fix test - #83
Merged
Merged
Conversation
romeokienzler
marked this pull request as draft
July 13, 2026 08:09
romeokienzler
marked this pull request as ready for review
July 13, 2026 08:20
- Bump version 0.8.0rc3 -> 0.8.0 - Integration tests assert trained metrics against a committed calibration_baseline.json (pf + opf bounds) via read_baseline(), with an environment fingerprint (warn-on-mismatch, non-gating) - conftest: add --pad option's calibrate_pad fixture - Fix NODE_RANK computation to tolerate FQDN vs short LSF hostnames - gitignore build/ and dist/ Validated on H100 (cu130): calibration run + assertion run both pass (2 passed). Note: this restores the read-baseline/fingerprint approach over main's hardcoded-bounds tests, per request. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Co-authored-by: PUECH Alban <72336171+albanpuech@users.noreply.github.com> Signed-off-by: Romeo Kienzler <5694071+romeokienzler@users.noreply.github.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Signed-off-by: Alban Puech <nabla.hceup@gmail.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Co-authored-by: PUECH Alban <72336171+albanpuech@users.noreply.github.com> Signed-off-by: Romeo Kienzler <5694071+romeokienzler@users.noreply.github.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Co-authored-by: PUECH Alban <72336171+albanpuech@users.noreply.github.com> Signed-off-by: Romeo Kienzler <5694071+romeokienzler@users.noreply.github.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
- print_calibration_stats no longer recomputes the confidence interval. It now takes the bounds from compute_bounds and prints those, so the displayed interval matches what is written to the baseline and asserted against (previously the printer used a bare mean±t-margin with no pad, which collapsed to lo==hi==mean on deterministic runs while the baseline stored the padded interval). compute_bounds is now the single source of truth. Addresses PR #82 review comments r3553876356 / r3553883472. - CONTRIBUTING.md: add "Running the Integration Tests" section explaining contributors must calibrate a baseline on the unchanged code first, then change code and run the tests to detect drift. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
- Untrack integrationtests/calibration_baseline.json and git-ignore it: the bounds are machine-specific (hardware/CUDA/library versions), so each environment calibrates its own. Matches the long-standing code comment. - CONTRIBUTING.md: add "Running the Integration Tests" to the TOC; document the --calibrate/--ci/--pad flags, the MLFLOW_ALLOW_FILE_STORE=true prerequisite, and what to do on an environment-fingerprint mismatch (warn -> recalibrate). - tests/test_node_rank.py: unit-test set_env() NODE_RANK computation, covering exact-match, FQDN-host/short-list fallback, short-host/FQDN-list, master=rank0, and the unresolvable-host RuntimeError. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Calibrating only PF (or only OPF) previously overwrote a single shared top-level fingerprint, falsely claiming the other test's bounds were measured in the current environment. Store the fingerprint per test under fingerprints[test_key]; only the calibrated test's bounds and fingerprint are touched. read_baseline checks the per-test fingerprint, falling back to the legacy top-level key so older baselines still verify. Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
The PyG wheel index has no prebuilt torch-scatter for torch 2.13, so pip fell back to building it from sdist, which fails under build isolation (import torch) with no compiler in the python:3.12-slim image. Pin torch to 2.12.* (newest with a prebuilt torch-scatter wheel, and the version the baseline was validated on). Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
Allow --calibrate without an explicit N (defaults to 5) while keeping omit-flag assert mode unchanged. Update CONTRIBUTING.md and integration test messages to document the three invocation modes and align fingerprint docs. Signed-off-by: Alban Puech <nabla.hceup@gmail.com> Signed-off-by: Romeo Kienzler <romeo.kienzler1@ibm.com>
romeokienzler
force-pushed
the
bump_version_0.8.0
branch
from
July 13, 2026 08:25
9b53d7b to
6e3b8e7
Compare
Signed-off-by: Romeo Kienzler <5694071+romeokienzler@users.noreply.github.com>
albanpuech
approved these changes
Jul 13, 2026
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
2 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.
No description provided.