refactor(web): split tokenization realignment from evaluateTransition 🚂#15191
Merged
jahorton merged 2 commits intoepic/autocorrectfrom Mar 16, 2026
Merged
refactor(web): split tokenization realignment from evaluateTransition 🚂#15191jahorton merged 2 commits intoepic/autocorrectfrom
jahorton merged 2 commits intoepic/autocorrectfrom
Conversation
User Test ResultsTest specification and instructions User tests are not required |
keymanapp-test-bot
bot
changed the title
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
from
b31bcad to
c303355
Compare
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
3 times, most recently
from
beafeb6 to
36df714
Compare
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
from
49391d5 to
3473c6f
Compare
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
from
36df714 to
c2e0427
Compare
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
from
3473c6f to
4f257f5
Compare
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
from
c2e0427 to
3713e6a
Compare
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
from
4f257f5 to
3ef2d2a
Compare
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
from
3713e6a to
0fa7e6b
Compare
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
from
3ef2d2a to
4450e14
Compare
jahorton
force-pushed
the
feat/web/search-space-node-propagation
branch
3 times, most recently
from
c5f9b66 to
91cf42e
Compare
jahorton
changed the base branch from
feat/web/search-space-node-propagation
to
feat/web/test-quotient-specialized-splits
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
2 times, most recently
from
a4843aa to
7e07e77
Compare
jahorton
commented
Mar 11, 2026
| readonly transitionEdits?: { | ||
| addedNewTokens: boolean, | ||
| removedOldTokens: boolean, | ||
| // NOTE: slated for removal in an upcoming PR. Exists in this form to |
With the various ways that tokenizations can transition depending upon which potential inputs are applied, it's possible for multiple different tokenizations to transition into the same one. As such, there will no longer be "just one" way that a tokenization is reached. Accordingly, it's best to perform word-boundary realignment operations (splits, merges) separately from text-editing operations (inserts, deletes). Build-bot: skip build:web Test-bot: skip
jahorton
force-pushed
the
refactor/web/realign-tokenization
branch
from
7e07e77 to
41beca3
Compare
jahorton
changed the base branch from
feat/web/test-quotient-specialized-splits
to
refactor/web/root-and-legacy-spur-tests
jahorton
requested review from
ermshiperete and
mcdurdin
and removed request for
ermshiperete
ermshiperete
approved these changes
Mar 13, 2026
web/src/engine/predictive-text/worker-thread/src/main/correction/context-tokenization.ts
Outdated
Show resolved
Hide resolved
|
|
||
| // Assumption: inputs.length > 0. (There is at least one input transform.) | ||
| const inputTransformKeys = [...inputs[0].sample.keys()]; | ||
| const baseTailIndex = (tailTokenization.length - 1); |
Contributor
There was a problem hiding this comment.
Shouldn't this be done after removing the tokens from tailTokenization? Otherwise baseTailIndex might point to an index that is no longer valid.
Contributor
Author
There was a problem hiding this comment.
No, this is still correct. Inputs to be applied are tokenized elsewhere, and those tokens are indexed relative to this specific index - the location of the last pre-edit context token. For such cases, the 'first' (and possibly more!) such token index (as obtained from inputs[0].sample.keys() will be negative.
This is enforced in ContextTokenization.mapWhitespacedTokenization and .assembleTransforms, which together produce the key-values obtained by the block of code reviewed here.
…on/context-tokenization.ts Co-authored-by: Eberhard Beilharz <ermshiperete@users.noreply.github.com>
ermshiperete
approved these changes
Mar 16, 2026
Base automatically changed from
refactor/web/root-and-legacy-spur-tests
to
epic/autocorrect
March 16, 2026 13:12
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
3 participants
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.
With the various ways that tokenizations can transition depending upon which potential inputs are applied, it's possible for multiple different tokenizations to transition into the same one. As such, there will no longer be "just one" way that a tokenization is reached. Accordingly, it's best to perform word-boundary realignment operations (splits, merges) separately from text-editing operations (inserts, deletes).
Fortunately, it's possible to enact this before multi-tokenization. It may even be advantageous to do so for clarity's sake - this makes clear which portions of the operations are for context word-boundary realignment and which are for actual context transition.
Build-bot: skip build:web
Test-bot: skip