Prism ML
Ternary Bonsai 27B (27B parameters) requires approximately 9.7 GB of VRAM with Q2_0_G128 quantization. For the best balance of quality and speed, we recommend hardware with at least 12 GB of VRAM.
Get started
— copy & paste to run locallyCopy-paste commands to run Ternary Bonsai 27B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "prism-ml/Ternary-Bonsai-27B-gguf" \
--hf-file "Ternary-Bonsai-27B-gguf-Q2_0_G128.gguf" \
-c 4096 -ngl 99Quick specs
About this model
Related models
Inference speed
Estimated decode speed (tokens/sec) for Ternary Bonsai 27B at Q2_0_G128 across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 4080 Super 16GB at ~87 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.
| GPU / Mac | Memory | Quant | Speed (tok/s) | Fits? |
|---|---|---|---|---|
 RTX 4080 Super 16GB | 16 GB | Q2_0_G128 | 87.2 | Fits |
| RTX 5090 32GB | 32 GB | Q2_0_G128 | 76.2 | Fits |
RX 7900 XTX 24GB | 24 GB | Q2_0_G128 | 64.6 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q2_0_G128 | 60.3 | Fits |
MacBook Pro M4 Max 128GB | 128 GB | Q2_0_G128 | 59.6 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q2_0_G128 | 59.6 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q2_0_G128 | 50.2 | Fits |
Mac Studio M1 Ultra 128GB | 128 GB | Q2_0_G128 | 47.6 | Fits |
| RTX 4090 24GB | 24 GB | Q2_0_G128 | 43.8 | Fits |
| RTX 3090 24GB | 24 GB | Q2_0_G128 | 42.6 | Fits |
MacBook Pro M4 Pro 48GB | 48 GB | Q2_0_G128 | 37.5 | Fits |
| RTX 4070 12GB | 12 GB | Q2_0_G128 | 26.0 | Tight |
MacBook Pro M3 Max 64GB | 64 GB | Q2_0_G128 | 26.0 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q2_0_G128 | 23.8 | Fits |
| RTX 3060 12GB | 12 GB | Q2_0_G128 | 17.5 | Tight |
| RTX 4060 8GB | 8 GB | Q2_0_G128 | 6.6 | Too big |
Estimates for single-stream decoding at Q2_0_G128; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.
Quick picks
Best hardware
Run this model
Quantization
How much VRAM Ternary Bonsai 27B (27B) needs at each GGUF quant, and whether it fits a 24 GB card (RTX 4090 / 3090). The recommended Q2_0_G128 uses ~7.2 GB — about 75% less VRAM than Q8_0, at a small quality cost.
| Quant | Bits | VRAM (weights) | Quality | Fits 24 GB? |
|---|---|---|---|---|
| Q1_0_G128 | 1.125 | 3.9 GB | Very Low | Fits |
| Q2_0_G128recommended | 1.71 | 7.2 GB | Low | Fits |
| Q2_K | 2 | 10.5 GB | Low | Fits |
| Q3_K_S | 3 | 13.2 GB | Low | Fits |
| NVFP4 | 4 | 15.1 GB | Medium | Fits |
| Q4_K_M | 4 | 16.5 GB | Medium | Tight |
| Q5_K_M | 5 | 19.4 GB | High | Offloads |
| Q6_K | 6 | 22.1 GB | High | Heavy offload |
| Q8_0 | 8 | 28.9 GB | Very High | Too big |
| F16 | 16 | 55.4 GB | Maximum | Too big |
VRAM shown is quantized weights only; add ~1–3 GB runtime overhead plus KV cache for your context length. Lower quants trade quality for memory — Q4_K_M is the usual sweet spot; Q2/Q3 only when you must fit a bigger model.
Hardware compatibility
Computing compatibility...
Memory breakdown
Frequently asked questions
Ternary Bonsai 27B (27B parameters) requires approximately 9.7 GB of VRAM with Q2_0_G128 quantization. Lower quantizations like Q4_K_M use less memory but may reduce quality.
Yes, Intel Arc B580 12GB can run Ternary Bonsai 27B with a compatibility score of 87/100. It provides 12 GB of memory and achieves approximately 23.7 tokens per second.
The recommended quantization for Ternary Bonsai 27B is Q2_0_G128, which offers the best balance between model quality and memory efficiency. Higher quantizations preserve more quality but require more VRAM.
The top recommended hardware for Ternary Bonsai 27B: RTX 5080 16GB (score: 94/100), RTX 4080 Super 16GB (score: 94/100), RTX 4070 Ti Super 16GB (score: 94/100). These provide the best combination of memory, bandwidth, and compute for running this model locally.
Yes, Ternary Bonsai 27B is well-suited for chat as well as reasoning, coding, edge, vision. It was designed with these use cases in mind.
See also
