File	Quant	Size	Notes
`LFM-2.5-1.2B-Novel-JP-F16.gguf`	F16	~2.3 GB	Full precision, highest quality
`LFM-2.5-1.2B-Novel-JP-Q8_0.gguf`	Q8_0	~1.2 GB	Near-lossless
`LFM-2.5-1.2B-Novel-JP-Q5_K_M.gguf`	Q5_K_M	~843 MB	Balanced quality/size
`LFM-2.5-1.2B-Novel-JP-Q4_K_M.gguf`	Q4_K_M	~731 MB	Smallest, recommended for most uses

Usage (llama.cpp)

Base model: LiquidAI/LFM2.5-1.2B-JP-202606
Architecture: LFM2 ( Lfm2ForCausalLM )
Format: GGUF (llama.cpp compatible)

bash

huggingface-cli download TeamDelta/LFM-2.5-1.2B-Novel-JP-GGUF \
  LFM-2.5-1.2B-Novel-JP-Q4_K_M.gguf --local-dir .

llama-cli -m LFM-2.5-1.2B-Novel-JP-Q4_K_M.gguf -p "こんにちは" -cnv

Usage (Ollama)

ollama run hf.co/TeamDelta/LFM-2.5-1.2B-Novel-JP-GGUF:Q4_K_M

Converted and uploaded with llama.cpp inside the Unsloth Docker image.