Data science expert

Inspiration

To have our own assistant who can do tedious task

What it does

Creates code by analysing your dataset and pattern discovery

How we built it

Fine tuning mistrals 7b instruct on 51k dataset on data science related questions

Challenges we ran into

Collecting dataset was first and foremost after using qlora fine tuneing ,hit a realization that to upload a qlora trained model deploy it there nis a need of gpu but no service provides a free gpu.

Accomplishments that we're proud of

Instead of just letting stayed choose another base model phi3-4k insteuct model fine tuned it on 10 k dataset among the 51 k orignal still it need 10 hrs and more to fine tune this time model was fine tuned using lora not qlora so that it can be deployed on a cpu

What we learned

Fine tuning deployment struggles and lot about llm why they require much time to train and how it can be reduced while not reducing much of quality

What's next for Data science expert

Adding a voice input method

Built With

• fastapi
• nextjs
• python