Skip to main content

mellea.plugins.builtin_debug.sampling

Built-in debug plugin for sampling pipeline.

Provides tracing for sampling strategies including iteration tracking, validation results, repair events, and success/failure analysis.

Examples: Enable sampling tracing:

from mellea.plugins.builtin_debug.sampling import ( log_sampling_loop_start, log_sampling_iteration, log_sampling_repair, log_sampling_loop_end, ) from mellea.plugins import register

register([ log_sampling_loop_start, log_sampling_iteration, log_sampling_repair, log_sampling_loop_end, ])

with start_session() as m: result = m.instruct("...", strategy=SamplingStrategy(...))

Functions​

FUNC log_sampling_loop_start ​

log_sampling_loop_start(payload: SamplingLoopStartPayload, ctx: PluginContext) -> None

Log sampling strategy initialization with budget and requirement count.

Args:

  • payload: SamplingLoopStartPayload with strategy_name, loop_budget, requirements.
  • ctx: Plugin context for hook execution.

FUNC log_sampling_iteration ​

log_sampling_iteration(payload: SamplingIterationPayload, ctx: PluginContext) -> None

Log validation results for each sampling attempt.

Args:

  • payload: SamplingIterationPayload with iteration, valid_count, validation_results.
  • ctx: Plugin context for hook execution.

FUNC log_sampling_repair ​

log_sampling_repair(payload: SamplingRepairPayload, ctx: PluginContext) -> None

Log when repair is triggered during sampling iterations.

Args:

  • payload: SamplingRepairPayload with repair_iteration, repair_type, failed_validations.
  • ctx: Plugin context for hook execution.

FUNC log_sampling_loop_end ​

log_sampling_loop_end(payload: SamplingLoopEndPayload, ctx: PluginContext) -> None

Log sampling completion with success status and attempt statistics.

Args:

  • payload: SamplingLoopEndPayload with success, iterations_used, all_results, all_validations.
  • ctx: Plugin context for hook execution.