Arcee AI | Use MergeKit to Extract LoRA Adapters from any Fine-Tuned Model

Arcee AI | Arcee AI and mergekit unite

Several months ago, I stumbled upon an innovative technique in the world of language model training known as Model Merging. This SOTA approach involves the fusion of two or more LLMs into a singular, cohesive model, presenting a novel and experimental method for creating sophisticated models at a fraction of

Sakana AI on Twitter / X

We’re excited to introduce Doc-to-LoRA and Text-to-LoRA, two related research exploring how to make LLM customization faster and more accessible.https://t.co/wGKDNhBcJXBy training a Hypernetwork to generate LoRA adapters on the fly, these methods allow models to instantly… pic.twitter.com/gId3J6hgEr— Sakana AI (@SakanaAILabs) February 27, 2026

Arcee AI | March is Merge Madness

To celebrate Arcee’s recent merger with mergekit, we’re bringing you a month of resources and knowledge on model merging.

Sakana AI on Twitter / X

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025!Paper: https://t.co/2FRiVF1UXJCode: https://t.co/rx4G7dq1SWBiological systems are capable of… pic.twitter.com/UdUYfqRXBS— Sakana AI (@SakanaAILabs) June 12, 2025

arcee-ai/mergekit

Tools for merging pretrained large language models.

Osaurus — All your AI. One app.

Chat with GPT-5, Claude, Llama, and more — or download local MLX models from Hugging Face. Supercharge Cursor with powerful tools. Open source and built for Mac.

Arcee AI | Announcing the Arcee Model Engine Public Beta

Get direct access to the small language models (SLMs) that power Arcee Orchestra, our new end-to-end, SLM-powered agentic AI platform. Sign up for the public beta of the Arcee Model Engine today.

Ellora: Enhancing LLMs with LoRA - Standardized Recipes for Capability Enhancement

A Blog post by Asankhaya Sharma on Hugging Face

swyx on Twitter / X

whoa so @thinkymachines is doing model merging + customized RLquite a come-up for merging in the past couple weeks, with @arcee_ai mergekit also featuring heavily in AFM. credit due to @jeremyphoward for being the first to make me take modelmerging seriously pic.twitter.com/DtXjX8li4t— swyx (@swyx) June 24, 2025

Seeed-Solution/MeshClaw

Bring AI to the Mesh — OpenClaw plugin for Meshtastic LoRa · No Internet Required

**An Edge-First Generalized LLM LoRA Fine-Tuning Framework for Heterogeneous GPUs**

A Blog post by QVAC on Hugging Face

Home | Mirascope

The complete toolkit for AI engineers

Maintain the unmaintainable - a Hugging Face Space by transformers-community

This application visualizes how over 400 machine learning models in the Transformers library are connected and related to each other. Users can see which models are derived from or import features ...

Introducing the v0 composite model family

Learn how v0's composite AI models combine RAG, frontier LLMs, and AutoFix to build accurate, up-to-date web app code with fewer errors and faster output.

Sebastian Raschka, PhD (@rasbt)

Is LoRA (Low Rank Adaptation) relevant in 2025 for reasoning models? I recently read "Tina: Tiny Reasoning Models via LoRA (https://arxiv.org/abs/2504.15777)", and it made me pause for a moment: when was the last time I heard someone excitedly talk/write about LoRA? LoRA was one of the most influential fine-tuning methods in the earlier LLM boom (as you may remember, I wrote about it a lot in recent years). The idea is simple but effective: avoid full model updates and instead inject a small number of trainable parameters for downstream tasks. This drastically reduces memory and compute costs. But in the age of ever-larger instruction-tuned models coupled with well-working distillation techniques (like popularized by DeepSeek-R1 etc), LoRA seemed to become more irrelevant recently. What about LoRA work for developing reasoning models? This paper tackles exactly that question. Instead of the usual supervised fine-tuning or instruction distillation pipeline, the authors use LoRA with reinforcement learning (RL) to improve reasoning capabilities. Specifically, they fine-tune a 1.5B base model using LoRA adapters while applying RL on reasoning benchmarks. Their baseline model is DeepSeek-R1-Distill-Qwen-1.5B, which is a model already fine-tuned for reasoning tasks. (I wish they started with the base Qwen-1.5B model; but this way, I guess they have more comparisons with other methods that further trained the DeepSeek-R1-Distill-Qwen-1.5B.) From there, the authors ran experiments across datasets, learning rates, LoRA ranks, and RL algorithms. Their best-performing model was trained on just 7k examples and cost just $9 to train. Even with hyperparameter sweeps and multiple ablations, the entire study cost just $526. So, how well does LoRA work? The top half of the results figure (highlighted in blue) compares models trained with LoRA-based RL versus standard RL (i.e., no LoRA). On every benchmark (AIME24, AIME25, AMC23, MATH500, GPAQ, Minerva), LoRA outperforms the regular RL baseline when applied to the same starting model. Insights from ablations 1) Surprisingly, the best-performing model came from the smallest dataset: just 7k examples from Open-RS. 2) The classic LoRA rank 16 emerged as the sweet spot, but ranks 8 and 32 also worked well. 3) It's nice that they included the recent Dr. GRPO (I recently discussed it in my latest Ahead of AI blog). It substantially reduces training time by length-normalizing rewards and addressing issues in GRPO Bottom line: Reasoning is certainly an interesting use case, and it's interesting (and a bit surprising) that LoRA does so well here. It might also be the first case where I've seen LoRA coupled with RL, which is another interesting aspect. LoRA certainly peaked in popularity 1-2 years ago, and more people now consider (more expensive) full-parameter updates (based on anecdotal perception); there's still a place for LoRA and LoRA-like methods. Let's not forget that one of the key advantages of LoRA is that it doesn't modify the underlying base model. This is key in applications where you either have lots of specialized use cases or lots of customers. For example, instead of storing 100 1B full-parameter tuned models, it would be much cheaper to store a 32B model with 100 sets of LoRA weights.