GitHub · Where software is built

8000 GitHub · Where software is built

[RFC] Proposed Changes to Feature Tracking & Classification for PyTorch Releases starting Release 2.8
#152134 · atalman opened on Apr 24, 2025
8
[v.2.8.0] Release Tracker
#156745 · atalman opened on Jun 24, 2025
70

Labels Milestones New issue

[CUDA] Add experimental green context support for SM carveout

matrix multiplication

pytorch/pytorch

#159104

· eqy opened

on Jul 25, 2025

torch.compile doesn't work with torch.mm(out_dtype=...) variant

matrix multiplication

module: inductor

#159023

· yf225 opened

on Jul 24, 2025

[CUDA][FP8] Expect SM110 to fail FP8 matmul tests

matrix multiplication

module: floatx (formerly float8)

topic: not user facing

pytorch/pytorch

#158309

· eqy opened

on Jul 15, 2025

[Inductor] addmm + activation function fusion

matrix multiplication

module: inductor

topic: not user facing

pytorch/pytorch

#158137

· AaronWang04 opened

on Jul 11, 2025

grouped_mm optional zero initialization of the output

matrix multiplication

module: python frontend

#154557

· IvanKobzarev opened

on May 28, 2025

[cuBLASLt][cuBLAS] Support 2D bias and <code>beta != 1.0</code> in cuBLASLt

ciflow/inductor-rocm

ciflow/rocm-mi300

matrix multiplication

topic: not user facing

pytorch/pytorch

#154170

· eqy opened

on May 23, 2025

matmul uses excessive memory in batch cases with more than 3 dimensions

matrix multiplication

module: memory usage

#154128

· JackBuck opened

on May 22, 2025

[cuBLAS] Strange behavior with <code>out_dtype</code> argument in <code>torch.addmm</code>

matrix multiplication

module: bfloat16

#154086

· eqy opened

on May 22, 2025

<code>torch.bmm</code> is slow on non-contiguous BF16 CPU tensors

matrix multiplication

module: bfloat16

module: performance

#151934

· ar0ck opened

on Apr 22, 2025

BF16 linear(matmul) operator 100x slower on odd matrix dimension sizes on A100

matrix multiplication

module: linear algebra

module: performance

#146817

· piubwd opened

on Feb 10, 2025

Addmm is not always better than add + mm

matrix multiplication

module: performance

#141210

· ConnollyLeon opened

on Nov 21, 2024

3D matrix support in _scaled_mm

matrix multiplication

module: floatx (formerly float8)

#129803

· rybakov opened

on Jun 28, 2024

0