Who contributed to PyTorch on June 15, 2026?

3 developers shipped this update, including oscarandersson8218, wwwjn, and tianyu-l.

What were the notable PyTorch updates?

Fix FakeTensor constructor fallback for custom ops without device (#185107), [BE][Ez]: Add missing functorch batch rules for copy views (#187256), and [aten] Fix CPU and MPS logit for eps > 0.5 (#181297).

@pytorch

PyTorch and the broader machine-learning ecosystem

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

PYTORCH FIXES CRITICAL FAKETEN OR EDGE CASE AS EXECUTORCH ADDS ARM TOSA SUPPORT

By RepoJournal · Filed 06:03 UTC on June 15, 2026 · About PyTorch

3 people shipped this

oscarandersson8218 @oscarandersson8218 1 cited

wwwjn @wwwjn 1 cited

tianyu-l @tianyu-l 1 cited

FakeTensor's constructor fallback broke custom ops without device arguments, potentially blocking symbolic tracing workflows across the ecosystem.

The pytorch/pytorch team shipped a fix [1] for FakeTensor's overly aggressive device kwarg injection, which was failing symbolic tracing on custom operators that don't accept device parameters. This is a silent correctness issue that hits anyone building custom ops without device arguments. Meanwhile, functorch batch rules got a substantial refresh [2] removing xfails and improving coverage for copy view operations, and the CPU and MPS logit implementations now correctly handle edge cases where eps > 0.5 [3], fixing a sign-flip bug that would have silently corrupted numerical results. On the X86 front, deprecated quantization fusion patterns have been removed and moved to torchao , streamlining the inductor codebase. The argmin/argmax fix [4] extends CPU support to match the Triton/GPU path, ensuring boolean tensor reductions compile correctly across all backends. ExecutorchArm backend work [5] adds TOSA dialect implementations for data layout operations (CONCAT, RESHAPE, REVERSE, TILE, TRANSPOSE), expanding mobile deployment options. TorchTitan now enables Expert Parallelism for MoE models in vLLM inference [6], pairing TP on dense layers with EP on experts for better scaling, while simultaneously fixing a critical Qwen3 MoE bug [7] where load balancing hooks weren't registered despite being configured.

Action items

→ Rebuild and test custom ops if using symbolic tracing or FakeTensor workflows pytorch/pytorch [immediate]
→ Review CPU/MPS logit usage for eps values and rerun validation pytorch/pytorch [plan]
→ Migrate quantization fusion patterns from X86 inductor to torchao pytorch/pytorch [plan]
→ Pull latest TorchTitan if running Qwen3 MoE or vLLM inference pytorch/torchtitan [monitor]

References

[1] Fix FakeTensor constructor fallback for custom ops without device (#185107) pytorch/pytorch
[2] [BE][Ez]: Add missing functorch batch rules for copy views (#187256) pytorch/pytorch
[3] [aten] Fix CPU and MPS logit for eps > 0.5 (#181297) pytorch/pytorch
[4] Fix argmin/argmax on boolean tensors for CPU and non-Triton backends (#185970) pytorch/pytorch
[5] Arm backend: Add TOSA dialect data layout ops ↗ pytorch/executorch
[6] [rl] Enable TP2EP for unified MoE model in vLLM wrapper ↗ pytorch/torchtitan
[7] support gpt-oss PP; fix qwen3 MoE expert bias udpate; some other cleanups ↗ pytorch/torchtitan

Quick answers

What shipped in PyTorch on June 15, 2026?: FakeTensor's constructor fallback broke custom ops without device arguments, potentially blocking symbolic tracing workflows across the ecosystem. In total, 35 commits and 5 pull requests landed.
Who contributed to PyTorch on June 15, 2026?: 3 developers shipped this update, including oscarandersson8218, wwwjn, and tianyu-l.
What were the notable PyTorch updates?: Fix FakeTensor constructor fallback for custom ops without device (#185107), [BE][Ez]: Add missing functorch batch rules for copy views (#187256), and [aten] Fix CPU and MPS logit for eps > 0.5 (#181297).

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.