Support exact INT8 matrix multiplication via INT4 building blocks in QWIX. - #371
Merged
Merged
Conversation
copybara-service
Bot
force-pushed
the
test_981365884
branch
4 times, most recently
from
September 17, 2026 17:13
98d7abf to
a747767
Compare
copybara-service
Bot
changed the title
copybara-service
Bot
force-pushed
the
test_981365884
branch
4 times, most recently
from
September 22, 2026 06:03
8e3c773 to
3b64d14
Compare
copybara-service
Bot
changed the title
copybara-service
Bot
force-pushed
the
test_981365884
branch
16 times, most recently
from
September 25, 2026 19:02
1b6c240 to
ebcb408
Compare
…QWIX. Enables exact signed INT8 matrix multiplication emulated through native INT4 sub-byte passes: - Support exact signed INT8 multi-pass modes: - four_pass_int4: Full cross product via 4 native INT4 passes (Ah Bh + Ah Bl + Al Bh + Al Bl). - INT8 inputs are decomposed into high and low 4-bit nibbles with algebraic offset adjustments. - Add unit tests verifying exact numerical equivalence (0 max diff) to signed INT8 GEMM across random and boundary inputs. ### INT8 Multi-Pass SQNR Benchmark Table (512x512, Channel-wise / Tensor Scaling) | Strategy | GEMMs | Compute DType | Execution Target | SQNR (dB) | Rel Error (%) | Equivalence vs INT8 | | :---| :---: | :---| :---| :---| :---: | :---| | 1-Pass INT8 Baseline | 1 | int32 / float32 | CPU Reference | 38.29 dB | 1.22% | Exact INT8 baseline | | Signed INT8 via INT4 Emulated Int | 4 | float8_e4m3fn | Physical TPU (FP8 Path) | 37.61 dB | 1.32% | Exact product on TPU FP8 MXU | PiperOrigin-RevId: 988453232
copybara-service
Bot
force-pushed
the
test_981365884
branch
from
September 25, 2026 20:12
ebcb408 to
5e57079
Compare
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.
Support exact INT8 matrix multiplication via INT4 building blocks in QWIX.
Enables exact signed INT8 matrix multiplication emulated through native INT4 sub-byte passes:
INT8 Multi-Pass SQNR Benchmark Table (512x512, Channel-wise / Tensor Scaling)