[Kernel] CUTLASS grouped gemm fp8 MoE kernel (#13972)

Signed-off-by: ElizaWszola <eliza@neuralmagic.com> Signed-off-by: ElizaWszola <ewszola@redhat.com> Co-authored-by: Lucas Wilkinson <wilkinson.lucas@gmail.com>
2025-03-27 01:54:44 +01:00
parent 7a6d45bc8a
commit 9239bf718e
22 changed files with 2317 additions and 15 deletions
--- a/benchmarks/kernels/benchmark_shapes.py
+++ b/benchmarks/kernels/benchmark_shapes.py
@@ -75,3 +75,19 @@ WEIGHT_SHAPES = {
        [7168, 8192],
    ],
 }
+
+WEIGHT_SHAPES_MOE = {
+    "nm-testing/Mixtral-8x7B-Instruct-v0.1": [
+        [8, 2, 4096, 28672],
+        [8, 2, 14336, 4096],
+    ],
+    "nm-testing/deepseekv2-lite": [
+        [64, 6, 2048, 1408],
+    ],
+    "ibm-granite/granite-3.0-1b-a400m": [
+        [32, 8, 1024, 1024],
+    ],
+    "ibm-granite/granite-3.0-3b-a800m": [
+        [40, 8, 1024, 1536],
+    ],
+}