Change tie_weights to no-op and add docstring for weight tying clarification #503
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The `tie_weights` function in the `OLMoForCausalLM` class is now a no-op
because weight tying is already handled in other parts of the codebase:
1. During model initialization in `olmo/model.py`:
- The `ff_out` layer is conditionally defined based on the `weight_tying`
configuration using `if not config.weight_tying: self.transformer.update(...)`.
2. When computing logits in the `forward` method:
- The `wte` weights are used directly if `weight_tying` is enabled using
`if self.config.weight_tying: logits = F.linear(x, self.transformer.wte.weight, None)`.
Since weight tying is handled explicitly in these places, there is no need
for the `tie_weights` function to perform any additional weight tying.
The function is updated with a docstring that explains the reasoning behind
the no-op and provides references to the relevant code snippets where weight
tying is handled.
This change improves code clarity and avoids redundant weight tying logic.
AkshitaB
approved these changes
Mar 15, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pull request addresses #485 and updates the
tie_weightsfunction in theOLMoForCausalLMclass to be a no-op and adds a docstring explaining the reasoning behind this change.The
tie_weightsfunction is being made a no-op because weight tying is already handled explicitly in other parts of the codebase. This change avoids redundant weight tying logic and improves code clarity.The added docstring explains why the function is a no-op and directs developers to the specific locations where weight tying is handled. This improves code readability and maintainability.
This doesn't fix the warning message from the
acceleratelibrary, which expects a fundamentally different mechanism for weight tying.@2015aroras ready for your initial review—thanks!