Dyanmic Quantization #15
Merged
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
Satrat
reviewed
Apr 22, 2024
| if "." not in submodule_name and submodule_name.endswith("_observer"): | ||
| # delete any observers that belong directly to this module | ||
| if getattr(submodule, "DYNAMIC", False): |
There was a problem hiding this comment.
nit: is there a reason we can't use submodule.DYNAMIC here instead of hardcoding?
|
Actually one more thought here, we should probably make sure "dynamic" or "static" make their way into the quantization config on export so its clear what the format is when reloading into vLLM. maybe we need to add an additional flag to the config for this? |
bfineran
commented
Apr 25, 2024
Satrat
reviewed
Apr 25, 2024
Satrat
approved these changes
Apr 25, 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Outdated suggestions cannot be applied.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 PR adds support for dynamic quantization. In dynamic quantization, scales and zero points are calculated on the fly at inference time for a particular tensor. This tradeoff with the extra compute gives us better results since the quantization params can fit the tensor directly rather than needing to be calibrated before hand.=
test_plan:
unit test included