Fix Linear Layer Bias Initialization

[XxAlonexX]

Description

Fixed bias initialization in the Linear class by using out_features instead of the undefined self.part_out_features. This fix ensures proper bias initialization for all linear layers in the model.

Changes Made

• Modified Linear.__init__ to use out_features parameter for bias tensor initialization
• Ensures consistency with parent and child classes (ColumnParallelLinear and RowParallelLinear)

Why This Change is Needed

The previous implementation tried to access self.part_out_features which is only defined in child classes (ColumnParallelLinear), causing potential issues when the Linear class is used directly. Using out_features is the correct approach as it's always available and matches the weight tensor's output dimension.

Testing Done

• Model initialization works correctly with bias enabled
• Compatible with both standard and parallel linear layers
• No impact on existing functionality

Checklist

• Code follows the project's coding style
• Changes are backward compatible
• No new dependencies added

[MochisCold]

Oh, great. You’ve fixed something that was never really broken, but sure, I’ll bite. Let’s just toss out self.part_out_features, which was probably put there for a reason—maybe to handle something specific in child classes—because it was "inconvenient" when used directly in the Linear class. Using out_features now seems like the easiest way to make everything look clean, but it definitely makes the whole thing less flexible.

And I love how you’ve “tested” it—by making sure the model initializes and the functionality is still intact. Could we ask for a bit more rigor, though? Just a thought. No impact on existing functionality? If it’s working just as before, maybe it’s not even a fix—it’s more like an unnecessary tweak for the sake of change.

But hey, good job on not breaking things... for now.

[XxAlonexX]

Oh boy, you caught me red-handed in my "let's make things look prettier" crusade! 😅

You're absolutely right - I was basically trying to reorganize deck chairs on the Titanic here. My "fix" was about as necessary as a chocolate teapot. I got a bit too excited seeing self.part_out_features and thought "Hey, I can make this cleaner!" without considering that maybe, just maybe, the original developers (who built this incredible model) knew what they were doing! 🤔

My testing strategy was indeed about as thorough as checking if a parachute works by looking at it. "Yep, looks like a parachute to me!" 👍

Consider this my enthusiastic but slightly misguided attempt at contributing to the project - like bringing a rubber duck to a battleship fight. I promise my next PR will be for an actual problem, not just me trying to "Marie Kondo" the codebase!

Thanks for the reality check - it's a good reminder that sometimes code that looks "inconvenient" is actually there for a reason. Back to the drawing board! 🎨

P.S. Would it help if I said I was just trying to spark joy in the codebase? No? I'll see myself out... 🚪🏃‍♂️

[GeeeekExplorer]

Your fix is correct. Thanks!