---
format: "aidr-story-markdown/v1"
id: "96189bd0e3d0b6dd5d938fa02439b8b2b20a705c697e548a8da2367065f48488"
canonical_url: "https://aidr.today/96189bd0?lang=en"
title: "Hugging Face TRL v1.15 Cuts Peak VRAM by 82%"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-09T09:54:28.000Z"
category: "Frameworks"
topics: ["llm","framework"]
source_urls: ["https://huggingnews.com/ai/hugging-face-trl-v115-cuts-peak-vram-by-82percent-d58f615d","https://x.com/QGallouedec/status/2108468004128424042","https://marketbrief.now/ai/hugging-face-trl-v115-cuts-peak-vram-by-82percent-d58f615d"]
summary: "Hugging Face released TRL v1.15, which reduces peak VRAM usage by up to 82% via a fused LM head enabled by default. The update supports sequence lengths up to 7 times greater than previous versions, increasing token capacity for DPO from 10k to 59k and GRPO from 29k to 115k."
---

# Hugging Face TRL v1\.15 Cuts Peak VRAM by 82%

> [Open the canonical story](<https://aidr.today/96189bd0?lang=en>)

**Published:** 2026-10-09T09:54:28.000Z
**Category:** Frameworks
**Topics:** llm, framework

## Summary

Hugging Face released TRL v1\.15, which reduces peak VRAM usage by up to 82% via a fused LM head enabled by default\. The update supports sequence lengths up to 7 times greater than previous versions, increasing token capacity for DPO from 10k to 59k and GRPO from 29k to 115k\.

## Sources

- [Story source](<https://huggingnews.com/ai/hugging-face-trl-v115-cuts-peak-vram-by-82percent-d58f615d>)
- [Story source](<https://x.com/QGallouedec/status/2108468004128424042>)
- [Story source](<https://marketbrief.now/ai/hugging-face-trl-v115-cuts-peak-vram-by-82percent-d58f615d>)