Dsxir.Optimizer.MIPROv2.Trial (dsxir v0.3.0)

Runs one MIPROv2 trial: apply a sampled config to a program's per-predictor state, run the program over a minibatch in parallel, score each prediction with the metric, and return a Dsxir.Optimizer.MIPROv2.Stats.Record.

Parallelism uses Task.Supervisor.async_stream_nolink/4 under Dsxir.TaskSupervisor. Each worker replays the caller's settings snapshot via Dsxir.Settings.run/2 so the LM, adapter, metadata, and call_plugs are preserved per-tenant.

Cache hits are tracked via Dsxir.Optimizer.Cache.fetch_or_put/3's {:hit | :miss, value} return. A miss increments lm_calls; a hit increments cached_calls. When the cache is disabled (tid == nil) every call is reported as a miss so lm_calls reflects the true number of program executions and cached_calls stays at zero.

Examples whose program forward function raises score 0.0 and the batch continues. Worker exits (timeout, crash) likewise score 0.0 without aborting the trial.

Summary

Types

args()

Functions

run(map)

Apply config to program, evaluate examples in parallel, return a Record summarising mean score, LM calls, cache hits, and duration.

Types

args()

@type args() :: %{
  program: Dsxir.Program.t(),
  config: map(),
  candidates: Dsxir.Optimizer.MIPROv2.Candidates.t(),
  examples: [Dsxir.Example.t()],
  metric: Dsxir.Metric.t(),
  cache_tid: Dsxir.Optimizer.Cache.tid(),
  settings_snapshot: %{globals: map(), stack: [map()]},
  trial_index: non_neg_integer(),
  batch_size: pos_integer()
}

Functions

run(map)

@spec run(args()) :: Dsxir.Optimizer.MIPROv2.Stats.Record.t()

Apply config to program, evaluate examples in parallel, return a Record summarising mean score, LM calls, cache hits, and duration.