Pin jax to 0.11.0 in CI — jax 0.11.1 hangs the fori_loop cells on CPU - #617
Merged
Conversation
jax 0.11.1 (2026-08-17) regresses lax.fori_loop on the CPU path. numpy_vs_numba_vs_jax's jax cells jit with device=cpu, so the regression bites even on this repo's GPU runners; the unpinned 'pip install -U jax[cuda13]' would resolve the broken release at the next Monday cache rebuild (2026-08-24). Executed in 8.18s under 0.11.0 on 2026-08-17; the fr/fa siblings measured >1800s under 0.11.1. Full evidence and the family-wide exposure map are recorded in QuantEcon/workspace-lectures#49. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Pull request overview
Pins JAX to a known-good version in GitHub Actions workflows to avoid a newly introduced JAX CPU-path regression that causes lecture notebook execution (notably lax.fori_loop in numpy_vs_numba_vs_jax) to hang during CI/cache/publish builds.
Changes:
- Replace unpinned
pip install -U "jax[cuda13]"with a pinned installpip install "jax[cuda13]==0.11.0"in CI workflows. - Add inline workflow documentation explaining the regression context and rationale for pinning (including reference to QuantEcon/workspace-lectures#49).
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| .github/workflows/publish.yml | Pins JAX to 0.11.0 during publish builds to prevent hangs during notebook execution. |
| .github/workflows/ci.yml | Pins JAX to 0.11.0 for PR CI runs to keep execution times bounded and reliable. |
| .github/workflows/cache.yml | Pins JAX to 0.11.0 for scheduled cache rebuilds to avoid cache-time notebook execution hangs. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
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.
jax/jaxlib 0.11.1 (released 2026-08-17 20:31 UTC) regresses
lax.fori_loopon the CPU execution path:numpy_vs_numba_vs_jaxexecutes in seconds under 0.11.0 and exceeds even an 1800 s cell timeout under 0.11.1 (locally bisected — the same jitted fori_loop hangs with and without the deprecateddevice=cpuargument). This repo is one rebuild away: the unpinned -U install would resolve 0.11.1 at the Monday 2026-08-24 cron cache rebuild, and the lecture jits its jax cells with device=cpu, so the CPU-path regression bites despite the GPU runners (executed in 8.18 s under 0.11.0 on 2026-08-17). numpyro stays unpinned — it was already installed after jax and remains compatible with 0.11.0.The full evidence chain, family exposure map, and the unpin condition are recorded in QuantEcon/workspace-lectures#49. One PR per repo with this same change; with jax pre-installed, the lecture's unpinned
!pip install quantecon jaxcell is satisfied and does not upgrade.🤖 Generated with Claude Code