External LeLab compute provider for Seeed Cloud Platform.
This package is intentionally outside the LeLab source tree. It lets Seeed maintain the cloud compute integration without adding vendor-specific code to LeLab core.
pip install -e ./lelab-compute-seeed-cloud
export SEEED_CLOUD_API_URL="https://sensecraft-gpu.seeed.cc/api"
export SEEED_CLOUD_TOKEN="<seeed-cloud-token>"
export SEEED_CLOUD_FLAVORS="RTX 4090"
lelab-seeed doctor
lelab-seeedlelab-seeed is a thin launcher. It adds this package to
LELAB_COMPUTE_PROVIDER_MODULES, then delegates to the normal lelab command.
Run lelab-seeed doctor --api when you also want to verify /api/health on
the configured Seeed Cloud API.
When running LeLab from source without the launcher, this fallback also works:
export LELAB_COMPUTE_PROVIDER_MODULES="lelab_compute_seeed_cloud.provider:create_provider"
lelabFor the MVP, the plugin sends LeLab's normal dataset repository id as
datasetRepoId. If you need to run from a prepacked archive instead, pass a
downloadable .zip, .tar.gz, or .tgz URL as the dataset value, or set
SEEED_CLOUD_DATASET_URL. The model/training output flow is not changed by this
plugin.
The plugin sends the full LeLab TrainingRequest as trainingConfig to the
Seeed Cloud API. The platform keeps the existing top-level policyType and
trainSteps fields for compatibility, while the agent can use
trainingConfig to preserve batch size, W&B, optimizer, checkpoint, and other
LeLab training options.