Research Notebooks
Reproduce our training environment, evaluation pipelines, and dataset validations directly in Google Colab.
01. Environment Setup
Basic verification of the training stack (uv, torch, unsloth, trl) on a Colab T4.
02. Model Validation (E4B)
Load the base Gemma 4 E4B model in 4-bit and run inference checks.
03. Baseline Evaluation
Run the Layer 1 Canonical Baseline Benchmark against the unmodified model.
04. Full Training Pipeline
End-to-end execution of SFT and DPO on the curated datasets.