mare_mark
Reproducible benchmarking, statistical comparison, tuning, and self-contained reports for MoonBit payloads.
benchmark
experiment
statistics
tuning
report
moon add Luna-Flow/mare_mark@0.3.0
Download zipAuthor
Version
0.3.0
License
Apache-2.0
Repository
Last updated
last month
Downloads
32
Dependencies
- moonbitlang/x0.4.46
- moonbitlang/async0.20.1
README#mare_mark
mare_mark is a reproducible experiment toolkit for MoonBit payloads. It keeps
dataset generation, fixture setup, timed execution, raw observations,
statistical comparison, tuning policy, and report rendering in separate
packages so measurement code stays small and reviewable.
#Current Baseline
Version 0.3.0 provides a stateful differential runner with shared warmup,
seeded balanced execution order, asynchronous synchronization hooks, typed
execution outcomes, reference and relational oracles, native subprocess
isolation, failure shrinking/replay artifacts, streaming JSONL adapters,
median-based statistics with deterministic paired bootstrap intervals,
environment compatibility checks, tuning data models, artifact-versioned JSONL,
and self-contained differential HTML reports.
The CLI is an artifact-edge tool and deliberately stays outside benchmark
execution. It provides report generation and native replay:
moon run src/cli --target native -- report \
testdata/report/sample.jsonl report.htmlreplay executes the worker command captured in a validation_failure event
on the native target. Applications construct and run typed benchmark cases
through the package API, so CLI parsing, filesystem access, and rendering can
never enter a timed payload.
#Installation
moon add Luna-Flow/mare_mark@0.3.0#Package Map
| Package | Responsibility |
|---|---|
| model | Protocol, environment, observation, validation, crossover, and deployment data. |
| generator / fixture | Reproducible input context and isolated setup/reset lifecycle. |
| runner | Oracle validation, sticky contexts, isolated workers, setup/reset lifecycles, timing, and shrinking. |
| event / ir_sink | In-memory and callback-streaming JSONL event sinks. |
| stats / experiment | Robust summaries, paired comparison, outlier reporting, and crossover diagnosis. |
| ir_model / report | Plot and differential IR, mismatch/capability/corpus tables, SVG, and HTML reports. |
| tune / tune_gemm | Candidate scoring, Pareto filtering, and GEMM tuning models. |
| cli | Artifact replay and JSONL-to-HTML report adapter; never imported by the runner. |
#Documentation
#Development
moon check --target js
moon check --target native
moon test
moon info
moon fmtGenerated pkg.generated.mbti files are the public interface snapshots.
Reproducible benchmarking, statistical comparison, tuning, and self-contained reports for MoonBit payloads.
benchmarkexperimentstatisticstuningreport
0.3.0
Apache-2.0
Updated last month
32 downloadsDependencies
- moonbitlang/x0.4.46
- moonbitlang/async0.20.1