TRF026

A module whose forward only delegates to its single submodule adds nothing; inline it.

Default	Enabled
Scope	Models added on or after 2026-06-20
Source	mlinter/trf026.py
Show in terminal	mlinter --rule TRF026

What it does

In modeling_*.py and modular_*.py, flags a class that is not a PreTrainedModel subclass, defines only __init__ and forward, assigns exactly one self.<attr> in __init__, and whose forward body is exactly return self.<attr>(...) for that same attribute. A leading docstring is ignored. Classes with any other method, any additional attribute, or any statement before the return are left alone because they do work of their own.

Why is this bad?

The wrapper adds a level to every weight name and to _no_split_modules, tensor-parallel and pipeline plans, and to every conversion mapping, while contributing no computation. Readers then have to open one more class to discover that nothing happens in it, which is the single most common structural review comment on new models. PreTrainedModel subclasses are exempt: those exist for from_pretrained and the auto classes even when the forward only delegates.

Example

-class AcmeAtomTransformer(nn.Module):
-    def __init__(self, config):
-        super().__init__()
-        self.encoder = AcmeEncoder(config)
-
-    def forward(self, hidden_states, **kwargs):
-        return self.encoder(hidden_states, **kwargs)
-
 class AcmeModel(AcmePreTrainedModel):
     def __init__(self, config):
         super().__init__(config)
-        self.atom_transformer = AcmeAtomTransformer(config)
+        self.encoder = AcmeEncoder(config)

Suppressing this rule

Add a # trf-ignore: TRF026 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.