LangChainHub

The LLMBashChain is a chain designed to take a natural language task description and generate a sequence of bash commands required to execute that task.

Input Variables

• question: A string describing the task you want to perform in bash.

The LLMCheckerChain is a specialized chain designed to improve the quality of answers to factual questions through a multi-step verification process.

When you run this chain, it performs the following workflow:

1. Drafting: Generates an initial draft answer to the question.
2. Assumption Extraction: Asks the model to list the assumptions underlying the draft statement.
3. Verification: Asks the model to determine if each assertion is true or false, providing explanations for false assertions.
4. Revision: Prompts the model to revise the original answer based on the findings from the verification step.

API URL Prompts are specialized prompts designed to construct an API URL to answer a specific question. These prompts require two specific inputs:

1. api_docs: The documentation for the target API.
2. question: The specific question being asked of the API.

You can use these prompts within an APIChain by loading them via the lc:// URI scheme.

from langchain.prompts import load_prompt
from langchain.chains import APIChain

llm = ...
api_docs = ...
prompt = load_prompt('lc://prompts/api/api_url/<file-name>')
chain = APIChain.from_llm_and_api_docs(llm, api_docs, api_url_prompt=prompt)

These prompts are specifically designed for summarization chains utilizing the stuff method. The prompt expects a single input variable:

• text: The raw text content that needs to be summarized.

You can load any prompt from the LangChainHub by using the lc:// prefix followed by the path relative to the langchain-hub repository.

For example, if a prompt is located at langchain-hub/prompts/qa/stuff/basic/prompt.yaml, you should use the path lc://prompts/qa/stuff/basic/prompt.yaml with the load_prompt function.

from langchain.prompts import load_prompt

prompt = load_prompt('lc://prompts/qa/stuff/basic/prompt.yaml')

These prompts are designed for question-answering tasks where the chain uses the refine method to iteratively improve an answer using multiple pieces of context.

Expected Inputs

The prompt expects the following input variables:

1. question: The original question to be answered.
2. existing_answer: The answer generated from previous documents.
3. context_str: The new piece of context used to refine the existing answer.

from langchain.prompts import load_prompt
from langchain.chains.qa_with_sources import load_qa_with_sources_chain

llm = ...
prompt = load_prompt('lc://prompts/qa_with_sources/refine/<file-name>')
chain = load_qa_with_sources_chain(llm, chain_type="refine", refine_prompt=prompt)

The llm_math prompt is designed to answer mathematical questions by optionally generating iPython syntax. This allows the model to execute code to solve complex math problems more accurately.

Expected Input:

• question: The user's mathematical question string.

from langchain.prompts import load_prompt
from langchain.chains import LLMMathChain

llm = ... # Initialize your LLM
prompt = load_prompt('lc://prompts/llm_math/<file-name>')
chain = LLMMathChain(llm=llm, prompt=prompt)

This prompt is designed for the 'reduce' step of a Map Reduce QA chain. It takes the individual summaries generated during the 'map' phase and reduces them into a final answer that includes relevant sources.

Expected Inputs:

• summaries: The summaries generated during the map step.
• question: The original question being answered.

from langchain.prompts import load_prompt
from langchain.chains.qa_with_sources import load_qa_with_sources_chain

llm = ...
# Replace <file-name> with the specific prompt file name
prompt = load_prompt('lc://prompts/qa_with_sources/map_reduce/reduce/<file-name>')
chain = load_qa_with_sources_chain(llm, chain_type="map_reduce", combine_prompt=prompt)

To upload a prompt to the LangChainHub, you must provide two files:

1. The prompt file: Supported formats are json, yaml, and python.
2. An associated README file: This provides a description and usage patterns. If you are uploading to an existing directory, a README should already exist.

Organization Rules:

• Prompts must be organized by use case within the directory structure.
• When adding to an existing folder, ensure the prompt serves the same use case and has the same inputs as existing prompts in that folder.