CLI

Use the embedl-hub CLI to record results from shell scripts or track an existing command as a run. It works with the same projects as the Python client. Complete SDK setup first.

ehub is a shorter name for the same command. Use --help on any command to see its options.

Start and finish a run#

run=$(embedl-hub run start --type eval --name "Evaluate classifier")
embedl-hub log param --run "$run" model_id classifier_int8_v1
embedl-hub log tag --run "$run" sample_data true
embedl-hub log metric --run "$run" accuracy 0.92 --step 1
embedl-hub run finish "$run"
embedl-hub run url "$run"

This logs illustrative data. Replace it with measurements from your own work. A run started with run start stays open until you finish it.

Use --use on run start to select it for subsequent commands in this directory and those below it, through the nearest .ehub/use.cfg. See CLI targets for how that file is chosen. embedl-hub use shows the selected defaults; use clear removes them. Explicit --run options are useful in scripts that track several runs.

Wrap an existing command#

embedl-hub exec --type eval python evaluate.py

Replace evaluate.py with your script. exec starts a run, provides the run and its project to the command through EMBEDL_HUB_RUN and EMBEDL_HUB_PROJECT, and finishes it based on the command’s outcome. Commands inside the script can use embedl-hub log to attach results to that run.

A Python script can adopt the run with client.use_inherited_run() instead of starting a second run. The wrapper remains responsible for finishing it.

exec --safe is the CLI’s only safe-mode switch. See Safe integration for its behavior and how to handle logging inside the command.

Queue log entries#

Use a batch to collect log entries locally and send them together. Create a run and select it before starting a batch:

embedl-hub run start --type eval --use
embedl-hub batch start --name "Evaluation metrics" --use
embedl-hub log metric accuracy 0.92 --step 1
embedl-hub log param dataset validation_v1
embedl-hub batch show
embedl-hub batch send
embedl-hub run finish

Batch entries stay on the local machine until sent successfully. Starting and finishing a run still requires a connection. Use batch list to find local queues and batch discard to remove one you no longer need.

Inspect recorded work#

embedl-hub run list
embedl-hub run show RUN_ID
embedl-hub run url RUN_ID

run list shows running, pending, and finished runs by default, and reports when failed or killed runs are hidden. Use --failed for only failed and killed runs, or --all for every status. run list and run show support --json.

Command groups#

Next steps#

• Explore results to inspect the runs you recorded.