Competition: https://www.kaggle.com/competitions/h2oai-predict-the-llm/overview
I'd like to once again thank h2o.ai, Kaggle and organizers for setting up this wonderful competition. While I have learnt a ton of new things working with decoder models, I had a lot of fun :-)
The decoder part of the solution is available here - Link.
Here are a few more resources if you're interested.
| Resource | Link |
|---|---|
| ML Experiments Logbook | Link |
| H2O LLMStudio Training yaml | Link |
| ML Experiment Runs | Link |
| Beating the Benchmark Solution | Link |
| Data Exploration via Clustering (which hinted the GroupKFold validation) | Link |