AION
Open-source releaseEfficiency & Inference · Training & Scaling · MLOps, Tooling & Infrastructure1 source · Jan 21, 2026

pytorch/pytorch v2.10.0: PyTorch 2.10.0 Release

<td> Reduced kernel launch overhead with <strong>combo-kernels</strong> horizontal fusion in torchinductor </td>

PyTorchIntelONNX
← Back to brief

Key points

  • Python 3.14t (freethreaded build) is experimentally supported as well.
  • <td> A new <strong>varlenattn()</strong> op providing support for ragged and packed sequences </td>

Sources (1)

  • [1]pytorch/pytorch v2.10.0: PyTorch 2.10.0 Release
    GitHub: pytorch/pytorch · Jan 21, 05:05 PM
    <td> Reduced kernel launch overhead with <strong>combo-kernels</strong> horizontal fusion in torchinductor </td>
    Python 3.14t (freethreaded build) is experimentally supported as well.

Extractive summary: sentences quoted from the sources.