Who contributed to PyTorch on June 28, 2026?

2 developers shipped this update, including pytorchbot and vmoens.

What were the notable PyTorch updates?

Don't link libdl when cross-compiling (#20522), [ExecuTorch][WebGPU] Add cat op + ValueList graph support (aten.cat.default), and [Feature] Replay buffer consume-after-sample support.

@pytorch

PyTorch and the broader machine-learning ecosystem

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

EXECUTORCH CLOSES CROSS-COMPILE GAP, RL SUITE SHIPS MAJOR REPLAY BUFFER OVERHAUL

By RepoJournal · Filed 06:02 UTC on June 28, 2026 · About PyTorch

2 people shipped this

vmoens @vmoens 4 cited

pytorchbot @pytorchbot 1 cited

ExecuTorch fixes a critical cross-compilation bug that was pulling host libraries into bare-metal targets, while TorchRL lands four major features that reshape how distributed RL training manages memory and replay state.

The cross-compilation fix [1] stops CMake from linking the host's libdl into executorch_core when targeting bare-metal systems like ARM Cortex-M or ESP32, a bug that cascaded through the entire exported interface and broke downstream builds. In parallel, ExecuTorch adds WebGPU support for the cat operator with ValueList graph integration [2], expanding the ops available for browser-based inference. On the RL side, four interconnected PRs ship together: replay buffer consumption semantics [3] that let you recycle slots after sampling instead of only writing sequentially, MCAdvantage shared state fixes [4] so multi-actor setups don't duplicate buffer metadata, collector device normalization [5] for CPU LazyTensorStorage writes, and mjlab environment wrapper support [6] to integrate MuJoCo Lab physics-engine environments. PyTorch core continues hardening: the dataloader test suite becomes device-agnostic [7] so pin_memory and multiprocessing tests run on any accelerator backend, not just CUDA, and torch.compile's partitioner gets a fix [8] for models using explicit CUDA streams in training mode.

Action items

→ Upgrade ExecuTorch if cross-compiling for bare-metal targets - this was silently breaking downstream CMake builds pytorch/executorch [immediate]
→ Review RL PRs 3910-3913 if you ship distributed RL training - four linked changes that reshape replay buffer semantics pytorch/rl [plan]
→ Monitor torch.compile fixes for stream-dependent models heading to next release pytorch/pytorch [monitor]

References

[1] Don't link libdl when cross-compiling (#20522) pytorch/executorch
[2] [ExecuTorch][WebGPU] Add cat op + ValueList graph support (aten.cat.default) ↗ pytorch/executorch
[3] [Feature] Replay buffer consume-after-sample support ↗ pytorch/rl
[4] [BugFix] Share MCAdvantage replay-buffer state ↗ pytorch/rl
[5] [BugFix] Normalize collector replay buffer device metadata ↗ pytorch/rl
[6] [Feature] Add mjlab environment wrapper ↗ pytorch/rl
[7] [Test] Make test_dataloader.py device-generic for pin_memory and multiprocessing tests (#185156) pytorch/pytorch
[8] Fix control_deps handling in partitioner for forward/backward extraction (#187695) pytorch/pytorch

Quick answers

What shipped in PyTorch on June 28, 2026?: ExecuTorch fixes a critical cross-compilation bug that was pulling host libraries into bare-metal targets, while TorchRL lands four major features that reshape how distributed RL training manages memory and replay state. In total, 19 commits and 12 pull requests landed.
Who contributed to PyTorch on June 28, 2026?: 2 developers shipped this update, including pytorchbot and vmoens.
What were the notable PyTorch updates?: Don't link libdl when cross-compiling (#20522), [ExecuTorch][WebGPU] Add cat op + ValueList graph support (aten.cat.default), and [Feature] Replay buffer consume-after-sample support.

EXECUTORCH ROUTES REASONING SEPARATELY, PYTORCH MODERNIZES COMPILER STACK

ExecutorTorch's shared OpenAI adapter now exposes optional reasoning content as a first-class output channel, while PyTorch core moves to clang-21 and strips dead C++ macros.

python 20 shipped 1-min read

@pytorch 2 days ago

DYNAMO FIXES GRAPH-BREAKING SDPA PARAMS, PAT SVD LEAK BREAKS DOWNSTREAM TESTS

PyTorch shipped fixes for two critical compilation and module patching bugs overnight that were silently breaking downstream workflows.

python 19 shipped 1-min read

@pytorch 3 days ago

CRITICAL OIDC INJECTION IN DOCS PREVIEW WORKFLOW PATCHED

PyTorch's docs-preview CI trusted fork-controlled artifacts in a context with token-write permissions, exposing the entire build pipeline to code injection.

python 66 shipped 1-min read

@pytorch 4 days ago

PYTORCH AUTOGRAD GETS 7% FASTER, AOTI FIXES SILENT FAILURES

Interned attribute names in autograd.Function shaved microseconds off the hot path while AOTI's scatter operations now properly report errors instead of silently corrupting results.

python 64 shipped 1-min read

Elsewhere on the wire

AI Agents about 21 hours ago

CLINE SHIPS FIVE CRITICAL AGENT FIXES AFTER END-TO-END TESTING BLITZ

Saoudrizwan just landed a cascade of fixes that prevent the agent loop from running wild, stop file reads from resolving to the wrong directory, and plug the compaction pipeline.

ai-agents 16 shipped 2-min read

Local LLMs about 21 hours ago

LLAMA.CPP SHIPS MINIMAX-M3 SPARSE ATTENTION, VLLM UNLOCKS HETEROGENEOUS PREFILL-DECODE

The inference stack gets a new model architecture and a fundamental shift in how prefill and decode can run on different hardware configurations.

+10

llms 125 shipped 1-min read

@CachyOS about 21 hours ago

GLIBC 2.43 REBUILD SHIPS, INSTALLER JUMPS TO C++26

CachyOS pushed glibc 2.43 through the pipeline while the CLI installer upgraded its C++ standard to C++26, signaling a major refresh across core infrastructure.

infra 97 shipped 1-min read

Elixir & Phoenix about 21 hours ago

ECTO FIXES SUBQUERY PLANNING IN JOINS, ELIXIR PATCHES IEX PARSING

Ecto's query planner now handles subqueries correctly in interpolated join filters, while Elixir fixes a syntax error that broke 'not in' as a continuation operator in IEx.