---

license: apache-2.0

library_name: llama.cpp

pipeline_tag: text-generation

tags:

• 1-bit
• gguf
• llama-cpp
• cuda
• metal
• on-device
• prismml
• bonsai

base_model:

• prism-ml/Bonsai-8B-unpacked

---

<p align="center">

<img src="./assets/bonsai-logo.svg" width="280" alt="Bonsai">

</p>

<p align="center">

<a href="https://prismml.com"><b>Prism ML Website</b></a> &nbsp;|&nbsp;

<a href="https://github.com/PrismML-Eng/Bonsai-demo/blob/main/1-bit-bonsai-8b-whitepaper.pdf"><b>Whitepaper</b></a> &nbsp;|&nbsp;

<a href="https://github.com/PrismML-Eng/Bonsai-demo"><b>Demo &amp; Examples</b></a> &nbsp;|&nbsp;

<a href="https://colab.research.google.com/drive/1EzyAaQ2nwDv_1X0jaC5XiVC3ZREg9bdG?usp=sharing"><b>Colab Notebook</b></a> &nbsp;|&nbsp;

<a href="https://discord.gg/prismml"><b>Discord</b></a>

</p>

Bonsai-8B-GGUF-1bit

End-to-end 1-bit language model for llama.cpp (CUDA, Metal, CPU)

> 14.1x smaller than FP16 | 6.2x faster on RTX 4090 | 4-5x lower energy/token

Highlights

• 1.15 GB parameter memory (down from 16.38 GB FP16) — fits on virtually any device with a GPU
• End-to-end 1-bit weights across embeddings, attention projections, MLP projections, and LM head
• GGUF Q1_0 (g128) format with inline dequantization kernels — no FP16 materialization
• Cross-platform: CUDA (RTX/datacenter), Metal (Mac), Android, CPU
• Competitive benchmarks: 70.5 avg score across 6 categories, matching full-precision 8B models at 1/14th the size
• MLX companion: also available as MLX 1-bit g128 for native Apple Silicon inference

<p align="center">

<img src="./assets/frontier.svg" width="680" alt="Frontier Efficiency">

</p>

Resources

• Google Colab — try Bonsai in your browser, no setup required
• Whitepaper — for more details on Bonsai, check out our whitepaper
• Demo repo — comprehensive examples for serving, benchmarking, and integrating Bonsai
• Discord — join the community for support, discussion, and updates
• 1-bit kernels: llama.cpp fork (CUDA + Metal) · MLX fork (Apple Silicon) · mlx-swift fork (iOS/macOS)
• Locally AI — we have partnered with Locally AI for iPhone support

Model Overview

| Item | Specification |

| :------------- | :--------------------------------------------------------------------- |

| Parameters | 8.19B (~6.95B non-embedding) |

| Architecture | Qwen3-8B dense: GQA (32 query / 8 KV heads), SwiGLU MLP, RoPE, RMSNorm |

| Layers | 36 Transformer decoder blocks |

| Context length | 65,536 tokens |

| Vocab size | 151,936 |

| Weight format | GGUF Q1_0 |

| Deployed size | 1.15 GB (14.2x smaller than FP16) |

| 1-bit coverage | Embeddings, attention projections, MLP projections, LM head |

| License | Apache 2.0 |

Quantization Format: Q1_0

Each weight is a single bit: 0 maps to −scale, 1 maps to +scale. Every group of 128 weights shares one FP16 scale factor.

Effective bits per weight: 1.125 (1 sign bit + 16-bit scale amortized over 128 weights).

Memory Requirement

Parameter memory only (weights and scales loaded into memory):

| Format | Size | Reduction | Ratio |

| :----------------- | ----------: | --------: | --------: |

| FP16 | 16.38 GB | — | 1.0x |

| GGUF Q1_0 | 1.15 GB | 93.0% | 14.2x |

| MLX 1-bit g128 | 1.28 GB | 92.2% | 12.8x |

The GGUF file on disk is 1.16 GB (~6.6 MB larger) because the format embeds the tokenizer, chat template, and model metadata alongside the weights.

Best Practices

Generation Parameters

| Parameter | Default | Suggested range |

| :----------------- | :------ | :-------------- |

| Temperature | 0.5 | 0.5 -- 0.7 |

| Top-k | 20 | 20 -- 40 |

| Top-p | 0.9 | 0.85 -- 0.95 |

| Repetition penalty | 1.0 | |

| Presence penalty | 0.0 | |

System Prompt

You can use a simple system prompt such as:

You are a helpful assistant

Quickstart

llama.cpp (CUDA)

# Clone the PrismML fork of llama.cpp (includes Q1_0 kernels)
git clone https://github.com/PrismML-Eng/llama.cpp
cd llama.cpp

# Build with CUDA support
cmake -B build -DGGML_CUDA=ON && cmake --build build -j

# Run inference
./build/bin/llama-cli \
    -m Bonsai-8B-Q1_0.gguf \
    -p "Explain quantum computing in simple terms." \
    -n 256 \
    --temp 0.5 \
    --top-p 0.85 \
    --top-k 20 \
    -ngl 99

llama.cpp (Metal / macOS)

# Clone the PrismML fork of llama.cpp (includes Q1_0 kernels)
git clone https://github.com/PrismML-Eng/llama.cpp
cd llama.cpp

# Build with Metal support (default on macOS)
cmake -B build && cmake --build build -j

# Run inference
./build/bin/llama-cli \
    -m Bonsai-8B-Q1_0.gguf \
    -p "Explain quantum computing in simple terms." \
    -n 256 \
    --temp 0.5 \
    --top-p 0.85 \
    --top-k 20 \
    -ngl 99

llama.cpp Server

./build/bin/llama-server \
    -m Bonsai-8B-Q1_0.gguf \
    --host 0.0.0.0 \
    --port 8080 \
    -ngl 99

Open the web UI at http://127.0.0.1:8080, or see our llama.cpp fork for more examples.

Cross-Platform Throughput

| Platform | Backend | TG128 (tok/s) | FP16 TG (tok/s) | TG vs FP16 | PP512 (tok/s) | FP16 PP512 (tok/s) |

| :---------------- | :--------------- | ------------: | --------------: | ---------: | ------------: | -----------------: |

| RTX 4090 | llama.cpp CUDA | 368 | 59 | 6.2x | 11,809 | 10,453 |

| RTX L40S | llama.cpp CUDA | 327 | 52 | 6.3x | 9,592 | 8,325 |

| RTX 3060 Laptop | llama.cpp CUDA | 81 | 3.5¹ | 23x¹ | 1,871 | 94¹ |

| M4 Pro 48 GB | llama.cpp Metal | 85 | 16 | 5.4x | 498 | 490 |

| Samsung S25 Ultra | llama.cpp OpenCL | 19.6 | — | — | 30.4 | — |

¹ FP16 only fits partially on GPU's 6 GB VRAM; 1-bit fits entirely in VRAM.

<p align="center">

<img src="./assets/speeds_8B.png" width="680" alt="Cross-platform throughput">

</p>

Energy Efficiency

| Platform | Bonsai E_tg (mWh/tok) | Baseline E_tg | Advantage |

| :----------------- | --------------------: | ------------: | --------: |

| RTX 4090 (CUDA) | 0.276 | 1.134 (FP16) | 4.1x |

| Mac M4 Pro (Metal) | 0.091 | 0.471 (FP16) | 5.1x |

<p align="center">

<img src="./assets/energy_8B.png" width="680" alt="Energy efficiency">

</p>

Benchmarks

Evaluated with EvalScope v1.4.2 + vLLM 0.15.1 on NVIDIA H100 under identical infrastructure, generation parameters, and scoring. All models are in the 6B–9B parameter range.

| Model | Company | Size | Avg | MMLU-R | MuSR | GSM8K | HE+ | IFEval | BFCL |

| :------------------ | :------------ | ----------: | -------: | -----: | ---: | ----: | ---: | -----: | ---: |

| Qwen 3 8B | Alibaba | 16 GB | 79.3 | 83 | 55 | 93 | 82.3 | 84.2 | 81 |

| RNJ 8B | EssentialAI | 16 GB | 73.1 | 75.5 | 50.4 | 93.7 | 84.2 | 73.8 | 61.1 |

| Mistral3 8B | Mistral | 16 GB | 71.0 | 73.9 | 53.8 | 87.2 | 67.4 | 75.4 | 45.4 |

| Olmo 3 7B | Allen Inst | 14 GB | 70.9 | 72 | 56.1 | 92.5 | 79.3 | 37.1 | 38.4 |

| 1-bit Bonsai 8B | PrismML | 1.15 GB | 70.5 | 65.7 | 50 | 88 | 73.8 | 79.8 | 65.7 |

| LFM2 8B | LiquidAI | 16 GB | 69.6 | 72.7 | 49.5 | 90.1 | 81 | 82.2 | 62.0 |

| Llama 3.1 8B | Meta | 16 GB | 67.1 | 72.9 | 51.3 | 87.9 | 75 | 51.5 | — |

| GLM v6 9B | ZhipuAI | 16 GB | 65.7 | 61.9 | 43.2 | 93.4 | 78.7 | 69.3 | 21.9 |

| Hermes 8B | Nous Research | 16 GB | 65.4 | 67.4 | 52.2 | 82.9 | 51.2 | 65 | 73.5 |

| Trinity Nano 6B | Arcee | 12 GB | 61.2 | 68.8 | 52.6 | 81.1 | 54 | 50 | 62.5 |

| Marin 8B | Stanford CRFM | 16 GB | 56.6 | 64.8 | 42.6 | 86.4 | 51 | 50 | — |

| R1-D 7B | DeepSeek | 14 GB | 55.1 | 62.5 | 29.1 | 92.7 | 81.7 | 48.8 | 15.4 |

Despite being 1/14th the size, 1-bit Bonsai 8B is competitive with leading full-precision 8B instruct models.

Intelligence Density

Intelligence density captures the ratio of a model's capability to its deployed size:

alpha = -ln(1 - score/100) / size_GB

| Model | Size | Intelligence Density (1/GB) |

| :------------------ | ----------: | --------------------------: |

| 1-bit Bonsai 8B | 1.15 GB | 1.062 |

| Qwen 3 8B | 16 GB | 0.098 |

| Llama 3.1 8B | 16 GB | 0.074 |

| Mistral3 8B | 16 GB | 0.077 |

Bonsai 8B achieves 10.8x higher intelligence density than full-precision Qwen 3 8B.

<p align="center">

<img src="./assets/intel_density_8B.png" width="680" alt="Intelligence density">

</p>

Use Cases

• On-device assistants: interactive AI on laptops and phones with low latency
• Mobile deployment: runs on a wide variety of phones due to low memory footprint
• Edge robotics and autonomy: compact deployment on devices with thermal, memory, or connectivity constraints
• Cost-sensitive GPU serving: higher throughput and lower energy per token on RTX-class and datacenter GPUs
• Enterprise and private inference: local or controlled-environment inference for data residency requirements

Limitations

• No native 1-bit hardware exists yet — current gains are software-kernel optimizations on general-purpose hardware
• Mobile power measurement is estimated rather than hardware-metered
• The full-precision benchmark frontier continues to advance; the 1-bit methodology is architecture-agnostic and will be applied to newer bases

Citation

If you use 1-bit Bonsai 8B, please cite:

@techreport{bonsai8b,
    title   = {1-bit Bonsai 8B: End-to-End 1-bit Language Model Deployment
               Across Apple, GPU, and Mobile Runtimes},
    author  = {Prism ML},
    year    = {2026},
    month   = {March},
    url     = {https://prismml.com}
}

Contact

For questions, feedback, or collaboration inquiries: contact@prismml.com