GitHub उपयोगकर्ता volotat ने mini-AGI जारी किया है, एक निरंतर सीखने वाला बाइट-स्तरीय भाषा मॉडल जिसे 8 GB VRAM के साथ उपभोक्ता हार्डवेयर पर से प्रशिक्षित करने के लिए डिज़ाइन किया गया है। परियोजना उस सीमा को संबोधित करती है कि 1B+ पैरामीटर स्केल पर मॉडल प्रशिक्षित करना आमतौर पर उद्यम-ग्रेड बुनियादी ढांचे की आवश्यकता होती है। एक मिश्रण विशेषज्ञ (MoE) वास्तुकला लागू करके जो प्रशिक्षण के दौरान dynamically adds and prunes experts（dynamic adds and prunes experts 此处意译为动态添加和剪枝专家，需确保符合 hindi 表达习惯） during training, and utilizing batch-1 training on a single continuous data stream, the model bounds parameter count by available disk space rather than VRAM capacity.

The system loads and unloads experts to GPU memory only when needed, eliminating the need to store large randomized batches and gradients. This allows the model to read continuous interleaved passages—each 32K characters long—mimicking human sequential learning. The current training run processes a 7.8B character corpus, expected to take several weeks to complete.

The author acknowledges extensive collaboration with Claude during development, stating the project would be impossible without AI assistance. While labeled a "toy-level model" not matching frontier capabilities, it demonstrates that continual learning without catastrophic forgetting is achievable on modest hardware. The repository is publicly available for cloning and observation, enabling anyone to train or continue training their own aligned models.