Back to all updates

4 months ago

Speaker Session: Swarnamalya Mohan

Hi everyone, 

We have our second speaker session today (June 26) at 3:00 PM EDT! You can join with this link: https://meet.google.com/vrm-zkxc-rqh

This session will be led by Swarnamalya Mohan, a Senior Full Stack Developer (AI/ML) at FedEx.

 

Beyond Prompting: Train and Adapt Your Own Language Model

Beyond Prompting: Train and Adapt Your Own Language Model is a hands-on session that takes participants through the full journey of adapting a language model for a specialized task. The session begins with the foundations of model training, including how language models learn from data, how training examples are structured, and how loss and evaluation metrics guide improvement. Participants will then explore how to prepare a task-specific dataset, establish a baseline, and fine-tune a pretrained model using parameter-efficient techniques such as LoRA and QLoRA. The session will also demonstrate how to compare model behavior before and after training, evaluate whether the model has genuinely improved, and identify common issues such as overfitting, poor data quality, and unstable outputs. By the end, participants will understand how to move beyond prompt engineering and build a repeatable workflow for training, evaluating, and adapting language models without requiring massive compute resources.

 

Updates: 

  • Our first session led by Molly Cantillon (CEO of NOX, Forbes 30 under 30) has been posted on our YouTube channel! You can watch it here: https://www.youtube.com/watch?v=ZS2s6Y9Zoso
  • Please disregard the "Final call for submissions - FutureHacks" email. It was automatically sent by Devpost. The deadline to submit is still June 28 at 8:00 PM EDT. 

 

See you at the presentation!

The TechShare Team