[transforms] TransformScheme.block_size, deprecate head_dim #466
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: Brian Dellabetta <bdellabe@redhat.com>
2 tasks
brian-dellabetta
changed the title
fynnsu
approved these changes
Sep 16, 2025
Thanks, yeah I played around with |
brian-dellabetta
added a commit
to vllm-project/llm-compressor
that referenced
this pull request
Sep 17, 2025
…field (#1806) SUMMARY: Resolves `INFERENG-1882` The research community [has pointed out](https://github.com/IST-DASLab/FP-Quant?tab=readme-ov-file#fp-format-quantization-harness) that the rotation/transform block size is important when performing transforms: > Key to efficiency is that the Hadamard block size matches the microscaling format group size (16 or 32) This exposes a new field on SpinQuantModifier and QuIPModifier to allow the user to set it to an arbitrary value, as long as the model's hidden_size and head_dim are both evenly divisible by it. - [x] Add to SpinQuant Modifier. Option to allow for different `transform_block_size`s for R1 vs. R2 can be added at a future time. - [x] Add to QuIPModifier. Option to allow for different `transform_block_size`s for U vs. V can be added at a future time. Merge in conjunction with: * neuralmagic/compressed-tensors#466 TEST PLAN: `transform_block_size` added to parameterized `tests/llmcompressor/modifiers/transform/(test_correctness.py|test_serialization.py)` --------- Signed-off-by: Brian Dellabetta <bdellabe@redhat.com>
dsikka
pushed a commit
to vllm-project/llm-compressor
that referenced
this pull request
Sep 21, 2025
…field (#1806) SUMMARY: Resolves `INFERENG-1882` The research community [has pointed out](https://github.com/IST-DASLab/FP-Quant?tab=readme-ov-file#fp-format-quantization-harness) that the rotation/transform block size is important when performing transforms: > Key to efficiency is that the Hadamard block size matches the microscaling format group size (16 or 32) This exposes a new field on SpinQuantModifier and QuIPModifier to allow the user to set it to an arbitrary value, as long as the model's hidden_size and head_dim are both evenly divisible by it. - [x] Add to SpinQuant Modifier. Option to allow for different `transform_block_size`s for R1 vs. R2 can be added at a future time. - [x] Add to QuIPModifier. Option to allow for different `transform_block_size`s for U vs. V can be added at a future time. Merge in conjunction with: * neuralmagic/compressed-tensors#466 TEST PLAN: `transform_block_size` added to parameterized `tests/llmcompressor/modifiers/transform/(test_correctness.py|test_serialization.py)` --------- Signed-off-by: Brian Dellabetta <bdellabe@redhat.com>
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 deprecates
TransformScheme.head_dimin favor ofTransformScheme.block_size, which is more meaningful now that we want to apply block-diagonal transforms with a user-configured block size.To be merged in conjunction with:
transform_block_sizefield vllm-project/llm-compressor#1806