Checks / pattern
Deep attention stack without depth-scaled init
Check R33. Runs in the editor as you build, in CI through the
GitHub Action, and over the wire at
POST /api/v1/check. Milliseconds, before any GPU is billed.
info
pattern
R33
| Trigger | ≥ 8 attention layers stacked in one model. |
|---|---|
| Why | Residual-branch outputs add up with depth; unscaled init lets activation variance grow layer over layer. GPT-2/LLaMA-family models scale residual output projections by depth: N(0, 0.02 / √(2L)). |
| Source | GPT-2 (Radford et al. 2019) init convention; carried forward by LLaMA/Mistral. |
Why it is not a lint you can ignore
A structural mistake does not fail at review time and it does not fail at import
time. It fails when the module is constructed on the training node, after the job was queued and
the dataset was downloaded. That is why this runs before the spend and not after it.
Run this check on your own model
Free, no account needed
Every check
41 structural checks: 6 guardrail gates and 35 architecture advisor rules. See the full catalogue.
← R32 Default init assumes ReLU, saturating activation follows ยท R34 LM head width disagrees with embedding vocab →