Repository navigation
Simplify CLI: one bundle-resolution ladder + CQ level in list - #715
Merged
Merged
Conversation
run, serve, transcribe and download now share a single ensure_runnable_bundle resolver (local path -> cache -> prebuilt download -> local convert+transpile), so any entry point builds a model on demand instead of erroring or using a partial path. Demote convert and transpile to an advanced section in --help. Signed-off-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Read the precision code from each tensor header and report the dominant CQ level (e.g. CQ4) as a column, since bit-width is stored per-tensor rather than in any single config field. Signed-off-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Signed-off-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
main grew a parallel bundle resolver for cactus test (direct push 8e7b71e) — a bespoke _ensure_bundle that hardcoded bits/platform and duplicated the download/convert/transpile ladder. Route cactus test through the shared ensure_runnable_bundle instead and delete the dead helpers. Carry over main's 'delete build artifacts, weights, venv' clean help wording.
main advanced again (DEFAULT_TEST_MODEL_ID/_TRANSCRIPTION split out for cactus test; --platform default 'auto'). Fold both in: cactus test now falls back to the DEFAULT_TEST_* ids through the shared ensure_runnable_bundle, help/wiring use 'auto'. Re-dropped main's re-introduced cactus test band-aid helpers.
Collaborator
|
tested, LGTM @jakmro |
kar-m
previously approved these changes
Jun 15, 2026
kar-m
left a comment
kar-m
left a comment
Collaborator
There was a problem hiding this comment.
LGTM (needed to write this in a review as well)
kar-m
approved these changes
Jun 15, 2026
kar-m
left a comment
kar-m
left a comment
Collaborator
There was a problem hiding this comment.
the PR removed a function that was used in a test, after this was fixed the rest of the logic is good
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.
Summary
Two independent CLI cleanups (one commit each).
1. Unify model resolution behind one fallback ladder
run,serve,transcribeanddownloadpreviously resolved bundles three different ways —runhad the full ladder,transcribeskipped the cache/prebuilt rungs, andservejust errored telling you to runconvertthentranspile. They now share a singleensure_runnable_bundleresolver:So every entry point builds a model on demand.
convert/transpileare unchanged but demoted to an "Advanced / build pipeline" section in--help, since they now run automatically. Net diff is ~neutral — the shared resolver replaces three bespoke ones.2. Show CQ quantization level in
cactus listBit-width isn't stored in any single config field — it lives per-tensor in the
CACTheader.listnow reads the precision code from.weightsheaders and reports the dominant CQ level as a column:Early-stops after a small sample, so it stays fast (~0.2s).
Test plan
cactus --helprenders; advanced section presentcactus listshows CQ columnconvert/transpile