Neuron steering

Steering changes selected internal activations during a forward pass. The checkpoint's stored weights remain unchanged. A saved steering Model stores an exact base and intervention profile; it is different from a learned weight patch produced by Neuron training.

For example, compare the same prompt under Off, a zero dose and negative/ positive doses for a selected coordinate. Observe both the output and runtime evidence; a desired change on one prompt may introduce regressions elsewhere.

Set up the experiment

Choose New → Neuron steering in Experiments.

Coordinates are model-specific. The normalization recipe can scale an intervention by the baseline prompt's median residual-vector norm after the first prompt position, divided by the selected down-projection column norm. The runtime owns calibration and records its result; dose 1 is not a checkpoint-independent physical unit. Use the exact normalization and token scope shown in the prepared request.

Off removes the intervention. A numeric zero dose is an enabled-path comparison and should not be silently substituted for Off. Keep calibration, model, prompts, sampling and token limits fixed when comparing signed doses. Current prompt/capture controls are shared with Activation inspection.

Interpret the experiment

An activation change does not guarantee a reliable personality, factual edit or task skill. Check nearby prompts and unrelated tasks. Inspect actual output, selected coordinates, baseline calibration, intervention scope and incomplete capture/work limits rather than judging only a single attractive response.

Review, run and recover

Use Review → Prepare review to resolve the exact inputs, compatible runtime and maximum spend. Inspect the retained preparation before approving Start. Opening the editor, changing tabs and saving a draft do not start paid compute. Preparing a run is separate from starting it.

If a response is lost, reopen the same experiment and retained execution before retrying. Preserve its operation identity and revision; creating a second run can duplicate work and charges. Use the existing refresh, retry or cancellation controls. Cancellation is a request: follow the execution through cleanup and inspect remaining capacity and funding before treating it as finished.

Runtime discovery is authoritative. A selectable action or saved configuration does not prove that a checkpoint/image/GPU combination is qualified. Unsupported combinations cannot start simply because sufficient credits are available.

Save and serve

After a successful execution with verified artifacts, choose Save Model. The saved version retains its exact base, modification and artifact identity. Saving is separate from allocating serving capacity. Open the saved Model, review compatible serving and approve activation explicitly. Use Stop and verify released capacity when finished.

Measure a baseline and candidate on the same held-out tasks and grader versions. Training loss, successful export and a working endpoint do not establish better answers. Keep runtime correctness and task quality as separate checks.

The saved steering profile is applied by its compatible serving runtime. It requires an implementation that applies the profile correctly for each request; an ordinary inference engine does not acquire steering support merely by loading the same base weights. Request state and calibration cannot leak between users or unrelated prompts.

API, SDK and harness

Use compileModificationExecution, prepareModificationExecution and startModificationExecution; CLI commands use the hosted-*-modification family. Save with training hosted-save-steering-model and control compatible serving with hosted-control-steering-serving.

Use the official CLI with your configured account and workspace. Run openpond training --help and the selected command's help for the current request schema; start from the SDK's typed draft/request builders rather than inventing an unversioned JSON payload. Web, API, TypeScript/Python SDKs and the CLI retain the same preparation, execution and Model identities. Paid Start and serving activation require explicit approval; a harness should retain these identities for recovery.