Who contributed to PyTorch on July 4, 2026?

2 developers shipped this update, including tom-arm and roman-janik-nxp.

What were the notable PyTorch updates?

Strip +PTX from CUDA arch list on release/RC builds (in build_env_setup.py) (#188914), Fix quantized ONNX gather export (#188272), and vllm-benchmark: read shared HF cache offline, refresh on nightly (#188659).

@pytorch

PyTorch and the broader machine-learning ecosystem

github ↗

Pick a date

Topics: Python AI / ML Full archive →

The Wire · Showcase

PYTORCH RELEASE WHEELS SHEDDING 80MB OF BLOAT, QUANTIZED EXPORTS FIXED

By RepoJournal · Filed 06:03 UTC on July 4, 2026 · About PyTorch

2 people shipped this

tom-arm @tom-arm 1 cited

roman-janik-nxp @roman-janik-nxp 1 cited

PyTorch 2.13 release builds are stripping PTX from CUDA architectures to reverse an 80-90 MB binary bloat that crept into 2.13.0, while core quantization and inference benchmarking fixes land across the stack.

The binary size explosion came from embedding compute_120 PTX in release wheels [1], a regression from 2.12.1 that only affects releases and RCs; nightly builds keep PTX for forward compatibility. That fix hits alongside quantized ONNX gather export repair [2], which unquantizes tensor inputs before lowering, closing the gap between eager and symbolic execution paths. On the inference front, vLLM benchmarking now mirrors the test-osdc offline cache pattern [3], reading shared HuggingFace cache at runtime and refreshing nightly, with FlashInfer's JIT workspace following the same strategy. Dynamo's control flow is tightening too: module-global random.random now routes through RandomVariable instead of graph-breaking [4], treating RNG values as symbolic. Meanwhile ExecuTorch's Arm backend gains memory-hungry test decorators [5] and Torch 2.12 compatibility patches for quantized decomposition [6], while NXP's Neutron flow now handles sum operations via the new MLIR path [7].

Action items

→ Plan upgrade to next PyTorch release once available; the binary size fix matters for mobile and edge deployments pytorch/pytorch [plan]
→ Review quantized ONNX exports in your pipelines; gather operations now export correctly [ref:2] pytorch/pytorch [monitor]
→ Verify vLLM benchmark runs use offline cache reads; nightly refreshes ensure model freshness without blocking CI pytorch/pytorch [plan]

References

[1] Strip +PTX from CUDA arch list on release/RC builds (in build_env_setup.py) (#188914) pytorch/pytorch
[2] Fix quantized ONNX gather export (#188272) pytorch/pytorch
[3] vllm-benchmark: read shared HF cache offline, refresh on nightly (#188659) pytorch/pytorch
[4] [dynamo] Route module-global random.random through RandomVariable (#188235) pytorch/pytorch
[5] Arm backend: Add xlarge pytest decorator ↗ pytorch/executorch
[6] Arm backend: fix Torch compatibility (#20671) pytorch/executorch
[7] NXP backend: Enable sum with new Neutron flow ↗ pytorch/executorch

Quick answers

What shipped in PyTorch on July 4, 2026?: PyTorch 2.13 release builds are stripping PTX from CUDA architectures to reverse an 80-90 MB binary bloat that crept into 2.13.0, while core quantization and inference benchmarking fixes land across the stack. In total, 37 commits and 4 pull requests landed.
Who contributed to PyTorch on July 4, 2026?: 2 developers shipped this update, including tom-arm and roman-janik-nxp.
What were the notable PyTorch updates?: Strip +PTX from CUDA arch list on release/RC builds (in build_env_setup.py) (#188914), Fix quantized ONNX gather export (#188272), and vllm-benchmark: read shared HF cache offline, refresh on nightly (#188659).

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.