System Support
Entry
Summary
Timestamp
What LoRA Rank Actually Changes
A grounded resource on what rank affects, where it matters, and where it does not.
A grounded resource on what rank affects, where it matters, and where it does not.
How LoRA Alpha Actually Behaves
A compact guide to what alpha scales, why it matters with rank, and when it can make a run feel unstable.
A compact guide to what alpha scales, why it matters with rank, and when it can make a run feel unstable.
Dataset Shape Before Training
Why example tone, repetition, and edge-case coverage matter more than raw row count when shaping an adapter.
Why example tone, repetition, and edge-case coverage matter more than raw row count when shaping an adapter.
Choosing a Base Model
How to think about instruction models, reasoning models, and larger MoE baselines before you start adapting behavior.
How to think about instruction models, reasoning models, and larger MoE baselines before you start adapting behavior.
Exporting Adapters Cleanly
What export means in Atelier today, how checkpoint-backed adapters travel, and why portability matters.
What export means in Atelier today, how checkpoint-backed adapters travel, and why portability matters.
Evaluating Voice After Training
A practical pass on checking tone drift, persona stability, and how to know when a checkpoint is ready to share.
A practical pass on checking tone drift, persona stability, and how to know when a checkpoint is ready to share.