[ROCm][Perf] Enable W4A16 FlyDSL MoE (#44400)

Signed-off-by: amd-asalykov <asalykov@amd.com>
Signed-off-by: Amanzhol Salykov <asalykov@amd.com>
This commit is contained in:
Amanzhol Salykov
2026-06-14 00:14:39 -07:00
committed by GitHub
parent 9548a1887f
commit 725c3bc808
15 changed files with 2173 additions and 0 deletions
+2
View File
@@ -133,6 +133,7 @@ MoEBackend = Literal[
"humming",
"triton_unfused",
"aiter",
"flydsl",
"emulation",
]
@@ -186,6 +187,7 @@ class KernelConfig:
- "humming": Use Humming Mixed Precision kernels
- "triton_unfused": Use Triton unfused MoE kernels
- "aiter": Use AMD AITer kernels (ROCm only)
- "flydsl": Use AMD FlyDSL kernels (ROCm only)
- "emulation": use BF16/FP16 GEMM, dequantizing weights and
running QDQ on activations.
"""