By using ~3 bit quantized model with llama.cpp, Unsloth makes good quants: https://docs.unsloth.ai/models/tutorials-how-to-fine-tune-and-run-llms/kimi-k2-how-t

By using ~3 bit quantized model with llama.cpp, Unsloth makes good quants: https://docs.unsloth.ai/models/tutorials-how-to-fine-tune-and-run-llms/kimi-k2-how-to-run-locally#model-uploads https://docs.unsloth.ai/models/tutorials-how-to-fine-tune-an... Note that llama.cpp doesn't try to be production-grade engine, more focused on local usage.