Full reference

Tune and evaluate a PEFT adapter

Train a compatible low-rank adapter from reviewed examples, evaluate it on held-out cases, and register the artifact before deciding how it should be served.

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 adapter tuning for a measured gap

A PEFT LoRA adapter changes a supported base model through a smaller set of learned parameters. Keep the exact base model and adapter format with the artifact. Try a prompt or data correction first when that addresses the measured failure.

Preview supervised tuning

bash
stimulir lab train sft \
  --family <supported-family> \
  --compute-offer <offer-id> --region <offered-region> \
  --preview --json

Choose data and training settings using the options below. Managed preview validates and prices the selection without creating a job. It requires both compute offer and region, and cannot be combined with --endpoint. Remove --preview only when ready to submit the reviewed job.

Use RL only with a suitable reward

The RL route has its own environment, policy and reward settings. It is not interchangeable with supervised tuning. See RL training before selecting it.

Validate an externally trained artifact

Check base-model identity, LoRA shape, tokenizer compatibility and artifact accessibility. A local directory on your laptop will not be available to a remote worker unless you stage it there. Registration and supported formats depend on the configured runtime.

Score on held-out evidence

bash
stimulir lab eval adapter \
  --family <supported-family> \
  --adapter-dir <runtime-adapter-path> --json

Inspect the baseline and adapted results under the same evaluation settings. Keep training examples separate from held-out cases and inspect regressions, not just a single aggregate gain.

Inspect registered adapters

bash
stimulir lab adapters list
stimulir lab adapters get <adapter-id>

Keep the job ID and registered artifact identity together. A completed training job does not necessarily mean an adapter has been loaded for inference.

Move to serving deliberately

Use Training jobs to inspect execution and Hot-swap inference to manage runtime loading. Preserve the baseline and a rollback plan when changing a production route.

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 adapters get

Show an adapter's details.

adapter_id required
Positional argument. Adapter id.
--endpoint
HybrIE runtime URL (bypasses the console proxy).
--json
Emit raw JSON output.
stimulir lab adapters list

List registered adapters.

--endpoint
HybrIE runtime URL (bypasses the console proxy).
--json
Emit raw JSON output.
stimulir lab adapters load

Load an adapter into the runtime.

adapter_id required
Positional argument. Adapter id.
--endpoint
HybrIE runtime URL (bypasses the console proxy).
--json
Emit raw JSON output.
stimulir lab adapters unload

Unload an adapter from the runtime.

adapter_id required
Positional argument. Adapter id.
--endpoint
HybrIE runtime URL (bypasses the console proxy).
--json
Emit raw JSON output.
stimulir lab eval adapter

Score a PEFT LoRA adapter directory on held-out retrieval (vs base model).

--family, -f required
Base model family: qwen3-4b, qwen3-0.6b, or mistral-7b.
--adapter-dir, -a required
Adapter directory with PEFT LoRA weights to score.
--examples
Held-out examples to score.
--seed
Seed for reproducible example generation.
--device
auto, metal, cuda, or cpu.
--model-dir
Optional base-model directory override.
--endpoint
HybrIE runtime URL (bypasses the console proxy).
--json
Emit raw JSON output.
stimulir lab train sft

Submit an SFT job training a plain PEFT LoRA adapter (job kind sft-peft-lora).

--family, -f required
See the installed command’s --help for accepted values.
--lora-rank
LoRA rank of the trained adapter (default: family preset).
--init-adapter-dir
PEFT adapter directory to warm-start from.
--examples
Number of training examples (sent as num_examples).
--epochs
Training epochs.
--lr
Learning rate.
--l1-coef
L1 regularisation coefficient.
--eval-examples
Held-out eval examples scored after training.
--min-blocks
Minimum synthetic context blocks.
--max-blocks
Maximum synthetic context blocks.
--log-every
Progress interval; 0 disables intermediate reports.
--seed
Seed for reproducible example generation.
--device
Target device (auto, metal, cuda, cpu).
--checkpoint-dir
Directory to write the trained adapter into.
--dataset
Immutable Lab data asset id.
--split-ratio
Fraction used for training.
--split-seed
Deterministic case split seed.
--eval-cohort
Held-out comparable eval run id.
--compute-offer
Managed GPU offer id.
--region
Managed GPU region.
--gpus
Managed GPU instance count.
--spot
Use spot pricing when the offer supports it.
--compute-provider
Managed compute provider.
--max-duration-hours
Hard managed-run time ceiling.
--max-cost-gbp
Optional GPU-spend ceiling.
--preview
Validate and price only; create no job and launch nothing.
--endpoint
HybrIE runtime URL (bypasses the console proxy).
--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_trainLabTrainClient
d2l(family, examples, eval_examples, epochs, lr, seed, checkpoint_dir, device, compute, timeout)
Doc-to-LoRA (hypernetwork context internalisation) training job.
sft(family, lora_rank, init_adapter_dir, examples, epochs, lr, l1_coef, eval_examples, min_blocks, max_blocks, log_every, seed, device, checkpoint_dir, dataset, split_ratio, split_seed, eval_cohort, compute, timeout)
Supervised fine-tuning of a plain PEFT LoRA adapter (sft-peft-lora).
rl(family, environment, prompts, group_size, policy, lora_rank, init_checkpoint_dir, lr, kl_beta, seed, device, checkpoint_dir, compute, timeout)
RL (GRPO-style) training job on an RL environment.
internalize(document_text, compute, name, session_key, ttl_seconds, pin, timeout)
Internalize a document on an explicitly selected managed worker.
client.lab_adaptersLabAdaptersClient
list()
Registered adapters (unwrapped from the runtime's "adapters"/"data" envelope).
get(adapter_id)
Returns the platform response; inspect its status and error detail.
load(adapter_id)
Load an adapter into the runtime (hot-swap for inference).
unload(adapter_id)
Returns the platform response; inspect its status and error detail.