Refactor data structures to reduce memory usage#142
Merged
copybara-service[bot] merged 3 commits intogoogle:devfrom Apr 10, 2024
Merged
Refactor data structures to reduce memory usage#142copybara-service[bot] merged 3 commits intogoogle:devfrom
copybara-service[bot] merged 3 commits intogoogle:devfrom
Conversation
jan-wassenberg
requested changes
Apr 10, 2024
Member
jan-wassenberg
left a comment
jan-wassenberg
left a comment
There was a problem hiding this comment.
Nice, thanks for resolving this TODO :) Some suggestions:
gemma/gemma.cc
Outdated
| static constexpr size_t kGatingEinsumWSize = 2 * kFFHiddenDim * kModelDim; | ||
| static constexpr size_t kConv1dWidth = TConfig::kConv1dWidth; | ||
| static constexpr bool kFFBiases = TConfig::kFFBiases; | ||
| static constexpr size_t kAOBiaseDim = |
Member
There was a problem hiding this comment.
Typo (extra e), maybe instead kAttnBiasDim?
gemma/gemma.cc
Outdated
| }; | ||
|
|
||
| struct { | ||
| ArrayT<float, kGriffinDim * kGriffinDim> griffin_linear_x_w; |
Member
There was a problem hiding this comment.
Consider naming the struct member griffin, so we can remove the griffin_ prefix from its names?
Contributor
Author
There was a problem hiding this comment.
I agree, your solution should be better.
gemma/gemma.cc
Outdated
| static constexpr size_t kGatingEinsumWSize = TLayer::kGatingEinsumWSize; | ||
| static constexpr size_t kConv1dWidth = TLayer::kConv1dWidth; | ||
| static constexpr bool kFFBiases = TLayer::kFFBiases; | ||
| static constexpr size_t kAOBiaseDim = TLayer::kAOBiaseDim; |
jan-wassenberg
approved these changes
Apr 10, 2024
Member
jan-wassenberg
left a comment
jan-wassenberg
left a comment
There was a problem hiding this comment.
Nice, thanks for making the change :)
jan-wassenberg
added
the
copybara-import
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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 PR moves the Gemma attention weights and Griffin weights into a union to share memory and reduces the additional overhead when not using the Griffin model.