{"slug": "if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning", "title": "If your GPU can run inference, it is now also capable of performing fine-tuning", "summary": "Developer tsuyu122 released USAF, an open-source sparse fine-tuning method for Mixture-of-Experts (MoE) models, enabling fine-tuning on consumer GPUs with as little as 12 GB VRAM. The method, demonstrated on an AMD RX 6750 XT with Qwen3-30B-A3B, trains sparse expert weights and the router instead of adapters, and is available on GitHub under the Apache 2.0 license.", "body_md": "I spent the last few months building a new sparse fine-tuning method for MoE models called **USAF**.\n\nThe goal was simple: if your GPU can run inference on an MoE model, it should also be able to fine-tune it.\n\nOn my AMD RX 6750 XT (12 GB), I can fine-tune Qwen3-30B-A3B by training sparse expert weights and the router instead of adapters.\n\nThe project is completely open source under the Apache 2.0 license. I’m not trying to build a business, sell anything, or monetize it in any way—I just wanted to share something I built that I think is genuinely interesting.\n\nI’d love to hear your feedback, especially from people working with MoE models.\n\nGitHub: [https://github.com/tsuyu122/usaf](https://github.com/tsuyu122/usaf)\n\nInteresting work, perhaps I can integrate this with AReno, which I am playing and developing now.\n\nThis is really interesting. Getting Qwen3-30B-A3B fine-tuning down to a 12GB consumer GPU is impressive.\n\nI’m working on a related problem from the infrastructure side: making model training easier without having to manually choose GPUs, provision compute, and configure the training environment each time.\n\nWould be interesting to see whether a method like USAF could eventually plug into that kind of automated training workflow.", "url": "https://wpnews.pro/news/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning", "canonical_source": "https://discuss.huggingface.co/t/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning/177456#post_3", "published_at": "2026-08-18 13:02:24+00:00", "updated_at": "2026-08-18 13:13:54.114831+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "ai-tools"], "entities": ["USAF", "tsuyu122", "AMD RX 6750 XT", "Qwen3-30B-A3B", "GitHub", "Apache 2.0"], "alternates": {"html": "https://wpnews.pro/news/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning", "markdown": "https://wpnews.pro/news/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning.md", "text": "https://wpnews.pro/news/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning.txt", "jsonld": "https://wpnews.pro/news/if-your-gpu-can-run-inference-it-is-now-also-capable-of-performing-fine-tuning.jsonld"}}