Pramit Choudhary

``` /* [CODE] SELECT AVG(sleep_health_and_lifestyle_study.sleep_duration) AS average_sleep_duration, sleep_health_and_lifestyle_study.gender FROM sleep_health_and_lifestyle_study GROUP BY sleep_health_and_lifestyle_study.gender LIMIT 100; */ SELECT ``` Fix above response.

- Validate dialect correctness for snowflake. - Validate workflow convenience cc @racoss

From understanding there r 2 parts of it, 1. LLM-as-a-judge, to generate reward score. As first step evaluate this strategy can help during re-generation request with exhaustive beam search. 2....

Add support for local/other LLMs

1

Experimentation in progress - NSQL, https://huggingface.co/NumbersStation/nsql-6B dataset used:  Initial benchmark shared:  -...

https://github.com/jupyterlab/jupyter-ai via @mmalohlava

Reference: https://docs.google.com/spreadsheets/d/1wM-k7fifYL7jLA2r0rforp4-sc3Tt9NF4pOxQfxQy0w/edit?usp=sharing Other datasets: (Known, may have been trained on) https://github.com/chiahsuan156/KaggleDBQA https://github.com/datadotworld/cwd-benchmark-data via @mmalohlava https://huggingface.co/datasets/Clinton/Text-to-sql-v1 https://huggingface.co/datasets/b-mc2/sql-create-context

Schema selection/linking

- Scalability - Better functionality

- Additional re-ranking metric - Vector DB for better caching and scalability