enhancementhelp wanted
Metriche repository
- Star
- (4313 stelle)
- Metriche merge PR
- (Metriche PR in attesa)
Descrizione
🚀 Feature
PyTorch utils module provides ThroughputBenchmark since 1.2.0
>>> from torch.utils import ThroughputBenchmark
>>> bench = ThroughputBenchmark(my_module)
>>> # Pre-populate benchmark's data set with the inputs
>>> for input in inputs:
# Both args and kwargs work, same as any PyTorch Module / ScriptModule
bench.add_input(input[0], x2=input[1])
>>> # Inputs supplied above are randomly used during the execution
>>> stats = bench.benchmark(
num_calling_threads=4,
num_warmup_iters = 100,
num_iters = 1000,
)
>>> print("Avg latency (ms): {}".format(stats.latency_avg_ms))
>>> print("Number of iterations: {}".format(stats.num_iters))
It would be interesting to provide a tiny wrapper over this to simplify usage with ignite.