HighFile-scopePython Plug & play

Call `model.eval()` or `model.train()` After Loading a PyTorch Model

Verify that after loading a PyTorch model, either `model.eval()` or `model.train()` is called. Failing to do so can result in incorrect behavior, especially for layers like dropout and batch normalization.

Add to Mesrai Open workspace

Why this matters

When loading a model’s state in PyTorch, failing to set it to training (`train()`) or evaluation (`eval()`) mode can lead to incorrect behavior.

Bad vs. good

Side-by-side examples engineers can pattern-match during review.

Avoid

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth')) # Noncompliant: model.train() or model.eval() was not called.

Prefer

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth'))
model.eval()

More snippets

Bad

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth')) # Noncompliant: model.train() or model.eval() was not called.

Good

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth'))
model.eval()

Related rules

From the same buckets as this rule.

Critical
Avoid equality operators in loop termination conditions
Check if loops use equality operators (== or !=) in termination conditions. These can lead to infinite loops if the condition is never met exactly. Instead, use relational operators like < or > for safer loop termination.
error-handlingreadability-refactor
Critical
Avoid Returning Null in Non-Async Task Methods
Returning null from a Task or Task<T> method causes a NullReferenceException if awaited. Use Task.CompletedTask or Task.FromResult<T>(null) instead.
error-handlingreadability-refactor+1
Critical
Avoid Using Boolean Expressions in `except` Statements
Detect `except` clauses that use `or` or `and` to catch multiple exceptions. This does not work as intended. Recommend using a tuple of exceptions instead.
error-handlingstack-python

Bad vs. good

Side-by-side examples engineers can pattern-match during review.

Avoid

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth')) # Noncompliant: model.train() or model.eval() was not called.

Prefer

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth'))
model.eval()

More snippets

Bad

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth')) # Noncompliant: model.train() or model.eval() was not called.

Good

import torch
import torchvision.models as models

model = models.vgg16()
model.load_state_dict(torch.load('model_weights.pth'))
model.eval()

Related rules

From the same buckets as this rule.

Critical

Avoid equality operators in loop termination conditions

Check if loops use equality operators (== or !=) in termination conditions. These can lead to infinite loops if the condition is never met exactly. Instead, use relational operators like < or > for safer loop termination.

error-handlingreadability-refactor

Critical

Avoid Returning Null in Non-Async Task Methods

Returning null from a Task or Task<T> method causes a NullReferenceException if awaited. Use Task.CompletedTask or Task.FromResult<T>(null) instead.

error-handlingreadability-refactor+1

Critical

Avoid Using Boolean Expressions in `except` Statements

Detect `except` clauses that use `or` or `and` to catch multiple exceptions. This does not work as intended. Recommend using a tuple of exceptions instead.

error-handlingstack-python