Fine-Tune Muse Glimmer 30B for Free with Unsloth (incl. GRPO)

Melvin Vivas · X post · 2026-08-14 · Open on X

Topics: Fine-tuning & Model Customization · Level: intermediate

Summary

Shares Unsloth's announcement that Meta's Muse Glimmer 30B can now be fine-tuned for free with Unsloth notebooks. The notebooks also support GRPO reinforcement-learning training. Unsloth says it trains 1.5x faster with 50% less VRAM than FlashAttention-2 setups, so training fits locally in 24GB of VRAM.

Key points

• Meta Muse Glimmer 30B can be fine-tuned for free with Unsloth's notebooks.
• The free notebook also supports GRPO, a reinforcement-learning training method.
• Unsloth claims 1.5x faster training and 50% less VRAM than FlashAttention-2 (FA2) setups.
• Local training is possible on a GPU with 24GB of VRAM (e.g., RTX 3090/4090).
• The guide and free notebooks are in Unsloth's docs under models/muse-glimmer/train.

Resources mentioned

• Unsloth Documentation: Muse Glimmer guide · docs · unsloth.ai · free
  Unsloth's guide and free notebooks for fine-tuning and GRPO-training Muse Glimmer 30B.
  Also in: Muse Glimmer 30B: Unsloth GGUF Release and Run Guide (Melvin Vivas on X · notes)
• Unsloth · tool · unsloth.ai · free
  Open-source library for fast, memory-efficient fine-tuning of open LLMs (LoRA/QLoRA) on a single GPU or in Colab.
  Also in: Run Laya Decision models locally with Unsloth on 4GB RAM (Melvin Vivas on X · notes), Unsloth passes 500M model downloads on Hugging Face (Melvin Vivas on X · notes), Run Qwen-Image-2.1 locally on 12GB VRAM with Unsloth GGUFs (Melvin Vivas on X · notes), Base vs fine-tuned Gemma 4 E2B as a model router (Melvin Vivas on X · notes) and 29 more
• Meta Muse Glimmer 30B · tool · huggingface.co · free
  A 30B open model from Meta that can be fine-tuned and run locally.
• FlashAttention-2 · tool · github.com · free
  Optimized attention kernel, used here as the baseline for Unsloth's speed and memory claims.

Try this

• Open the Unsloth Muse Glimmer guide and run the free fine-tuning notebook.
• Try the GRPO RL training option in the notebook.
• Fine-tune Muse Glimmer 30B locally on a 24GB GPU for your own task with Unsloth.