: List[str] messages: List[Union[BaseMessagePromptTemplate, BaseMessage]] + @classmethod + def from_template(cls, template: str, **kwargs: Any) -> ChatPromptTemplate: + prompt_template = PromptTemplate.from_template(template, **kwargs) + message = HumanMessagePromptTemplate(prompt=prompt_template) + return cls.from_messages([message]) + @classmethod def from_role_strings( cls, string_messages: List[Tuple[str, str]] ) -> ChatPromptTemplate: messages = [ ChatMessagePromptTemplate( - content=PromptTemplate.from_template(template), role=role + prompt=PromptTemplate.from_template(template), role=role ) for role, template in string_messages ] @@ -160,7 +166,7 @@ class ChatPromptTemplate(BaseChatPromptTemplate, ABC): cls, string_messages: List[Tuple[Type[BaseMessagePromptTemplate], str]] ) -> ChatPromptTemplate: messages = [ - role(content=PromptTemplate.from_template(template)) + role(prompt=PromptTemplate.from_template(template)) for role, template in string_messages ] return cls.from_messages(messages)