[Performance] Reduce compression memory requirements via structure change #301
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
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
…-compression-memory
…-compression-memory
kylesayrs
changed the title
map_module_to_schemeSigned-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
…uantized-compression-memory
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
…uantized-compression-memory
…uce-quantized-compression-memory
kylesayrs
changed the title
dsikka
reviewed
Apr 23, 2025
| @@ -362,6 +364,29 @@ def get_unexpected_file_keys(self, model: Module) -> List[str]: | |||
|
|
|||
| return list(unexpected_keys) | |||
|
|
|||
| def apply_compression_status(self, model: Module) -> Module: | |||
There was a problem hiding this comment.
Where does this get applied?
There was a problem hiding this comment.
After this lands, we'll add model_compressor. apply_compression_status(model) within the overridden save_pretrained function. In the future, this will be used/folded into apply_compression_status
|
Could we add a test to compress a model with sparsity+quantization? |
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
…-compression-memory
…-compression-memory
|
LGTM pending conflict, good work! |
rahul-tuli
reviewed
Apr 30, 2025
| def apply_compression_status(self, model: Module): | ||
| if self.quantization_config is None: | ||
| for module in model.modules(): | ||
| module.quantization_status = QuantizationStatus.COMPRESSED |
There was a problem hiding this comment.
Could we run compression here too, for cases when we only have sparsity!
rahul-tuli
requested changes
Apr 30, 2025
brian-dellabetta
previously approved these changes
Apr 30, 2025
Comment on lines
+316
to
+317
| def unwrap_module_forward_quantized(module: Module): | ||
| delattr(module, "forward") # revert to class implementation |
There was a problem hiding this comment.
does this then expose the forward method on the Linear or CompressedLinear class? why do we want to delete the attr?
There was a problem hiding this comment.
Yes, this exposes the forward method of the class implementation
dsikka
reviewed
Apr 30, 2025
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
…-compression-memory
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.
Purpose
Allocating separate compressed state dict (current)
Replacing model structure (new)
Prerequisites
Changes
apply_compression_statuswhich converts theLinearlayers of a model toCompressedLinearsmodule_replace_dfs, which is more likely performant than replacing modules using existingreplace_moduleutilityCompressedLinearparameters on the execution deviceTesting
test_apply_compression_status