Repository navigation
feat(compression): runnable A/B benchmark CLI (F2.4) - #4220
Merged
Merged
Conversation
The compression benchmark harness (benchmarkEngines / compareReports / BENCHMARK_CORPUS) existed and was unit-tested, but nothing invoked it, it wasn't re-exported from the harness barrel, and there was no report output — so F2.4 "benchmark L2" was test-only scaffold. - Add a markdown report formatter (formatBenchmarkTable) + a default deterministic engine set (DEFAULT_BENCHMARK_ENGINES, llmlingua excluded — needs the ONNX model at runtime). - Re-export the benchmark API from harness/index.ts. - Add a CLI (scripts/compression/benchmark.ts) + `npm run bench:compression [engineId ...]` that runs the corpus and prints a best-first A/B table. Deterministic and API-free (the harness already guarantees reproducibility). Real run: ccr 36.5% savings, ultra 19.7%, others ~0% on this single-string corpus — honest numbers. The N4 budget-gate ratchet (budgetGate.ts wired to CI with a committed baseline) and the real llmlingua A/B (needs the ONNX model) are tracked follow-ups. Part of the compression "100% functional" program (audit follow-up).
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Merged
tkgo11
pushed a commit
to tkgo11/OmniRoute
that referenced
this pull request
Sep 23, 2026
…ce#4220) Integrated into release/v3.8.29 (F2.4 runnable compression A/B benchmark CLI).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Contexto
Programa compressão 100% funcional — item F2.4. O harness de benchmark (
benchmarkEngines/compareReports/BENCHMARK_CORPUS) existia e era testado, mas nada o invocava, não era re-exportado do barrel, e não havia saída de relatório → F2.4 era scaffold test-only.O que muda
formatBenchmarkTable(best-first, melhor engine em negrito) + set defaultDEFAULT_BENCHMARK_ENGINES(8 engines determinísticos;llmlinguaexcluído — precisa do modelo ONNX em runtime).harness/index.ts.scripts/compression/benchmark.ts+ scriptnpm run bench:compression [engineId ...]que roda o corpus e imprime a tabela A/B.Saída real (validação real-environment)
Números honestos: o adapter envolve uma string única como 1 mensagem
user, então rtk/caveman (que comprimem tool-results/multi-msg) marcam ~0% nesse corpus — o benchmark surfacing isso é o ponto.Follow-ups rastreados
budgetGate.tscom baseline commitado + CI) —runBenchmarkGatejá existe.llmlingua(precisa do modelo ONNX) — framework é engine-agnostic, basta passar o id.Validação
formatBenchmarkTable(markdown, best bolded) +DEFAULT_BENCHMARK_ENGINESresolvem e rodam o corpus end-to-end.npm run bench:compression) → tabela acima, exit 0.benchmark.test.ts"reproducibility") ·typecheck:core✅ ·eslint✅.🤖 Generated with Claude Code