rlmesh::RunOptions
Options for run and Model::run.
struct RunOptions<'h>Build with ..Default::default(). Adding an options field is a minor
change by project policy.
Fields
- episodesThe exact number of episodes: one by default,
seeds.len()when only seeds are given; the two must agree when both are. - seedsOne reset seed per episode.
- max_episode_stepsTruncate an episode after this many steps (at least 1).
- max_episodeTruncate an episode after this wall time (must be positive).
- hooksObserver callbacks, borrowed for the call.
- instructionOverride each text input of a spec’d model for this call.
- viewLive viewer; a local
Model::runrefuses a set view. - close_envShut the caller's env down when the run ends.
- trust_entrypointsOverride the model's
trust_entrypointsfor this run. - execution_horizonActions of each predicted chunk executed before re-planning (1..=1024; default 1).
- prefetch_leadNative-loop async inference: predict the next chunk while this many replay frames remain (0 = off; at most
u32::MAX). - workflow_editionThe workflow edition this run declares, above
RLMESH_WORKFLOW_EDITIONand the model's own. - trial_index_baseThe first trial ordinal; episode
iwalkstrial_index_base + i. - cancelCancels the run.
Methods
episodes
field[source]episodes: Option<u64>The exact number of episodes: one by default, seeds.len() when only
seeds are given; the two must agree when both are. Some(0) runs
nothing.
seeds
field[source]seeds: Option<Vec<i64>>One reset seed per episode.
max_episode_steps
field[source]max_episode_steps: Option<u64>Truncate an episode after this many steps (at least 1).
max_episode
field[source]max_episode: Option<std::time::Duration>Truncate an episode after this wall time (must be positive).
hooks
field[source]hooks: Option<&'h mut dyn RunHooks>Observer callbacks, borrowed for the call.
instruction
field[source]instruction: Option<String>Override each text input of a spec’d model for this call.
view
field[source]view: Option<View>Live viewer; a local Model::run refuses a set view.
close_env
field[source]close_env: boolShut the caller’s env down when the run ends.
trust_entrypoints
field[source]trust_entrypoints: Option<bool>Override the model’s trust_entrypoints for this run.
execution_horizon
field[source]execution_horizon: u32Actions of each predicted chunk executed before re-planning (1..=1024; default 1).
prefetch_lead
field[source]prefetch_lead: u64Native-loop async inference: predict the next chunk while this many
replay frames remain (0 = off; at most u32::MAX).
workflow_edition
field[source]workflow_edition: Option<String>The workflow edition this run declares, above
RLMESH_WORKFLOW_EDITION and the model’s own.
trial_index_base
field[source]trial_index_base: u64The first trial ordinal; episode i walks trial_index_base + i.
cancel
field[source]cancel: Option<CancelToken>Cancels the run.
instruction
method[source]fn instruction(self, value: impl Into<String>) -> SelfSet the text override.
Trait implementations
- impl Default for RunOptions<'_>
- impl Debug for RunOptions<'_>