Skip to content

log of updates. Dockerfile revamp, pipeflush previous layers - #4

Merged
atharvas merged 12 commits into
pipeflushfrom
redo-dockerfile
Sep 2, 2025
Merged

log of updates. Dockerfile revamp, pipeflush previous layers#4
atharvas merged 12 commits into
pipeflushfrom
redo-dockerfile

Conversation

@atharvas

@atharvas atharvas commented Sep 1, 2025

Copy link
Copy Markdown
Member

No description provided.

@atharvas
atharvas merged commit 5491a8f into pipeflush Sep 2, 2025
4 checks passed
atharvas added a commit that referenced this pull request Sep 2, 2025
* adding info upto now

* adding info upto now

* update again

* adding scikit-learn context

* small changes

* upgrade pkgs

* saving progress till now; implemented an agent for filtering commits + start storing patch info

* validation working

* partial auto-installer working

* auto-installer working pipe-flush single

* partially working autobuilder

* added portkey

* Better backup sync

* update benchmarking

* use dotenv

* remove unneccessary type hints

* remove support for 3.13

* updating build agent to be reactive

* small errors

* lot of updates. Dockerfile revamp, pipeflush previous layers (#4)

* lot of updates. Dockerfile revamp, pipeflush previous layers

* Adding PORTKEY_MODEL_NAME function for easier control.

* Increase timeout for container conflict. This caused a crash.

* config_agent_backend should be on top level.

* Add additional logging for containers.

* Correcting small bugs in dockerfile building.

* Stop bypassing cache.

* Help avoid race conditions in caching.

* Update commit_collection code to remove caching + log tagged and perf improving commits.

* Update various scripts

* example notebook

* Update README
@atharvas
atharvas deleted the redo-dockerfile branch April 8, 2026 15:12
atharvas added a commit that referenced this pull request Aug 23, 2026
Human decision after round 1 review: at build time this invariant
cannot be a useful fatal gate either way -- not creating __init__.py
hard-fails repos that legitimately lack one, creating it
unconditionally makes the check tautological (the same dead-check
pattern this plan has hit before). Its real value is comparing the
sealed build-time value against a later trial-time observation
(agent-deleted __init__.py), which lands in a later plan. The
invariant stays registered and keeps recording the fact; it just
stops gating the build.

- manifest.py: INVARIANTS severity fatal -> warn for
  benchmark_init_missing, with a comment recording why and warning
  against restoring it without the trial-time comparison.
- local_ci.py: removed benchmark_init_missing from _FATAL_INVARIANTS
  and deleted the now-unused _c_init_present, with a pointer comment
  back to manifest.py's reasoning.
- Canonical fatal set is now six ids (was seven).
- test_manifest.py: renamed/updated the now-stale
  test_benchmark_init_missing_is_fatal to assert warnings instead of
  fatal; restructured the parity test's case table to prove both
  implementations agree the flip is no longer fatal, rather than
  silently dropping coverage.
- Verified the parity test added in round 1 does its job: ran it after
  changing manifest.py's severity but before touching local_ci.py --
  it failed immediately (via the case table's self-check, since
  manifest.py's fatal set no longer matched the test's still-fatal
  expectation for this id), demonstrating it catches exactly this
  one-sided-drift scenario.
- Updated the build-manifest-verification design doc's invariant table
  to match (row #4, with a footnote).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant