pytorch/pytorch v2.12.0: PyTorch 2.12.0 Release

Key points

• <tr><td><strong>torch.export.save</strong> now supports Microscaling (MX) quantization formats, enabling full export of aggressively compressed models.</td></tr>1
• <tr><td><strong>torch.cond</strong> control flow can now be captured and replayed inside CUDA Graphs.</td></tr>1
• <tr><td><strong>ROCm</strong> users gain expandable memory segments, rocSHMEM symmetric memory collectives, and FlexAttention pipelining.</td></tr>1
• Below are the full release notes for this release.1

Sources (1)

• [1]pytorch/pytorch v2.12.0: PyTorch 2.12.0 Release
  GitHub: pytorch/pytorch · May 13, 05:38 PM

  <tr><td><strong>Batched linalg.eigh on CUDA</strong> is up to 100x faster due to updated cuSolver backend selection.</td></tr>

  <tr><td><strong>torch.export.save</strong> now supports Microscaling (MX) quantization formats, enabling full export of aggressively compressed models.</td></tr>

Extractive summary: sentences quoted from the sources.