What kind of ML workload are you working on?

Select the primary model type for your use case.

Computer vision

Image classification, detection, segmentation, video, depth, LiDAR-derived vision.

Time series / sensor data

Forecasting, anomaly detection, vibration, IMU, telemetry, sequential signals.

Tabular / structured data

Fraud, risk, ranking, classification, regression.

NLP / small language models

Text classification, small LLMs, rerankers, embedded language systems.

Other

Another ML workload type.

Where does this model run, or where do you need it to run?

Select your deployment environment.

Edge device

Embedded, robotics, camera, sensor, industrial, field hardware.

Mobile / on-device

Phone, tablet, wearable, or other local device.

Cloud GPU infrastructure

AWS, GCP, Azure, or managed GPU inference/training.

CPU / on-prem infrastructure

Private data center, CPU-only inference, local infrastructure.

Hybrid

Mix of cloud, edge, on-prem, or staged deployment.

What is the primary bottleneck today?

Select the most important challenge you're facing.

Accuracy is not high enough

Model quality is the main blocker.

Optimization reduced model quality

Compression or speedup made the model worse.

Inference is too slow

Response time, FPS, or real-time requirements are hard to hit.

Model is too large for hardware

Model size or memory footprint is limiting deployment.

Compute cost is too high

Training or inference is too expensive.

Needs too much training data

Label volume or data efficiency is the bottleneck.

Iteration is too slow

Training, tuning, or deployment cycles take too long.

What optimization techniques have you already tried?

Select all that apply, or choose "None yet".

Quantization

Reduced precision to speed up or shrink the model.

Pruning

Removed weights, neurons, channels, or layers.

Distillation

Trained a smaller model from a larger teacher.

Smaller architecture

Switched to a smaller or lighter model.

Compiler/runtime optimization

ONNX, TensorRT, TVM, torch.compile, or similar.

None yet

You have not tried optimization yet.

Have optimization techniques reduced model quality?

Based on your optimization experience.

Yes, significantly

Optimization materially hurt the model.

Somewhat

Optimization helped efficiency but hurt quality.

No

Optimization did not hurt quality.

Not sure

You have not measured the tradeoff clearly.

What metric matters most for this workload?

Select the primary metric you track.

What is the current value for that metric?

Enter your current baseline (e.g., 0.86, 120ms, 450MB).

What stage or scale best describes this model?

Select your current deployment stage.

Research / prototype

Early exploration or internal experimentation.

Production pilot

Limited deployment or active evaluation.

Production model

Already deployed or near deployment.

Edge fleet / high-volume deployment

Many devices, high inference volume, or distributed deployment.

Enterprise platform

Shared ML platform, many teams, or multiple models.

Anything else we should know?

Optional: Share any additional context about your model or deployment.

How can we reach you?

We'll send your results and may follow up with relevant insights.

Name *

Email *

Company (optional)

Role (optional)

ML Optimization Diagnostic