Appendices
Experiment log
Teaching Gemma to Code
Why this handbook
How agents work
1
The agent loop
2
Tools
3
Context and tokens
4
Prompts and skills
Measuring
5
Evaluation
6
Error analysis
Making it better
7
First failing test
8
Running Gemma locally
9
Fine-tuning with LoRA
Looking back
10
Results
Appendices
Experiment log
Glossary
Appendices
Experiment log
Experiment log
One row per run. Change one thing at a time.
Date
Version
What changed
Local score
Kaggle score
Notes
v0
Baseline
10
Results
Glossary