Speed up and monitor inference

Track, profile and monitor inference to optimize latency and throughput. For any model and deployment.

Sign up for a free account to get an API key.
Add Graphsignal Profiler to any script or notebook.
Аnalyze ML speed and compute.

import graphsignal
from graphsignal.profilers.pytorch import profile_inference

graphsignal.configure(
    api_key='my_api_key', workload_name='my_inference')

for x in data:
  with profile_inference():
    preds = model(x)

See Quick Start Guide for instructions and more integrations.

Natively supported frameworks and technologies

Inference profiling for full visibility

Benchmark speed

Compare how changes in parameters, metrics and compute impact speed.

Analyze time distribution

Understand where the most time and resources are spend.

Analyze operator statistics

Understand which ML operators and compute kernels consume most time.

See detailed device utilization

Make sure all CPUs and GPUs are utilized as expected.

Monitor speed and usage metrics

Monitor run metrics to catch issues such as memory leaks.

Get visibility into distributed workloads

See all distributed or multi-worker inference performance data in one place.

Enable team access

Easily share improvements with team members and others.

Ensure data privacy

Keep data private. No code or data is sent to Graphsignal cloud, only run statistics and metadata.

Get to full visibility in minutes

Start now