Changes to make OpenAI key work for dataStoreAgent

[gaganahluwalia]

Description

Right now DatastoreAgent is failing to create a Cypher Query when using OpenAI Key, so a change in the code is required to parse the relationship returned by the OpenAI

Changelog

• This change involves changing the code inside semantic_layer_builder.py

[evpearce]

It seems like we might be fixing this later that we should be. Is the real problem not that OpenAI is outputting the nodes in the wrong format? in which case it would be better to fix the prompt that asks for the nodes and not just un-nest the nodes. Maybe we could provide an example of the format that we want in the prompt. This way switching between models would be more likely to work if we change it in the future.

[hsauve-scottlogic]

Nice work Gagan! I've tried with mistral and it breaks:
Error during data retrieval: 'relType'
I wonder if we should comment out the code for mistral, rather than remove it, and leave a comment?

[gaganahluwalia]

It seems like we might be fixing this later that we should be. Is the real problem not that OpenAI is outputting the nodes in the wrong format? in which case it would be better to fix the prompt that asks for the nodes and not just un-nest the nodes. Maybe we could provide an example of the format that we want in the prompt. This way switching between models would be more likely to work if we change it in the future.

This was an easy update to get it working with the Open AI key. With our dataset we could be facing some different problem, so we don't know how it's going to behave with the ESG dataset, that's why I didn't want to tinker the prompting as we may have to change that for our project later on.

[evpearce]

It seems like we might be fixing this later that we should be. Is the real problem not that OpenAI is outputting the nodes in the wrong format? in which case it would be better to fix the prompt that asks for the nodes and not just un-nest the nodes. Maybe we could provide an example of the format that we want in the prompt. This way switching between models would be more likely to work if we change it in the future.

This was an easy update to get it working with the Open AI key. With our dataset we could be facing some different problem, so we don't know how it's going to behave with the ESG dataset, that's why I didn't want to tinker the prompting as we may have to change that for our project later on.

Okay, could you add a comment explaining that it's temporary code please and why. Also we should probably change the code somehow so that it works for Mistral, commented out code like Helene said would be alright for now.

[hsauve-scottlogic]

The code works both with mistral and openai now, nice work!

[gaganahluwalia]

It seems like we might be fixing this later that we should be. Is the real problem not that OpenAI is outputting the nodes in the wrong format? in which case it would be better to fix the prompt that asks for the nodes and not just un-nest the nodes. Maybe we could provide an example of the format that we want in the prompt. This way switching between models would be more likely to work if we change it in the future.

This was an easy update to get it working with the Open AI key. With our dataset we could be facing some different problem, so we don't know how it's going to behave with the ESG dataset, that's why I didn't want to tinker the prompting as we may have to change that for our project later on.

Okay, could you add a comment explaining that it's temporary code please and why. Also we should probably change the code somehow so that it works for Mistral, commented out code like Helene said would be alright for now.

I have added a comment which tell what the change is and this is a temporary change and we should be changing it in future to have LLM return the same structure. For the rel type failure for Mistral, I had to make another change, which has also been commented. As of now the code is working for both OpenAI and Mistral.