[Flashinfer] Support Flashinfer TRTLLM FP8-qkv BF16/FP16-out Attention Kernel (#23647)

Signed-off-by: elvischenv <219235043+elvischenv@users.noreply.github.com>
2025-09-09 11:53:07 +08:00
parent b6fbc15634
commit bba1042c6f
5 changed files with 22 additions and 11 deletions
--- a/benchmarks/kernels/benchmark_trtllm_decode_attention.py
+++ b/benchmarks/kernels/benchmark_trtllm_decode_attention.py
@@ -259,6 +259,7 @@ if __name__ == "__main__":
        # (q_quant_dtype, kv_quant_dtype, o_quant_dtype)
        (None, None, None),
        (None, FP8_DTYPE, None),
+        (FP8_DTYPE, FP8_DTYPE, None),
        (FP8_DTYPE, FP8_DTYPE, FP8_DTYPE),
        (FP8_DTYPE, FP8_DTYPE, FP4_DTYPE),
    ]