Skip to content

v1.2.22

Choose a tag to compare

View all tags
@Hecate0821 Hecate0821 released this 10 Oct 03:46
· 9 commits to main since this release
v1.2.22
130a53d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

1.2.22 (2026-10-10)

Full Changelog: v1.2.21...v1.2.22

Features

  • training-sdk: preserve optional Parquet sampling-support references through deployment sampling, Tinker-style sampled sequences, model inputs, and TITO; target-policy support selection remains off by default
  • training-sdk: accept the trainer's built-in DPPO loss name

Bug Fixes

  • training-sdk: retain HTTP request/response diagnostics when an older serving image rejects sampling-support capture
Assets 2
Loading