---

Quantization	Size	Use Case
Q2_K	~3.8 GB	Extreme compression, lowest quality
Q3_K_S	~4.3 GB	Small footprint
Q3_K_M	~4.6 GB	Small footprint, balanced
Q3_K_L	~4.9 GB	Small footprint, higher quality
Q4_0	~5.3 GB	Good balance
Q4_K_S	~5.4 GB	Good balance
Q4_K_M	~5.7 GB	Recommended for most users
Q5_0	~6.3 GB	High quality
Q5_K_S	~6.3 GB	High quality
Q5_K_M	~6.5 GB	High quality, balanced
Q6_K	~7.4 GB	Near-lossless
Q8_0	~9.5 GB	Highest quality quantization
BF16	~17.9 GB	Full precision

# Install llama.cpp
brew install llama.cpp  # macOS
# or build from source: https://github.com/ggml-org/llama.cpp

# Interactive chat
llama-cli --hf-repo Tesslate/OmniCoder-9B-GGUF --hf-file omnicoder-9b-q4_k_m.gguf -p "Your prompt" -c 8192

# Server mode (OpenAI-compatible API)
llama-server --hf-repo Tesslate/OmniCoder-9B-GGUF --hf-file omnicoder-9b-q4_k_m.gguf -c 8192

---