GGUF Quantizations Released for IFM's K2-Horizon AI Model Series

Quantized GGUF weights for IFM's K2-Horizon model fleet—ranging from 0.9B to 36B parameters—have been published on Hugging Face. Local inference users can now access GGUF variants for the 0.9B, 3.7B, 7B, 32B, and MoVA-36B model sizes. Releasing GGUF quantizations makes IFM's new open-weights model family accessible on standard consumer GPUs and CPUs by significantly lowering RAM and VRAM requirements. This allows open-source AI enthusiasts and developers to run and benchmark these models locally without specialized enterprise hardware. Because official integration into the main llama.cpp repository is still pending approval via an active pull request, users must run these GGUF weights using a specific custom fork of llama.cpp. The uploaded models include standard dense architectures alongside the MoVA-36B Mixture-of-Experts model.