bench() — pytorch Function Reference

Architecture documentation for the bench() function in dynamo_guard_eval.py from the pytorch codebase.

Function python DynamoBenchmarks Configuration called by 1

Entity Profile

DynamoBenchmarks→ Configuration→ bench() — pytorch Function Reference

Dependency Diagram

graph TD
  9c23b932_759b_0edb_2bd4_7d00709b8aca["bench()"]
  0bbb4b04_28a5_9cb6_a2bc_cd88b5f1ed81["main()"]
  0bbb4b04_28a5_9cb6_a2bc_cd88b5f1ed81 -->|calls| 9c23b932_759b_0edb_2bd4_7d00709b8aca
  style 9c23b932_759b_0edb_2bd4_7d00709b8aca fill:#6366f1,stroke:#818cf8,color:#fff

Relationship Graph

Source Code

benchmarks/dynamo/microbenchmarks/dynamo_guard_eval.py lines 24–38

def bench(name, fn):
    torch._dynamo.reset()
    inps = [[torch.randn(i) for _ in range(100)] for i in range(10, 101, 10)]

    def run_fn():
        for inp in inps:
            fn(*inp)

    start = time.perf_counter()
    for _ in range(3):
        run_fn()
    end = time.perf_counter()

    results = timeit.repeat(lambda: run_fn(), number=1000, repeat=10)
    print(f"{name} {np.median(results) * 1000:.1f}us (warmup={end - start:.1f}s)")