Latest releases

GitHub releases from top ML projects, refreshed on a schedule.

Project	Release	Summary
PyTorch	ciflow/torchtitan/201768: Update October 24, 2026 00:04 UTC	[ghstack-poisoned]
PyTorch	ciflow/torchtitan/201769: Update October 24, 2026 00:04 UTC	[ghstack-poisoned]
PyTorch	ciflow/inductor/201767: Update October 24, 2026 00:04 UTC	[ghstack-poisoned]
PyTorch	ciflow/torchtitan/201767: Update October 24, 2026 00:04 UTC	[ghstack-poisoned]
PyTorch	v2.15.0-rc8 October 23, 2026 20:30 UTC	[release/2.15] [c10d][nccl2] Make endCoalescing cover all ops in the …
llama.cpp	b11845: CUDA: fix build after chunked CUDA graphs (#30804) (#31053) October 23, 2026 18:47 UTC	CUDA: fix build after chunked CUDA graphs ( #30804 ) Apply suggestion from @JohannesGaessler Co-authored-by: Johannes Gäßler johannesg@5d6.de Fix newlines in common.cuh Co-authore…
llama.cpp	b11844 October 23, 2026 17:56 UTC	allows the usage of 4-10 gpus for stepfun and laguna in tensor paral…
llama.cpp	b11843: qwen4exp : build the QSA selection without the kq_mask (#30965) October 23, 2026 17:49 UTC	qwen4exp : build the QSA selection without the kq_mask clean lang qwen4exp : reshape the fused QSA selection to the kq_mask streams ggml_top_k_mask with a single keep row returns …
llama.cpp	b11840 October 23, 2026 16:51 UTC	hexagon: add CEIL op ( #31006 ) hex-ceil: add CEIL(f32) op hex-ceil: add f16 version hex-ops: update supported ops hex-ops: update docs with supported ops Website: https://llama.a…
llama.cpp	b11841 October 23, 2026 16:26 UTC	CUDA: chunk graphs to amortize graph launch ( #30804 )
Diffusers	Diffusers 0.42.0: FLUX.3, Fibo 2.0, Kandinsky 6, and more October 23, 2026 15:39 UTC	Tip This release brings FLUX.3 Image Klein, Bria Fibo 2.0, Kandinsky 6, SANA-WM, and RFD4-Proteína to Diffusers, alongside new adapter support, hardware integrations, and modular …
PyTorch	trunk/23520dd5b595b832ace93bc5516a5c62488c6542 October 23, 2026 14:14 UTC	Revert "[Inductor] Don't specialize dynamic dims in the persistent-TM…
llama.cpp	b11838 October 23, 2026 13:39 UTC	ggml : fix time overflow in ggml_time_us ( #30969 ) fix time overflow use double Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/590936…
PyTorch	trunk/132fec6b15c8d0c9caf4a7e59b0c39f294a81517 October 23, 2026 10:23 UTC	[Feature Refactoring] Support non-CUDA accelerators in ShardedGradSca…
PyTorch	trunk/022c55e97874314b7d7fefde62bb6310384bfd27: Fix forward AD of triangular_solve and cholesky_solve (#198602) October 23, 2026 10:08 UTC	Fixing an Issue Before submitting, please review: The Ultimate Guide to PyTorch Contributions AI-Assisted Development policy Issue Fixes #196694 Fixes #196707 Part of #198017 . Su…
PyTorch	trunk/1776b988238f9978a57690ad7e9b37bff539555c: [MPS] Replace MPSGraph max_pool2d forward with Metal kernels (#201756) October 23, 2026 08:26 UTC	The generic max_pool Metal kernel decomposes the thread id with per-dim div/mod by runtime sizes and is ~8x slower than memory bandwidth allows. That is why stride-1 max_pool2d wa…
llama.cpp	b11836 October 23, 2026 08:23 UTC	CANN: missing const qualifiers added for consistency ( #30933 ) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/58989960 macOS/iOS: mac…
PyTorch	trunk/76bb73d4599d7fde5957268de00fcaacebadbec7 October 23, 2026 07:59 UTC	[Testcase Refactoring] Resolve default backend from device in common …
PyTorch	trunk/76b7f0188deaecd85777d1fdc1e03978c8366b2e: [Test][XPU] Refactor reductions tests by device scope (#194350) October 23, 2026 07:39 UTC	Split reductions tests into CPU-only and device-scoped classes, update device test instantiation, and add matching hardware classification (GENERIC for non-instantiated class, ACC…
llama.cpp	b11835 October 23, 2026 07:32 UTC	sycl : pass device into mul_mat_uses_onednn ( #31014 ) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/58974916 macOS/iOS: macOS Apple …