Full reference

Multimodal projector tuning

Configure a projector run for a supported modality, encoder and stage plan. Inspect its runtime job linkage and artifacts before using the result.

Updated

Reviewed against Console v0.2.2. Available actions depend on project access and runtime configuration; check your installed CLI version for command differences.

Choose the training route

RoutePurpose
PEFT LoRAAdapt a compatible language-model backbone
Doc-to-LoRATrain or use document internalization
Projector tuningAlign supported modality representations with the model

These routes have different inputs, artifacts and resource needs. A completed job from one route cannot be substituted for another without compatibility checks.

Define the encoder and stages

Choose a supported modality and encoder, then an explicit stage plan and dataset. Confirm that the runtime has the relevant model and data. Keep these selections in the run record so a later evaluation can identify exactly what changed.

Inspect a run before repeating it

bash
stimulir lab projectors list --json
stimulir lab projectors get <run-id> --json

Creating a run requires --modality and --stages; inspect the parameter descriptions below for their accepted format. Deleting the platform record does not cancel an in-flight runtime job.

Review status and outputs

Use the Console’s Lab training views to inspect the run, stage status and associated job. Follow the reported failure when execution is blocked. Evaluate the resulting artifact on modality-appropriate held-out cases rather than equating job completion with quality.

Command catalogue

Expand a command for its arguments. These commands follow the released CLI source; check stimulir --version and the command’s --help for your installation. Required flags are marked; some operations also validate combinations of inputs.

stimulir lab projectors create

Create a multimodal projector tuning run — the third training route.

--modality, -m required
Input modality: image (vision tower) or audio (Whisper encoder).
--encoder
Frozen encoder tap (default per modality: SigLIP for image, Whisper for audio).
--stages required
stage1 trains the projector only; stage1+2 also jointly tunes a LoRA adapter.
--lora-rank
Stage-2 LoRA rank (stage1+2 only).
--lora-alpha
Stage-2 LoRA alpha (stage1+2 only).
--dataset
Workspace data asset id with paired multimodal examples.
--name, -n
Display name for the run.
--json
Emit raw JSON output.
stimulir lab projectors delete

Delete a projector tuning run record (in-flight HybrIE jobs run to completion).

run_id required
Positional argument. Projector tuning run id.
--yes, -y
Skip the confirmation prompt.
stimulir lab projectors get

Show a projector tuning run — status, encoder, stages, and HybrIE job linkage.

run_id required
Positional argument. Projector tuning run id.
--json
Emit raw JSON output.
stimulir lab projectors list

List multimodal projector tuning runs.

--modality
Filter by modality: image or audio.
--status
Filter by queued, running, completed, failed, or cancelled.
--limit
See the installed command’s --help for accepted values.
--json
Emit raw JSON output.

SDK method catalogue

Expand a client to inspect public methods and parameter names. Read each operation’s task guide before supplying a payload or starting work. The SDK returns platform responses; handle errors and inspect completion state before using a result.

client.lab_projectorsLabProjectorsClient
list(modality, status, limit)
Returns the platform response; inspect its status and error detail.
get(run_id)
Returns the platform response; inspect its status and error detail.
create(modality, stages, encoder, lora_rank, lora_alpha, data_asset_id, name)
Create a projector tuning run.
delete(run_id)
Returns the platform response; inspect its status and error detail.