google-labs-jules[bot] 93ea4797c0 Add initial support for Nvidia Blackwell (SM120) ...

This change introduces the necessary compiler flags and CMake configurations to enable support for the Nvidia Blackwell SM120 architecture.

- Modified deep_gemm/jit/compiler.py to include sm_120 and compute_120 flags for NVCC and NVRTC.
- Updated CMakeLists.txt to add the new architecture flags for the build process.

Further testing on Blackwell hardware is required to validate MMA instruction compatibility and overall performance.

2025-06-24 00:30:35 +00:00

..

include/deep_gemm

Fix illegal memory address when skipping -1 m indices (#113)

2025-06-16 10:44:31 +08:00

jit

Add initial support for Nvidia Blackwell (SM120)

2025-06-24 00:30:35 +00:00

jit_kernels

Revert "Fixed the bug in get_swizzle_mode function related to elem_size setting. (#115)"

2025-06-23 17:13:36 +08:00

__init__.py

Weight gradient kernels for dense and MoE models (#95)

2025-05-14 14:47:58 +08:00

utils.py

Weight gradient kernels for dense and MoE models (#95)

2025-05-14 14:47:58 +08:00