Spiritfather released a quantized model Boulesis-26B-A4B

Developer spiritfather has presented Boulesis-26B-A4B-i1-GGUF — a new version of the language model optimized for local execution. The release is made in the GGUF format, which ensures compatibility with popular execution environments for personal computers. The model is based on an architecture with 26 billion parameters and uses the A4B quantization method. The use of the i1-GGUF format makes it possible to significantly reduce the requirements for video memory and RAM while maintaining the accuracy of text generation. In preliminary tests, the model received a rating of 11 stars, demonstrating stable operation when performing generation tasks for 1 hour of load. The tool is fully compatible with current libraries and environments for local inference, including llama.cpp, vLLM, and Ollama. This allows users to run the model on consumer equipment without the need to use cloud APIs or specialized server clusters. This release expands the availability of heavy language models for enthusiasts and developers working with local AI solutions. The GGUF format remains the standard for decentralized launch of neural networks, allowing flexible adjustment of quantization parameters for specific user hardware configurations.