Harness Adapters: RL Training for Agentic Tasks on LFM2.5, LFM3, and Twil-9B

X postMario ZechnerAnnouncement

Younes announced the live launch of LoRA adapter training for LFM2.5, LFM3, and Twil-9B. By applying reinforcement learning, the models aim to achieve higher accuracy and efficiency in agentic tasks. Mario Zechner highlighted the project.

Key points

  1. LoRA adapters are being trained on Liquid AI's LFM2.5 and LFM3, as well as thewebAI's Twil-9B model.
  2. Reinforcement learning is used to boost accuracy and efficiency for agentic workloads.
  3. The training progress is accessible live via a Hugging Face Space alongside a Hugging Face blog post.

Why it matters

Targeted LoRA adaptation through reinforcement learning allows developers to enhance smaller base models specifically for autonomous agent workflows.

Analysis

Live Training for AI Agents

Mario Zechner drew attention to a project by Younes (@yb2698) focusing on training LoRA adapters using reinforcement learning (RL). The training targets LFM2.5 and LFM3 by Liquid AI along with the recently released Twil-9B model by thewebAI.

Enhancing Agentic Workflows

The project specifically tackles performance during agentic tasks. By refining base models with LoRA and RL, the team aims to improve operational efficiency and task completion accuracy. Training metrics and runs can be monitored publicly via Hugging Face Spaces.

Original post

Mario Zechner

@badlogicgames · 6 November 2026

super cool! t.co/b0qPwl9nRp

91 likes2 replies92 bookmarks9.9k views

View on X
Selected replies (2)
  • @MixtTensi8198 @badlogicgames Finding that LFM3 works well in Pi, but with SFT and RL it ends up doing much better at recovering from typicaly failure modes and spooky loops! :)
  • @yb2698 @badlogicgames Thanks !

Mentioned links

Summary generated by AI (Gemini 3.8 Flash, 8 November 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Article:Shuai Shao, Kangning Zhang, Qingyao Li, Shijian Wang, Hao Wang, Wenxiang Jiao, Yuan Lu, Yi Guo, Weiwen Liu, Weinan Zhang

    Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories

    Researchers introduce Harness-R1, a framework that trains a dedicated agent to adapt and patch the runtime harness of LLM agents based on failure logs. Using online reinforcement learning, a separate 9B model learns to generate executable patches for context handling, tool mediation, and error recovery to directly improve task completion.

    AI & LLMs· Research

  • Repository:ornith-ai/Ornith-1

    Ornith: Self-Improving Open-Source Models for Agentic Workflows

    The Ornith-1 repository provides an open-source model family designed for agentic tasks using autonomous self-improvement. Spanning Ornith-1.0 and Ornith-1.5, the project implements self-scaffolding and automated generation of training rollouts for reinforcement learning.

    2kstars

    AI & LLMs· Tool

  • Repository:amosblomqvist/learnharness

    LearnHarness: Open-Source Agent Harness for Adaptive Learning

    LearnHarness uses AI agents to deliver structured, step-by-step tutoring based on the learner's existing comprehension. It captures reasoning paths and retains insights in a persistent knowledge graph. It helps learners and developers master complex subjects systematically.

    3.2kstarsTypeScript

    AI & LLMs· Tool

  • Link:Apodex AI

    Apodex-1.1-mini: Multimodal Agentic Model Released on Hugging Face

    Apodex AI has released Apodex-1.1-mini under the Apache-2.0 licence on Hugging Face. Based on the qwen3_5_moe architecture, the model is designed for long-horizon research and agentic workflows with native tool execution.

    AI & LLMs· Tool

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.