Tuned H100/H200 triton fp8 block configs for fused_qkv_a_proj (#23939)

Signed-off-by: mgoin <mgoin64@gmail.com>
2025-08-29 13:28:35 -04:00
parent 4d7fe40fc0
commit b7adf94c4a
4 changed files with 271 additions and 3 deletions
--- a/benchmarks/kernels/benchmark_w8a8_block_fp8.py
+++ b/benchmarks/kernels/benchmark_w8a8_block_fp8.py
@@ -141,6 +141,7 @@ def get_weight_shapes(tp_size):
    # cannot TP
    total = [
        (512 + 64, 7168),
+        (2112, 7168),
        ((128 + 64) * 128, 7168),
        (128 * (128 + 128), 512),
        (7168, 16384),