Skip to content

Conversation

@LeiWang1999
Copy link
Contributor

This pull request focuse on improving backend handling, updating data type mappings, and enhancing matrix multiplication operations. The most important changes are summarized below:

Backend Handling Improvements:

  • Modified the backend parameter in the __init__ method of bitblas/ops/general_matmul/__init__.py to use "tl" instead of "tir".
  • Updated the dispatch_tir method to dispatch in bitblas/ops/general_matmul/__init__.py.

Data Type Mappings:

  • Updated type mappings in bitblas/builder/wrapper/tl.py to use new type names for bfloat16, e4m3_float8, and e5m2_float8.

Matrix Multiplication Enhancements:

  • Added support for with_bias in bitblas/ops/general_matmul/tilelang/dense/matmul_simt.py by including bias shape and updating the main computation loop. [1] [2] [3] [4] [5]
  • Improved the retrieve_weight_shape method in bitblas/ops/general_matmul/__init__.py to handle cases where prim_func is None.

Testing Enhancements:

  • Added an assertion to check the backend type in testing/python/builder/test_backend_tir_builder.py.
  • Modified test_repack_from_gptq.py to include a package requirement decorator for auto_gptq and refactored the import statement. [1] [2]

@LeiWang1999
Copy link
Contributor Author

Luckily to get all tests passed, but we should also implement NF Format and implement relevant test case.

@LeiWang1999 LeiWang1999 merged commit 08b56e8 into microsoft:main Dec 17, 2024
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant