Move quantized weight handling out of the Weights class
#2194
Conversation
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
danieldk
force-pushed
the
refactor/quantizer-weights
branch
3 times, most recently
from
a93a398 to
eca19d4
Compare
Narsil
reviewed
Jul 8, 2024
There was a problem hiding this comment.
Nice change.
I expected a net negative amount of lines from this (or neutral) but I guess the extra interface make it a bit more verbose.
Regardless it's a nice change.
I think we can make the interface slightly simpler and remove the need for default checking in the weights load.
danieldk
force-pushed
the
refactor/quantizer-weights
branch
2 times, most recently
from
397657e to
8aaacc3
Compare
danieldk
force-pushed
the
refactor/quantizer-weights
branch
from
8aaacc3 to
2f19a24
Compare
Quantized weights were loaded in the `Weights` class, but this was getting quite unwieldy, where every higher level method to load weights was a long conditional to cover all the different quantizers. This change moves loading of quantized weights out of the `Weights` class. This is done by defining a simple `WeightsLoader` interface that is implemented by `Exl2WeightsLoader`, `GPTQWeightsLoader`, and `MarlinWeightsLoader`. These implementations are in the quantizers' respective modules. The `Weights` class provides the low-level load operations (such as loading tensors or sharded tensors), but delegates loads that need quantizer-specific weight processing to a loader. The loaders still use the low-level functionality provided by `Weights`. I initially tried making a hierarchy where a class like `GPTQWeights` would inherit from `Weights`. But it is not very flexible (e.g. does not work well with the new weight storage mock used in tests) and the implicit indirections made the code harder to follow.
danieldk
force-pushed
the
refactor/quantizer-weights
branch
from
2f19a24 to
800ee99
Compare
ErikKaum
pushed a commit
that referenced
this pull request
Jul 26, 2024
Quantized weights were loaded in the `Weights` class, but this was getting quite unwieldy, where every higher level method to load weights was a long conditional to cover all the different quantizers. This change moves loading of quantized weights out of the `Weights` class. This is done by defining a simple `WeightsLoader` interface that is implemented by `Exl2WeightsLoader`, `GPTQWeightsLoader`, and `MarlinWeightsLoader`. These implementations are in the quantizers' respective modules. The `Weights` class provides the low-level load operations (such as loading tensors or sharded tensors), but delegates loads that need quantizer-specific weight processing to a loader. The loaders still use the low-level functionality provided by `Weights`. I initially tried making a hierarchy where a class like `GPTQWeights` would inherit from `Weights`. But it is not very flexible (e.g. does not work well with the new weight storage mock used in tests) and the implicit indirections made the code harder to follow.
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.
What does this PR do?
Quantized weights were loaded in the
Weightsclass, but this was getting quite unwieldy, where every higher level method to load weights was a long conditional to cover all the different quantizers.This change moves loading of quantized weights out of the
Weightsclass. This is done by defining a simpleWeightsLoaderinterface that is implemented byExl2WeightsLoader,GPTQWeightsLoader, andMarlinWeightsLoader. These implementations are in the quantizers' respective modules. TheWeightsclass provides the low-level load operations (such as loading tensors or sharded tensors), but delegates loads that need quantizer-specific weight processing to a loader. The loaders still use the low-level functionality provided byWeights.I initially tried making a hierarchy where a class like
GPTQWeightswould inherit fromWeights. But it is not very flexible (e.g. does not work well with the new weight storage mock used in tests) and the implicit indirections made the code harder to follow.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.