Repository navigation
Prepare v0.2.0: version bump, release notes and rechecked README commands - #63
Merged
Merged
Conversation
…ands v0.2.0 serves the same CRF bundle as v0.1.0 and ships the declared test evaluation, two-reference scoring, the ONNX/INT8 check, the prospective check, fresh-clone reproduction and the public demo. Every README command was rerun: the pilot-gate command now names pilot-2 (which reproduces docs/pilot_gate.md byte for byte), the import comment notes that re-importing rewrites pilot-1 with identical labels, and app run mentions --port.
…xture Dynamic INT8 kernels differ between CPUs, so INT8 predictions of the tiny fixture model vary by runner; a GitHub runner moved spans to typed F1 0.93 against 1.0, failing the 0.05 allowance (the same failure delayed v0.1.0). INT8 accuracy is decided by the typed-F1 gate on a trained model (reports/onnx-int8/, where it fails for DeBERTa run 3). The test still requires exact FP32 parity, and now that every INT8 output loads and passes the validator, with a smaller file and a distinct weights digest.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
pyproject.toml,__init__.py,uv.lock— one line each).docs/release_notes/v0.2.0.md: preview release (gates not judgeable), same default CRF bundle as v0.1.0, changes since v0.1.0, measured dev/test results, limitations, deployment and rollback.annotate gateneeds--gold annotations/pilot/pilot-2.jsonlto reproducedocs/pilot_gate.md(verified byte-identical);annotate importcomment notes it rewritespilot-1.jsonl(labels identical, schema version updated);app runmentions--port. New v0.2.0 paragraph in the deployment section.After merge: tag
v0.2.0→release.ymlbuilds, smoke-tests and pushes the image; then the CRF bundle is attached, the release notes applied, and the local rollback drill v0.1.0 ↔ v0.2.0 run and recorded.Also:
test_onnx_export_and_int8_parityfailed on one GitHub runner (INT8 typed F1 0.93 vs 1.0 on the tiny fixture; CPU-dependent INT8 kernels, the same failure that delayed v0.1.0). The test now checks that the INT8 path works (loads, every output passes the validator, smaller file, distinct digest) and still requires exact FP32 parity; INT8 accuracy is decided by the trained-model gate inreports/onnx-int8/.