RepoJournal
PyTorch @pytorch
77 wires and counting

$ follow PyTorch

Keep up with PyTorch in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-09-03
stories 125
GitHub RSS Archive Topics

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

PyTorch 2.14.0 ships, import reload guarded, AI reviewer unblocks merges

By RepoJournal · Filed · About PyTorch

ethche albanD Jean Schmidt izaitsevfb atalman vmoens Anthony Shoumikhin
16 people shipped this
ethche @ethche 1 cited albanD @albanD 1 cited Jean Schmidt @jeanschmidt 1 cited izaitsevfb @izaitsevfb 1 cited atalman @atalman 1 cited vmoens @vmoens 1 cited Anthony Shoumikhin @shoumikhin 1 cited Eugenio "Jay" Zuccarelli @jayzuccarelli Klaus Zimmermann @zklaus srhust @srhust apbose @apbose lanluo-nvidia @lanluo-nvidia github-actions[bot] @github-actions[bot] Per Åstrand @per Jiseong-oh @Jiseong-oh oscarandersson8218 @oscarandersson8218

PyTorch 2.14.0 is out with NVGEMM on Inductor, while a safety guard stops reload(torch) segfaults and the AI CI Advisor now clears non-blocking failures.

PyTorch 2.14.0 released, bringing NVGEMM, CuTeDSL-generated CUTLASS kernels to Inductor, alongside the usual backwards incompatible changes, deprecations, and security fixes [1]. The release matrix now advertises 2.14.0 as stable, with the getting-started page and generated entries updated accordingly [2].

Re-running `import torch` after a failed import is now rejected; previously a `reload(torch)` could pass once but then segfault from the second call onwards. This is technically breaking for that edge case, but the old state was "a precarious state primed for segfault at any moment" [3].

On the test-infra side, the AI CI Advisor's cleared verdicts now move failures into a new non-blocking `AI_NOT_RELATED` bucket, so jobs the advisor clears no longer block merges [4]. The greenlight reviewer also gained a time budget hook that nudges progress as deadlines approach [5]. In TorchRL, the `AsyncEnvPool` shutdown deadlock with `exchange="queue"` is fixed; unread results left in the result queue caused `shutdown()` to hang indefinitely once buffers filled [6]. Executorch fixed causal attention masking where query length was shorter than keys, with error against eager up to 3.4 in that case [7].

Action items

References

  1. [1] PyTorch 2.14.0 Release ↗ pytorch/pytorch
  2. [2] Release 2.14 go live. Update release matrix ↗ pytorch/test-infra
  3. [3] Reject re-running `import torch` after a failed import (#194654) ↗ pytorch/pytorch
  4. [4] Add an AI_NOT_RELATED bucket for advisor-cleared failures ↗ pytorch/test-infra
  5. [5] Give the greenlight reviewer a review time budget ↗ pytorch/test-infra
  6. [6] [BugFix] Fix AsyncEnvPool queue-exchange shutdown deadlock ↗ pytorch/rl
  7. [7] Give causal attention the mask PyTorch asks for (#22443) ↗ pytorch/executorch

$ ls pytorch/reviews/ # the step-back reads

the week in review PyTorch 2.14.0 ships, security docs clarified the month in review ExecuTorch MoE quantization, Helion runtime splits, and a broader cleanup wave
Topics: Python AI / ML Full archive →

Quick answers

What shipped in PyTorch on September 3, 2026?
PyTorch 2.14.0 is out with NVGEMM on Inductor, while a safety guard stops reload(torch) segfaults and the AI CI Advisor now clears non-blocking failures. In total, 88 commits, 35 pull requests, and 2 releases landed.
Who contributed to PyTorch on September 3, 2026?
16 developers shipped this update, including ethche, albanD, Eugenio "Jay" Zuccarelli, Klaus Zimmermann, srhust, apbose, lanluo-nvidia, and Jean Schmidt, and 8 more.
What were the notable PyTorch updates?
PyTorch 2.14.0 Release, Release 2.14 go live. Update release matrix, and Reject re-running `import torch` after a failed import (#194654).
♥

$ log @pytorch --recent

MPS complex solves land, Arm round() matches torch, dynamo fixes repro fidelity 24 shipped Torchtitan trains MoE routers in fp32 with bf16x9 on Blackwell 73 shipped executorch fixes dropped edge-manager passes; TorchTitan overhauls MxFP8 160 shipped Inductor fixes stale captured-tensor mutations in while_loop 117 shipped

$ log --all # elsewhere on the wire

Meta React Native adds Stable API guards to renderer attributed string FastAPI & Pydantic Pydantic fixes Mypy plugin crash, tightens datetime constraints

$ ls showcases/

Rails Django Laravel Node.js Go Vercel Vue.js Kubernetes HashiCorp Arch Linux Linux Anthropic

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.