Ternary Bonsai 2 27B — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
PrismML 对 Qwen3.8-27B 的三值重建：语言模型和嵌入共 26.9B 参数，另加 0.46B 视觉塔，64 个块中约四分之一使用全注意力。PTQ1_0 文件为 5.95 GB，PQ2_0 为 7.21 GB，而 FP16 约为 54 GB。PrismML 称其在 14 项思考模式基准中保留了 FP16 模型平均分的 98.2%。标准 llama.cpp 无法运行这些文件：它会拒绝 PTQ1_0 和 PQ2_0，而且据模型卡说明，它会无警告地加载 Q2_0 版本并输出乱码。请使用 PrismML 的 llama.cpp 或 MLX 分支。
Ternary Bonsai 2 27B 在 PTQ1_0 ternary 下约需 7 GB 显存——量化权重加框架开销，不含 KV 缓存。在 Apple Silicon 上，这部分来自统一内存。
按量化级别的显存与速度
计算基准：NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存，因此数值高于上方的主数字。
| 量化 | 显存 | 显存 | 速度（估算） | 适配 |
|---|---|---|---|---|
| PTQ1_0 ternary 1.78 bpw | 7.3 GB | ~95 tok/s | 可运行 | |
| PQ2_0 ternary 2.16 bpw | 8.5 GB | ~82 tok/s | 可运行 |
黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型，详见 方法说明页. Ternary Bonsai 2 27B 显存计算器 →
运行 Ternary Bonsai 2 27B
目录中能运行 Ternary Bonsai 2 27B 的最便宜 GPU 是 Intel Arc B570 (10 GB).
或在 Vast.ai 比较，低至 $0.35/小时 (typical low · varies)
作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。
如何运行 Ternary Bonsai 2 27B
No Ollama library entry, and stock llama.cpp cannot run these files. Use PrismML's llama.cpp fork (CUDA, Metal, CPU) or its MLX fork on Apple Silicon.
规格
Corroborated — Two or more independent sources agree on these figures, but the model card itself was not retrieved. Treat the numbers as good rather than confirmed. 尚未确认： parameter count (model card and GGUF metadata agree on 26.9B for the language model; no independent source).
- 参数量
- 26.9 Billion (+0.46B vision tower)
- 上下文窗口
- 262,144
- 架构
- Qwen3.8-27B hybrid attention (~3 linear : 1 full), ternary weights
- 提供商
- PrismML
- 许可证
- Apache 2.0
- Availability
- Weights downloadable
- Practical deployment
- Laptop
- 规格量化
- PTQ1_0 ternary
- Vision projector
- +0.6 GB when you send images (not included above)
- 系统内存
- 16 GB
- 记录更新于
- 2026-10-15
Commercial use permitted. No usage restrictions beyond attribution.
限制与注意事项
- Ships only in PrismML's ternary packings; there is no Q4_K_M, Q8_0 or FP16 GGUF to choose instead.
- Stock llama.cpp, Ollama and LM Studio cannot run it; the PrismML forks are required.
- The card's speed table was measured on an earlier build and is marked pending re-measurement.
- The vision projector is a separate file (0.63 GB at Q8_0).
另请参阅 Ternary Bonsai 27B
质量与使用场景
评分由模型作者或独立评测方发布——衡量质量而非吞吐量，并非我们实测。
我的 GPU 能运行 Ternary Bonsai 2 27B 吗？
Ternary Bonsai 2 的其他尺寸
Ternary Bonsai 2 27B — 常见问题
How much VRAM does Ternary Bonsai 2 27B need?
About 7 GB at PTQ1_0 ternary — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Ternary Bonsai 2 27B run on an RTX 4090 (24 GB)?
Yes. Ternary Bonsai 2 27B needs about 7 GB at PTQ1_0 ternary, inside a 24 GB card, at an estimated 95 tokens/sec.
How do I run Ternary Bonsai 2 27B locally?
No Ollama library entry, and stock llama.cpp cannot run these files. Use PrismML's llama.cpp fork (CUDA, Metal, CPU) or its MLX fork on Apple Silicon. Running the published tag would send your prompts to a hosted GPU rather than your own machine.
What other sizes does Ternary Bonsai 2 come in?
Ternary Bonsai 2 2B (1 GB), Ternary Bonsai 2 4B (2 GB), Ternary Bonsai 2 9B (3 GB), Ternary Bonsai 2 27B (7 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.