Input in GraphQAChain Prompt is Malformed When Only 1 Triplet Is Found

[msp10003]

System Info

Mac OS Ventura 13.3.1 (a)
Python 3.10.8
LangChain 0.0.224

Who can help?

@hwchase17
@hinthornw

Information

• The official example notebooks/scripts
• My own modified scripts

Related Components

• LLMs/Chat Models
• Embedding Models
• Prompts / Prompt Templates / Prompt Selectors
• Output Parsers
• Document Loaders
• Vector Stores / Retrievers
• Memory
• Agents / Agent Executors
• Tools / Toolkits
• Chains
• Callbacks/Tracing
• Async

Reproduction

Reproduction Steps:

1. Run the following

    from langchain.llms import OpenAI
    from langchain.indexes import GraphIndexCreator
    from langchain.chains import GraphQAChain
    from langchain.prompts import PromptTemplate

    text = "Apple announced the Vision Pro in 2023."

    index_creator = GraphIndexCreator(llm=OpenAI(openai_api_key='{OPEN_AI_KEY_HERE}', temperature=0))
    graph = index_creator.from_text(text)
    chain = GraphQAChain.from_llm(
        OpenAI(temperature=0, openai_api_key='{OPEN_AI_KEY_HERE}'),
        graph=graph, 
        verbose=True
    )
    chain.run("When did Apple announce the Vision Pro?")

2. Observe the "Full Context" output in your terminal and notice that the two triplets are concatenated onto a single line with no spacing in between them.

I believe the issue is in the code here. When only 1 triplet is found in an iteration, .join does not add any \n characters, resulting in a context string with no separation between triplets.

Expected behavior

Expected: A multi-line string with each triplet text on its own line (delimited by "\n")

In the above repro steps, I would expect

Full Context:
Apple announced Vision Pro
Vision Pro was announced in 2023

[dosubot]

Answer generated by a 🤖

Answer

I understand that you're experiencing an issue with the GraphQAChain in LangChain where the input prompt is malformed when only one triplet is found. This is causing the triplets to be concatenated onto a single line without any spacing in between them.

The solution I propose involves modifying the way the context is joined in the GraphQAChain. Currently, when there's only one triplet, the .join method doesn't add any newline characters, which results in a context string without separation between triplets.

Here's how you can address this issue:

# This is the line of code you pointed out in your question
context = "\n".join(triplet.text for triplet in triplets)

This modification ensures that each triplet is on a new line, even if there's only one triplet. The "\n".join() method will join the triplets with a newline character in between each one, creating the expected multi-line string.

Please note that this is one potential solution based on the information provided. If this doesn't resolve your issue or contradicts your experience, I recommend checking if you're using the latest version of LangChain, as the issue might have been addressed in a more recent update.

If you believe this solution is correct and it's a bug that could impact other users, I encourage you to make a pull request with this change. Your contribution would be greatly appreciated by the LangChain community.

Sources

• langchain/chains/graph_qa/prompts.py

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

[Bearnardd]

Hi! It seems like an easy bug to fix, I am working on it :)