tp sweep - TunePlane

tp sweep <EXP> -g KEY=V1,V2 [options]

Each variant goes through the ordinary submit path: local validation, quota, queue. The walkthrough is running a sweep.

Worked examples

# 3 x 2 = 6 variants. Look at the list before spending anything.
tp sweep my-grpo --profile h200:4 \
  -g policy.optimizer.kwargs.lr=1e-6,2e-6,5e-6 \
  -g loss_fn.reference_policy_kl_penalty=0.01,0.05 \
  --dry-run

tp sweep my-grpo --profile h200:4 \
  -g policy.optimizer.kwargs.lr=1e-6,2e-6,5e-6 \
  -g loss_fn.reference_policy_kl_penalty=0.01,0.05 \
  -s grpo.num_generations_per_prompt=8

# cut the hopeless variants at step rungs
tp sweep my-grpo --profile h200:4 \
  -g policy.optimizer.kwargs.lr=1e-6,1e-5,1e-4 \
  --early-stop --rungs 100,300,900 --keep 3

# stop the whole group
tp job stop-sweep sweep-my-grpo-20260912-101500

There is no flag for the group name. The sweep_id is sweep-<experiment>-<timestamp>, and tp sweep prints it.

`tp sweep`

Submit a hyperparameter sweep

tp sweep <EXP> [options]

Argument	Description
`EXP`	Experiment name or path (required)

Option	Description
`--set-grid`, `-g`	A grid axis, repeatable; several axes form a cartesian product. e.g. -g policy.lr=1e-5,2e-5 -g grpo.kl=0.01,0.05 (required · repeatable)
`--set`, `-s`	A fixed override shared by every variant, repeatable (repeatable)
`--profile`	Target hardware and resources in one argument: h200 (the registry default shape), h200:4 (4 GPUs), h200:16 (2 full nodes). Repeat it to split pools per role (the heterogeneous extension point): —profile train=h200:8 —profile rollout=h100:2. Comma-separate roles that share a pool: —profile actor,rollout=h200:8 —profile teacher=h200:8 (repeatable)
`--method`, `-m`	Method identifier; read from the experiment’s recipe.lock.json when omitted
`--allow-dirty`	Allow uncommitted changes in the working tree
`--no-validate`	Skip the pre-submission validation
`--dry-run`	Only print the variant list, submit nothing
`--early-stop`	Enable ASHA early stopping: cut the clearly hopeless variants at a few step rungs
`--stop-metric`	Which metric early stopping ranks by; defaults to the method’s declared objective, held-out first (see tp methods <method>)
`--rungs`	Early-stopping rungs (steps), comma separated. Defaults to 100,300,900
`--keep` `<int range>`	Keep 1/keep of the variants at each rung. 3 is the usual value: cutting harder hurts slow starters (default `3`)