{"slug": "mini-agi-actually-trains-on-8gb-vram-without-exploding", "title": "Mini-AGI actually trains on 8GB VRAM without exploding", "summary": "A project called Mini-AGI is training a dynamic continual learning model on 8GB of VRAM by combining dynamic Mixture of Experts, which adds and prunes experts during training, with batch-1 training on a single continuous stream. The model is currently processing a corpus of 7.8B characters in interleaved 32K-character passages, and its weights are not expected to be released for another couple of weeks. The code is available via git clone from the volotat/mini-AGI GitHub repository, and the author says the architecture was brainstormed with Claude.", "body_md": "# Mini-AGI actually trains on 8GB VRAM without exploding\n\nCalling a project \"Mini-AGI\" is a bold move that usually invites a shower of rocks, but this one actually has some technical meat on its bones. The core claim here is a dynamic continual learning model that manages to train on a measly 8GB of VRAM. For those of us tired of the \"you need a H100 cluster to breathe\" era of LLMs, this is a refreshing change of pace.\n\nThe project is basically a rebellion against the fact that we can fine-tune 1B+ models on consumer gear, but we can't actually train them from scratch without selling a kidney. The goal was to get full control over the training data instead of just trusting whatever corporate slurry OpenAI or Google fed their models.\n\n## How it handles the VRAM bottleneck\n\nThe author managed to dodge the massive memory requirements using two specific architectural gambles:\n\n- **Dynamic MoE (Mixture of Experts):** Instead of a static wall of parameters, the model adds and prunes experts during training. Only a tiny subset of experts are active at any given moment. This effectively shifts the bottleneck from VRAM to disk space, as experts are loaded and unloaded on the fly.\n- **Batch 1 Training:** This is the real trick. By training on a single continuous stream of data, the model avoids the need to store massive randomized batches and their corresponding gradients, which is usually what kills your GPU memory.\n\n## Current progress and training stats\n\nIf you're expecting to download the weights today, stop. The model is currently chewing through a corpus of 7.8B characters. The training process involves reading interleaved passages of 32K characters each as a single, continuous stream.\n\nAccording to the project's current pace, the weights are still \"cooking\" and won't be ready for another couple of weeks. The author even shared a scaling law graph that they claim looks promising, though we'll see if that holds up once the weights actually hit the public.\n\n## Getting it running\n\nThe setup is straightforward if you want to watch the process yourself. Since it's designed for low-VRAM environments, you don't need a server farm to test the implementation.\n\n```\ngit clone https://github.com/volotat/mini-AGI\n```\n\nThe author admits to brainstorming the architecture with [Claude](https://promptcube3.com/en/tags/claude/), which proves that using AI to build AI is the only way we're actually getting anything done these days. It's a lean approach to training that prioritizes disk swapping and stream-processing over raw hardware brute force. Whether it actually achieves \"AGI\" (even a mini one) remains to be seen, but the 8GB VRAM footprint is a win for the home-lab crowd.\n\n[Next Opening an untrusted repository with an AI coding agent is riskier than just reading the files yourself →](https://promptcube3.com/en/threads/9521/)", "url": "https://wpnews.pro/news/mini-agi-actually-trains-on-8gb-vram-without-exploding", "canonical_source": "https://promptcube3.com/en/threads/9545/", "published_at": "2026-09-21 13:46:17+00:00", "updated_at": "2026-09-21 13:53:51.082564+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-research", "ai-infrastructure"], "entities": ["Mini-AGI", "GitHub", "Claude", "OpenAI", "Google", "volotat"], "alternates": {"html": "https://wpnews.pro/news/mini-agi-actually-trains-on-8gb-vram-without-exploding", "markdown": "https://wpnews.pro/news/mini-agi-actually-trains-on-8gb-vram-without-exploding.md", "text": "https://wpnews.pro/news/mini-agi-actually-trains-on-8gb-vram-without-exploding.txt", "jsonld": "https://wpnews.pro/news/mini-agi-actually-trains-on-8gb-vram-without-exploding.jsonld"}}