This pull request includes several updates to the bitblas library, focusing on the matmul operations and their testing. The most important changes involve making parameters optional in the apply_config function, adding a new scheduler for fine-grained dequantization, and enhancing test coverage for the new scheduler.
This pull request includes several updates to the
bitblas
library, focusing on thematmul
operations and their testing. The most important changes involve making parameters optional in theapply_config
function, adding a new scheduler for fine-grained dequantization, and enhancing test coverage for the new scheduler.Enhancements to
apply_config
function:bitblas/ops/general_matmul/tilelang/dense/matmul_tensorcore.py
: Updatedapply_config
to accept optional parameters and added assertions to ensure required parameters are provided.Addition of Fine-Grained Dequantization Scheduler:
bitblas/ops/general_matmul/tilelang/dequantize/__init__.py
: IntroducedMatmulDequantizeFineGrainedScheduler
to the module.testing/python/operators/test_general_matmul_tilelang_kernel.py
: Updated imports to includeMatmulDequantizeFineGrainedScheduler
and added a new test functiontest_matmul_fine_grained_dequant_with_default
. [1] [2]Test Coverage Enhancements:
testing/python/operators/test_general_matmul_tilelang_kernel.py
: Removed commented-out code and added a new functionassert_matmul_fine_grained_dequant_with_default_correctness
to test the correctness of the fine-grained dequantization. [1] [2]testing/python/tilelang/test_tilelang_dequantize_gemm.py
: Re-enabled the main testing function call.