NOTE: The views and opinions expressed in this blog are my own
Lets jump right in with an actual question I just gave to a Hana Cloud Database:
The response I got back was correct :
Ok so you're guessing I used ChatGPT, so it's either hallucinating and made a lucky guess, or perhaps I pre-loaded the data (context) into ChatGPT and it just summarised the results for me.
Nope.
I utilsed OpenAI's API's to converted the question into SQL, the SQL was then executed on a Hana Cloud instance and the results were then further summarised by OpenAI. LangChain's SQLDatabaseChain does most of the work.
Heres the full interaction:
Bonus Points to anyone who noticed I 'deliberately' miss-spelt Australia in the question. It wasn't a significant obstacle for OpenAI.
The SQL generated really does run in Hana:
The data flow is approximately like this:
To test it yourself you can try on Google Colab here:
Hana_LLM_Experiments
The main inspirations and references include:
llms-and-sql
sqllite examples
sqlalchemy for Hana
So as you can see it's now possible to ask Hana questions in Natural language.
I hope you found this glimpse of a potential future interesting.
In the next 2 blogs I explore some of the issues and potentials solutions of working with Large Lanaguage Models with sensitive Enterprise data:
Under the Hood – Text to SQL with LangChain, LLM’s and Hana
Into the SQL weeds - Fine-Tuning an LLM with Enterprise Data
What use case would you find this useful for?
SAP notes that posts about potential uses of generative AI and large language models are merely the individual poster's ideas and opinions, and do not represent SAP's official position or future development roadmap. SAP has no legal obligation or other commitment to pursue any course of business, or develop or release any functionality, mentioned in any post or related content on this website.
Lets jump right in with an actual question I just gave to a Hana Cloud Database:
The response I got back was correct :
Ok so you're guessing I used ChatGPT, so it's either hallucinating and made a lucky guess, or perhaps I pre-loaded the data (context) into ChatGPT and it just summarised the results for me.
Nope.
I utilsed OpenAI's API's to converted the question into SQL, the SQL was then executed on a Hana Cloud instance and the results were then further summarised by OpenAI. LangChain's SQLDatabaseChain does most of the work.
Heres the full interaction:
Click to Enlarge
Bonus Points to anyone who noticed I 'deliberately' miss-spelt Australia in the question. It wasn't a significant obstacle for OpenAI.
The SQL generated really does run in Hana:
Click to Enlarge
The data flow is approximately like this:
To test it yourself you can try on Google Colab here:
Hana_LLM_Experiments
The main inspirations and references include:
llms-and-sql
sqllite examples
sqlalchemy for Hana
So as you can see it's now possible to ask Hana questions in Natural language.
I hope you found this glimpse of a potential future interesting.
In the next 2 blogs I explore some of the issues and potentials solutions of working with Large Lanaguage Models with sensitive Enterprise data:
Under the Hood – Text to SQL with LangChain, LLM’s and Hana
Into the SQL weeds - Fine-Tuning an LLM with Enterprise Data
What use case would you find this useful for?
SAP notes that posts about potential uses of generative AI and large language models are merely the individual poster's ideas and opinions, and do not represent SAP's official position or future development roadmap. SAP has no legal obligation or other commitment to pursue any course of business, or develop or release any functionality, mentioned in any post or related content on this website.
- SAP Managed Tags:
