Zlio
Everything in AI and robotics, on one page.
· Free & non-commercial
Updated 02:31 UTC · Next update 03:00 UTC
Zlio Daily, the newsletter, is coming soon. The day's essential AI and robotics news in your inbox. Leave your email and we'll tell you when it launches.
One email when we launch. No spam, unsubscribe anytime. Privacy
Developer Tools · first seen 14 November 2026, 12:15 UTC

Hugging Face fixes TRL CISPO default that shrank gradients fivefold

Hugging Face's TRL v1.18.0 sets the CISPO default to 5.0 after an inherited 0.2 ceiling clipped on-policy weights near 1, shrinking gradients fivefold in reproduction.

Hugging FaceOpenFreeCode
Domain
general

Read it at RuntimeWire →

Share LinkedIn X Bluesky WhatsApp

Coverage: 1 sources, in order

  1. RuntimeWire Read the article →

← Front page