[!IMPORTANT] Compatibility: These GGUF files require a llama.cpp build with K2 Horizon architecture support. Until upstream support lands, use the MBZUAI-IFM fork.

K2-Horizon-3.7B GGUF

GGUF quantizations of IFM/K2-Horizon-3.7B, a 3.7B dense decoder-only model for reasoning, coding, long-context work, and tool use. The source checkpoint supports a native context length of 524,288 tokens (512K).

Benchmarks

Benchmark results reported by IFM for the original IFM/K2-Horizon-3.7B checkpoint.

GGUF files

Quantization	File	Size
Q4_0	K2-Horizon-3.7B-Q4_0.gguf	3.02 GB
Q4_K_M	K2-Horizon-3.7B-Q4_K_M.gguf	3.16 GB
Q5_K_M	K2-Horizon-3.7B-Q5_K_M.gguf	3.64 GB
Q6_K	K2-Horizon-3.7B-Q6_K.gguf	4.16 GB
Q8_0	K2-Horizon-3.7B-Q8_0.gguf	5.39 GB

Chat template

Each GGUF embeds the llama.cpp-compatible chat template. chat_template.jinja is a matching external copy for tools that require one. The original source template is retained as chat_template.upstream.jinja for runtimes with full Jinja support.

xml is the default tool-call format. Use --chat-template-kwargs to select json or xml_typed when required.

Usage

Use the IFM K2 Horizon llama.cpp fork.

llama-cli \
  -m K2-Horizon-3.7B-Q4_K_M.gguf \
  -c 8192 -n 512 --jinja \
  -p "Explain why reproducible builds matter."

Source