Squeezenet reshape outputs fix

[glenn-jocher]

@AyushExel

Signed-off-by: Glenn Jocher glenn.jocher@ultralytics.com

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced convolutional layer bias handling in YOLOv5's model utility.

📊 Key Changes

• Modified the nn.Conv2d constructor in the reshape_classifier_output function to correctly handle whether a bias should be applied.

🎯 Purpose & Impact

• Consistency: Ensures that when a convolutional layer is reshaped, the bias parameter is maintained according to the original configuration (presence or absence of bias).
• Reliability: Prior to this change, models that originally did not use bias could have a bias mistakenly introduced, which is fixed now.
• User Impact: Users will experience improved model manipulation functionality, particularly in custom object classification, with no unexpected behavior regarding bias in convolutional layers.