[Bugfix] Correctly call cudaProfilerStop in benchmarks script (#14183)

Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca>
2025-03-06 19:42:49 -05:00
parent ad60bbb2b2
commit c34eeec58d
6 changed files with 5 additions and 6 deletions
--- a/benchmarks/kernels/benchmark_quant.py
+++ b/benchmarks/kernels/benchmark_quant.py
@@ -40,7 +40,7 @@ def main(num_tokens: int,

        end_time = time.perf_counter()
        if profile:
-            torch.cuda.cudart().cudaProfilerStart()
+            torch.cuda.cudart().cudaProfilerStop()
        return (end_time - start_time) / num_iters

    # Warmup.