Experiment Tracking
Explore experiments, compare results, and share what you find.
Use Experiment Tracking to understand how a model performs and decide what to try next. Explore individual runs, compare candidates in dashboards, and share results with your team in the Embedl Hub web app.
Compare accuracy and loss across four training configurations.
You can start with results already in a project. To try it with sample data, create a project using the Hub demo template.
Explore your experiments#
Open Runs to find a training session, evaluation, or hardware profile. Each run keeps its measurements alongside the settings and files used to produce them. Open a result to inspect accuracy, check the dataset, or download a report. See Exploring results.
A run can include:
- Parameters: settings such as precision, batch size, or learning rate.
- Metrics: measurements such as accuracy, loss, or latency, including progress over steps.
- Tags: labels for finding related work.
- Artifacts: files such as model checkpoints, plots, or reports.
Compare candidates#
Open Experiments to select runs and compare them in a dashboard. Follow training curves, compare an evaluation with a baseline, or inspect how quantization changes accuracy and latency.
Choose charts and tables for the comparison, then save the dashboard and share it with teammates. See Dashboards.
Understand a result#
Follow run lineage to find the steps that led to an evaluation. Inspect artifacts for the model file or detailed report. When comparing hardware results, check the device, runtime, and test conditions alongside the measurements.
If a result needs review against a requirement, use Evidence Collection to submit it with supporting context.
Record your own work#
Connect your training and evaluation code with the SDK to record new runs. Your code runs locally or in your existing pipelines; the Hub stores the results for you and your team to explore.