Skip to content

docs(site): explain CPU transcription memory, timing and workload limits - #3933

Merged
LauraGPT merged 1 commit into
mainfrom
codex/site-cpu-suitability-guide-20261122
Nov 21, 2026
Merged

LauraGPT merged 1 commit into
mainfrom
codex/site-cpu-suitability-guide-20261122

Conversation

@LauraGPT

Copy link
Copy Markdown
Collaborator

Summary

Substantially revise the existing bilingual CPU article, lightweight-speech-recognition-cpu, for beginners and application agents. Explain how to evaluate a recording-to-final-text workload using quality, memory, timing and concurrency instead of unqualified benchmark promises.

  • Separate toolkit, model weights, runtime and service; qualify standalone C++/Python and offline/disconnected boundaries.
  • Explain peak memory, RTF, cold/warm requests, client latency and concurrency with explicitly invented arithmetic examples.
  • Add an application-owned evaluation record and authorization/failure boundaries; do not present it as an SDK response schema.
  • Preserve both URLs and the original 2026-06-22 publication date; mark the 2026-11-22 revision. Update one catalog entry and the two legacy hashes.

Validation

  • Focused bilingual regression: 12 failures on old content, then 12 passes.
  • Full product-site suite: 789 tests; two deterministic builds with 281 matching files.
  • Browser suite: 154 checks; four additional Chinese/English mobile/desktop article views with screenshots inspected.
  • Five private static transition checks and ten isolated transfer contract cases; neither substitutes for production acceptance.
  • Exact-head CI and its checkout/log evidence are required before normal merge. Main CI and actual public HTML/resource bytes are separate pre-publication and post-deployment gates.

Scope

This is a conceptual guide, not a new installation tutorial or performance report. Official source/documents at 581150d8e1de2bc055cc390b6f2efc9ea78b48b7 support the boundaries. No new model inference, weights, API purchases or production model-service changes. No accuracy, speed, adoption or growth claim. Nano remains deferred.

Only five website content/metadata/test paths change. Static deployment is a separate authorized operation after normal merge; this PR alone does not prove the page is live.

Rewrite the bilingual CPU guide around model/runtime boundaries, memory, RTF, request latency and workload evaluation. Preserve URLs and publication dates; label invented arithmetic and remove unbounded benchmark claims. No new model inference or downloads.

Signed-off-by: LauraGPT <zhifu.gzf@alibaba-inc.com>
@LauraGPT
LauraGPT merged commit 47ba463 into main Nov 21, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant