U2 dq mat mul rebase#36329
Draft
akodanka wants to merge 2 commits into
Draft
Conversation
Introduce a new graph-rewrite pass (DQMatMulCWuAsymm) that restructures the asymmetric dequantize-then-MatMul subgraph: cast(W) -> Subtract(Z) -> Multiply(S) -> MatMul(Act) into a form where the quantized weights remain at the MatMul input: MatMul(Act, cast(W)) -> Subtract(ReduceSum(Act)*Z) -> Multiply(S) This allows the NPU sub-compiler to fuse the Convert+MatMul into a native quant-MatMul kernel, while the zero-point and scale corrections are applied post-MatMul on small tensors. Supported weight types: u2, u4, u8, i4, i8 with f32 zero-point and scale of shape [N, 1] (channel-wise). Register the new matcher in Partitioner::optimize() alongside DQMatMulCWi. Signed-off-by: Anoob Anto Kodankandath <anoob.anto.kodankandath@intel.com>
The CWAI path in getPartitioning() was missing the optimize() call that the FOLD path already makes. Without it, the DQ graph-rewrite passes are never run. Add p.optimize so that NPUW_DQ-gated rewrites apply on the CWAI path. Signed-off-by: Anoob Anto Kodankandath <anoob.anto.kodankandath@intel.com>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Details:
Tickets:
AI Assistance: