Bonsai 27B — PrismML 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
PREVIEW (July 2026). The first 27B-class model that runs on a phone. Bonsai 27B is PrismML's extreme low-bit rebuild of Qwen3.6-27B: the architecture is unchanged, but the weights are squeezed to roughly 1–1.7 bits each. The 1-bit build is just 3.9 GB — small enough for an iPhone — while the ternary build is 5.9 GB and aimed at laptops. Both are multimodal (text + images), carry a 262K-token context, and ship under Apache 2.0. PrismML reports the ternary build keeps more than 95% of the full-precision (FP16) baseline across a 15-benchmark suite and the 1-bit build more than 90%, running about 11 tok/s on an iPhone 17 Pro. Specs are from launch coverage — verify against the Hugging Face model cards.
变体
Bonsai 27B 最小的变体在 1-bit (binary, ~1.125 bpw) 下约需 5 GB 显存——量化权重加框架开销，不含 KV 缓存。
| 模型 | Q4 下显存 | 显存 | 上下文 | 运行 |
|---|---|---|---|---|
| 1-bit Bonsai 27B → 27B | ~4.6 GB | 262,144 | ollama run bonsai-27b | |
| Ternary Bonsai 27B → 27B | ~6.6 GB | 262,144 | ollama run bonsai-27b |
显存为 Q4_K_M 下的量化权重加开销，与 GPU 与显存检测器使用同一引擎计算。
如何在本地运行 Bonsai 27B
安装 Ollama，然后拉取标签。
ollama run bonsai-27b在上方选择一个尺寸，查看它自己的显存、速度估算和安装命令。
许可证
Commercial use permitted. No usage restrictions beyond attribution.
适用于: 1-bit Bonsai 27B, Ternary Bonsai 27B推荐 GPU
目录中能在本地运行 Bonsai 27B（至少 5 GB 显存）的最便宜 GPU 是 Intel Arc B570 (10 GB).
我的 GPU 能运行 Bonsai 27B 吗？
- Bonsai 27B 在 AMD Radeon RX 9060 XT 8GB 上
- Bonsai 27B 在 Intel Arc B570 上
- Bonsai 27B 在 Intel Arc B580 上
- Bonsai 27B 在 NVIDIA GeForce RTX 3060 (12GB) 上
- Bonsai 27B 在 NVIDIA GeForce RTX 3070 上
- Bonsai 27B 在 NVIDIA GeForce RTX 3070 Ti 上
- Bonsai 27B 在 NVIDIA GeForce RTX 3080 (10GB) 上
- Bonsai 27B 在 NVIDIA GeForce RTX 4060 上
- Bonsai 27B 在 NVIDIA GeForce RTX 4070 上
- Bonsai 27B 在 NVIDIA GeForce RTX 4070 Super 上
- Bonsai 27B 在 NVIDIA GeForce RTX 4070 Ti 上
- Bonsai 27B 在 NVIDIA GeForce RTX 5060 上
- Bonsai 27B 在 NVIDIA GeForce RTX 5060 Ti 8GB 上
- Bonsai 27B 在 NVIDIA GeForce RTX 5070 上
Bonsai 27B — 常见问题
How much VRAM does Bonsai 27B need?
Bonsai 27B needs about 5 GB VRAM at 1-bit (binary, ~1.125 bpw) quantization for its smallest variant. Variants: 1-bit Bonsai 27B (5 GB, 1-bit (binary, ~1.125 bpw)); Ternary Bonsai 27B (7 GB, Ternary (1.58-bit, ~1.71 bpw)). On Apple Silicon, unified memory counts toward this requirement.
Can I run Bonsai 27B on an RTX 4090 (24 GB)?
Yes — Bonsai 27B runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at 1-bit (binary, ~1.125 bpw).
What quantization should I use for Bonsai 27B?
Bonsai 27B is published only pre-quantised, so there is no Q4_K_M or Q8_0 build to choose between: each variant ships in the format listed beside it above (1-bit (binary, ~1.125 bpw)). Use the vendor's runtime for that format, as the variant pages describe.
How do I run Bonsai 27B with Ollama?
Bonsai 27B has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.