Hongxia Yang f2beaa80c8 [ROCm][Quant] mxfp8 moe/linear gfx950 tuning for MiniMax-M3 (#45725) ...

Signed-off-by: Hongxia Yang <hongxia.yang@amd.com>

2026-06-16 18:50:40 +00:00

..

mixed_precision

[Kernel] Consolidate Marlin thread-tile padding across all dense Marlin paths (#45295)

2026-06-12 13:46:21 -07:00

mxfp4

Fix the E8M0 scale computation in the MXFP4 (W4A4) MOE CUTLASS kernel (#43557)

2026-06-15 06:04:54 -07:00

mxfp8

[ROCm][Quant] mxfp8 moe/linear gfx950 tuning for MiniMax-M3 (#45725)

2026-06-16 18:50:40 +00:00

nvfp4

Add the QuantizedActivation linear-kernel contract (#44260)

2026-06-12 13:48:15 -07:00

scaled_mm

[Perf] SM90 cutlass fp8 mm supports odd M by swap_ab, 180~290% kernel performance improvement (#44572)

2026-06-13 12:05:45 -07:00

__init__.py

[Model] Add MiniMax M3 support (#45381)

2026-06-16 01:01:25 +08:00

base.py

Add the QuantizedActivation linear-kernel contract (#44260)

2026-06-12 13:48:15 -07:00

zentorch_utils.py

[CPU][Zen] Route W8A8 and W4A16 linear inference through zentorch on AMD Zen CPUs (#41813)

2026-05-30 14:17:21 -05:00