The GGUF file on disk is 0.57 GB (~6.4 MB larger) because the format embeds the tokenizer, chat template, and model metadata alongside the weights.
1# Clone the PrismML fork of llama.cpp (includes Q1_0 kernels)
2git clone https://github.com/PrismML-Eng/llama.cpp
3cd llama.cpp
4
5# Build with CUDA support
6cmake -B build -DGGML_CUDA=ON && cmake --build build -j
7
8# Run inference
9./build/bin/llama-cli \
10 -m Bonsai-4B-Q1_0.gguf \
11 -p "Explain quantum computing in simple terms." \
12 -n 256 \
13 --temp 0.5 \
14 --top-p 0.85 \
15 --top-k 20 \
16 -ngl 99
1# Clone the PrismML fork of llama.cpp (includes Q1_0 kernels)
2git clone https://github.com/PrismML-Eng/llama.cpp
3cd llama.cpp
4
5# Build with Metal support (default on macOS)
6cmake -B build && cmake --build build -j
7
8# Run inference
9./build/bin/llama-cli \
10 -m Bonsai-4B-Q1_0.gguf \
11 -p "Explain quantum computing in simple terms." \
12 -n 256 \
13 --temp 0.5 \
14 --top-p 0.85 \
15 --top-k 20 \
16 -ngl 99
1./build/bin/llama-server \
2 -m Bonsai-4B-Q1_0.gguf \
3 --host 0.0.0.0 \
4 --port 8080 \
5 -ngl 99
1@techreport{bonsai,
2 title = {Bonsai: End-to-End 1-bit Language Model Deployment
3 Across Apple, GPU, and Mobile Runtimes},
4 author = {Prism ML},
5 year = {2026},
6 month = {March},
7 url = {https://prismml.com}
8}
For questions, feedback, or collaboration inquiries:
contact@prismml.com
