[Kernel] Marlin MoE: include SM 12.x in default arch list (#40923)

Signed-off-by: Tony Liu <tonyliu0512@gmail.com>
Co-authored-by: Tony Liu <tonyliu0512@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Shengqi Chen <harry-chen@outlook.com>
This commit is contained in:
tonyliu312
2026-05-28 15:30:26 +08:00
committed by GitHub
co-authored by Tony Liu mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Shengqi Chen
parent d6b48f928f
commit 6cc8577421
2 changed files with 26 additions and 6 deletions
+4 -2
View File
@@ -400,10 +400,12 @@ void marlin_mm(const void* A, const void* B, void* C, void* C_tmp, void* b_bias,
"Turing only support FP16 or INT8 activation.");
}
if (a_type == vllm::kFE4M3fn) {
TORCH_CHECK(major_capability * 10 + minor_capability >= 89,
"FP8 only support Ada Lovelace or newer GPUs.");
TORCH_CHECK(
major_capability * 10 + minor_capability == 89 ||
major_capability * 10 + minor_capability == 120,
"Marlin W4A8-FP8 only support SM89 or SM120 device (It is slower than "
major_capability == 12,
"Marlin W4A8-FP8 only support SM89 or SM12x device (It is slower than "
"Marlin W4A16 on other devices).");
}