Skip to content
Trending storyDeveloping

Lewis Tunstall shares TRL v1.15.0 release notes and a Triton kernel for long-context training

2 articles1 sourcesince Oct 10Last article 1h ago ·

Overview

AISummary of 2 articles

Hugging Face's Lewis Tunstall announces the release of TRL v1.15.0, with release notes linked on GitHub.

In a second post, he says v1.15 includes a new Triton kernel that scales SFT and RL training to over 100k tokens on the same GPU, and urges TRL users to upgrade. The posts do not describe other changes in the release, and the 100k-token figure is Tunstall's own claim.

Written by AI from the articles below · updated Oct 10, 1:42 PM ET

Check the sources:

Latest developmentOct 10, 1:38 PM ET
TRL v1.15.0 is released on GitHub

Developments

2 developments

  1. Oct 10, 1:38 PM ET · 1 article
    Lewis Tunstall shares release notes for TRL v1.15.0
    Lewis Tunstall: TRL v1.15.0 is released on GitHub
  2. Oct 10, 1:38 PM ET · 1 article
    TRL v1.15 adds Triton kernel scaling SFT/RL beyond 100k tokens on one GPU
    Lewis Tunstall: TRL v1.15 adds Triton kernel scaling SFT and RL past 100k tokens

Article timeline

The articles in this story. Times are ET.

Oct 10
  1. Lewis Tunstall @ COLM 🌉X
    TRL v1.15.0 is released on GitHub

    AIHugging Face's Lewis Tunstall announces the release of TRL v1.15.0, with release notes linked on GitHub. The post gives no further details of the changes.

  2. Lewis Tunstall @ COLM 🌉X
    TRL v1.15 adds Triton kernel scaling SFT and RL past 100k tokens

    AIHugging Face's Lewis Tunstall says TRL v1.15 includes a new Triton kernel that scales SFT and RL training to over 100k tokens on the same GPU. He urges users of TRL to upgrade to the release.

    Image from @_lewtun's post

Heat trend

Not enough continuous observations to show a trend yet.