TRF012

_init_weights must use init primitives, not in-place operations on module weights.

Default	Enabled
Scope	All models
Source	mlinter/trf012.py
Show in terminal	mlinter --rule TRF012

What it does

Checks that _init_weights(self, module) does not use in-place operations (e.g. .normal_(), .zero_()) directly on module weights.

Why is this bad?

We rely on internal flags set on parameters to track whether they need re-initialization. In-place ops bypass this mechanism. Use the init primitives instead.

Example

+from transformers import initialization as init
+
 def _init_weights(self, module):
-    module.weight.normal_(mean=0.0, std=0.02)
+    init.normal_(module.weight, mean=0.0, std=0.02)

Suppressing this rule

Add a # trf-ignore: TRF012 comment on the flagged line or the line directly above it. See Suppressing rules for whole-file directives and when a suppression is the wrong answer.