Repository navigation
Qualify bounded local runtimes, Python retrieval and R2 for server 0.3.6 - #153
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
R2 live qualification exposed an empty-list regression in AWS SDK v4 and an ignored conditional-delete precondition. Empty S3 pages now work, and R2 refuses conditional deletion before sending it. Eight isolated live checks and nine local transport checks pass; the temporary bucket was removed. R2 remains preview with this explicit contract limitation.
Adds opt-in local native chat and a portable Choice log-probability judge over an injected session contract. Chat enforces an explicit output-token ceiling; the native adapter binds model bytes, runtime build, original/effective template, context and single-slot configuration. Scoring validates single-token labels, complete requested probabilities, rotations, coverage and agreement. Local runtime registration is independent of remote CLI registration; authenticated workspace-write guards remain enforced. These targets remain prototypes, with production host/accounting/issuer qualifications explicitly held.
Python flat-vector retrieval keeps exact float32 operation ordering and full eligible-record validation, while reducing scalar conversion overhead and bounding retained candidates. The frozen baseline probe preserves 2,075 numeric/error controls and nine complete timed 384-dimensional responses, plus nine complete budget responses at 10/30/100 candidates and a ten-item output cap. Adds the same parity probe to the nine-platform qualification matrix. Returned rows and candidate pools are recorded separately. Distinct-item grouping remains caller-owned; rank-before-hydration is deferred to preserve fail-closed storage validation.
Includes a bounded Cloudflare Queues bridge design and authorizes server image 0.3.6. Existing registry package cohorts are unchanged.
Validation: full Python runtime qualification (246 tests, coverage, strict typing, clean wheel/sdist/server installs); 33 focused provider cases including two pinned real-model smoke cases; 189 general provider tests; three server-registration controls; JavaScript and Python client suites; SDK/schema/version/publication checks; deterministic export; credential/history/private ownership scans. Canonical release evidence and the protected publisher must pass before publication.