Keep L6, warm on load, show a download percentage - #6
Open
aashutoshrathi wants to merge 2 commits into
Open
Conversation
Items that ended with a secondary '[Source]' or '[Read]' link (often tweets) left a dangling '... [Source.' tail in the description. tidy_desc now removes a trailing bracketed source/read/tweet/link label. Regenerated index.html + vectors.f32.
Reverts the model swap/deferral direction in favour of less *net* wait:
- Stay on all-MiniLM-L6-v2 (better retrieval quality).
- Warm the model on load in the background (Web Worker), so the ~23 MB download
overlaps with reading time instead of blocking a first search.
- Progress readout is now a download percentage ('Setting up smart search… 47%')
from transformers.js's progress_callback, instead of an item count (which only
ever appeared on the rare no-precomputed-vectors fallback path anyway).
Built on the empty-description fix (PR #4); supersedes the L3/defer branch (PR #5).
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Goal: less net wait, without giving up quality
Direction correction after the L3/defer experiment (#5). Keeping all-MiniLM-L6-v2 for its better retrieval quality, and cutting the felt wait a different way:
Setting up smart search… 47%(from transformers.js'sprogress_callback) instead of an item count. The item count only ever appeared on the rare fallback path where precomputed vectors are missing; on the normal path there's no per-item embedding, so a % of the model download is the meaningful signal.Relationship to open PRs
Built on top of #4 (empty-description fix), so it includes those commits. It supersedes #5 (which swapped to the smaller L3 model and deferred the download — reverted here). Suggested: merge this, close #4 and #5.
Verified
On load, the model warms in the background with no interaction and the status settles to "on · ask in plain English" when ready; semantic search works on L6. The % updates during a real (non-localhost) download.