|
@@ -167,31 +167,30 @@ def benchmark_throughput(
|
|
|
)
|
|
|
total_examples = batch_size * num_clients * num_batches_per_client
|
|
|
|
|
|
- print("Benchmark finished, status:" + ["Success", "Failure"][benchmarking_failed.is_set()])
|
|
|
-
|
|
|
- print(
|
|
|
+ logger.info("Benchmark finished, status:" + ["Success", "Failure"][benchmarking_failed.is_set()])
|
|
|
+ logger.info(
|
|
|
f"Server parameters: num_experts={num_experts}, num_handlers={num_handlers}, "
|
|
|
f"max_batch_size={max_batch_size}, expert_cls={expert_cls}, hid_dim={hid_dim}, device={device}"
|
|
|
)
|
|
|
- print(
|
|
|
+ logger.info(
|
|
|
f"Client parameters: num_clients={num_clients}, num_batches_per_client={num_batches_per_client}, "
|
|
|
f"batch_size={batch_size}, backprop={backprop}"
|
|
|
)
|
|
|
|
|
|
- print("Results: ")
|
|
|
- print(
|
|
|
+ logger.info("Results: ")
|
|
|
+ logger.info(
|
|
|
f"\tServer startup took {time_between('began_launching_server', 'server_ready') :.3f} s. "
|
|
|
f"({time_between('began_launching_server', 'created_experts') :.3f} s. experts + "
|
|
|
f"{time_between('created_experts', 'server_ready') :.3f} s. networking)"
|
|
|
)
|
|
|
- print(f"\tProcessed {total_examples} examples in {time_between('server_ready', 'clients_finished') :.3f}")
|
|
|
- print(
|
|
|
+ logger.info(f"\tProcessed {total_examples} examples in {time_between('server_ready', 'clients_finished') :.3f}")
|
|
|
+ logger.info(
|
|
|
f"\tThroughput for {'forward + backward' if backprop else 'forward'} passes: "
|
|
|
f"{total_examples / time_between('server_ready', 'clients_finished') :.3f} samples / s."
|
|
|
)
|
|
|
- print(f"\tBenchmarking took {time_between('started', 'server_shutdown_finished') :.3f} s.")
|
|
|
+ logger.info(f"\tBenchmarking took {time_between('started', 'server_shutdown_finished') :.3f} s.")
|
|
|
if benchmarking_failed.is_set():
|
|
|
- print("Note: benchmark code failed, timing/memory results only indicate time till failure!")
|
|
|
+ logger.info("Note: benchmark code failed, timing/memory results only indicate time till failure!")
|
|
|
print_device_info(device)
|
|
|
sys.stdout.flush()
|
|
|
sys.stderr.flush()
|