K2-Horizon-0.9B-System1 (GGUF)
These are GGUF builds of K2-Horizon-0.9B-System1. That model is a no-thinking ("System 1") fine-tune of IFM/K2-Horizon-0.9B: it answers right away instead of writing a long think block first. Results, method and limitations are on the main card.
| File | Size |
|---|---|
K2-Horizon-0.9B-System1-Q8_0.gguf |
1.15 GB |
K2-Horizon-0.9B-System1-Q4_K_M.gguf |
0.67 GB |
- Format: llama architecture with the
k2-horizonpre-tokenizer and YaRN RoPE (factor 16, beta 128/4, 8K original context). - Tokens: end-of-turn is set to
<|ifm|im_end|>. - Chat template: embedded, defaulting to no-think mode.
- BOS: the template adds BOS itself, so
add_bos_tokenis false. With raw prompts, start with<|begin_of_text|>.
llama-cli -hf Lucasgitguy-co/K2-Horizon-0.9B-System1-GGUF:Q8_0 -p "What is 17*23?" -cnv
llama-server -hf Lucasgitguy-co/K2-Horizon-0.9B-System1-GGUF:Q4_K_M
Checked: with llama.cpp, greedy answers to 5 test prompts match the bf16 model, and generation stops at end-of-turn. Converted with llama.cpp convert_hf_to_gguf.py (Nov 2026). Apache-2.0.