cd /news/large-language-models/model-ternary-bonsai-2-27b · home › topics › large-language-models › article
[ARTICLE · art-172522] src=quantized.uk ↗ pub= topic=large-language-models verified=true sentiment=· neutral

Model: Ternary Bonsai 2 27B

Prism ML has published Ternary Bonsai 2 27B, a ternary-quantized derivative of Qwen3.8 27B distributed as a 7.63 GB GGUF Q2_0 file (7,626,008,928 bytes) that the publisher says requires llama.cpp build b11556 or later because of its Hadamard-rotated weights. The 64-layer decoder keeps 16 growing-cache layers, supports a 262K maximum context, and adds a separate 0.63 GB Q8_0 vision projector for image input; the specialized PQ2_0 and PTQ1_0 packages use different runtime kernels. No speed or perplexity loss is recorded for this file, and the site's reviewed link index has no community run for the model yet.

key takeaway

Ternary Bonsai 2 27B ships as a 7.63 GB GGUF Q2_0 file with a 262K context and needs llama.cpp b11556 or later for its Hadamard-rotated weights, with image input requiring a separate 0.63 GB Q8_0 projector.

read7 min views5 publishedNov 19, 2026
Model: Ternary Bonsai 2 27B
Image: Quantized (auto-discovered)

27B Prism ML Ternary Bonsai

Prism ML’s ternary Qwen3.8 27B derivative, indexed here in the official Q2_0 packing. Its Hadamard-rotated weights require llama.cpp b11556 or later according to the publisher. The 64-layer decoder retains 16 growing-cache layers. Image use adds the matching Q8_0 projector. The specialized PQ2_0 and PTQ1_0 packages use different runtime kernels. No speed or perplexity loss is recorded for this file.

262K

Max Context

1

Quant Variants

GGUF Q2_0

Best Quality

—

Perplexity rise (best build)

Quantization Variants #

Per-quant VRAM, quality loss, and inference speed on RTX 4090

Measured = run on this site · Estimated = not run here: calculated, or a figure this site has not verified · Community = public reports

Which file each size is for

A file size is not a memory figure: the VRAM column is the calculator's total at 4K context — weights plus KV cache plus a 10% activation buffer. Units differ on purpose: file sizes are decimal GB, as Hugging Face lists them; the VRAM column is GiB (1024³ bytes) with 2% added to the weights, so 7.12 GB of file counts as about 6.8 GiB there. A vision projector (mmproj), where the repo ships one, is a separate download and is in neither number.

  • GGUF Q2_0prism-ml/Ternary-Bonsai-2-27B-gguf · Ternary-Bonsai-2-27B-Q2_0.gguf · 7.63 GB (7,626,008,928 bytes) · + vision projector 0.63 GB, separate · listing read 2026-11-19

Related community runs

These runs describe their own settings. Their speeds do not replace this estimate or prove your configuration fits.

Our reviewed link index has no run for this model yet. You can browse community reports or share your own test below.

Browse more runs on llamaPerf ↗

Similar models #

[Compare with Qwen3-VL 30B-A3B](https://quantized.uk/tools/compare/?a=ternary-bonsai-2-27b&b=qwen3-vl-30b-a3b)

[30B-A3BConsumer GPUPro GPUQ4_K_M19.9 GB+2.6%perplexity rise](https://quantized.uk/quant-hub/qwen3-vl-30b-a3b/)

### Qwen3-VL 30B-A3B Instruct

Alibaba Qwen3-VL

Multimodal MoE with only ~3B active parameters, so it stays responsive on Apple unified memory and survives CPU offload far better than a dense 30B. At Q4_K_M it needs about 19.9 GB at 4K, comfortable on a 24 GB card; Q3_K_M brings that to 15.1 GB. The vision encoder is a separate 1.1 GB file.

24BConsumer GPUMac / Apple SiliconQ4_K_M15.7 GB+2.9%perplexity rise

Magistral Small 1.2 24B

Mistral Magistral

Mistral's reasoning model on a Mistral Small 3.2 base, with reasoning wrapped in [THINK] tags. At Q4_K_M it needs about 15.7 GB at 4K, tight on a 16 GB card and comfortable on 20 GB — explicit chain-of-thought a level below the 70B-class reasoners most guides assume.

26B-A4BConsumer GPUPro GPUQ4_K_M18.2 GB—perplexity rise

Gemma 4 26B-A4B IT

Google Gemma 4

Google's Gemma 4 mixture-of-experts model: about 25B parameters in the text model, roughly 4B active per token, a 256K window and image input. Only 5 of its 30 layers keep a cache that grows with context, and those use fewer, wider heads, so the cache stays small — about 0.9 GB at 32K and 5.3 GB at the full 256K. At Q4_K_M that is about 18.8 GB at 32K: comfortable on a 24 GB card, which can still load the whole window (tight, about 23.6 GB). A 16 GB card cannot hold it, since the Q4_K_M file alone is 17.0 GB. The vision encoder is a separate 1.2 GB file on top. Sizes come from bartowski's GGUF files, whose Q4_K_M works out at 5.40 bits per weight, well above the generic 4-bit rate; Unsloth's UD-Q4_K_M is a different file, 0.09 GB smaller. No per-level quality loss has been published.

31BConsumer GPUPro GPUQ4_K_M22.1 GB—perplexity rise

Gemma 4 31B IT

Google Gemma 4

Google's dense Gemma 4 flagship: about 31B parameters, a 256K window and image input. Only 10 of its 60 layers cache the full context, and those use four wide KV heads, so 128K of context adds about 11 GB of cache where a conventional 60-layer design would need about 120 GB. At Q4_K_M it needs about 22.1 GB at 4K, tight on a 24 GB card; Q3_K_M is about 18.3 GB at 4K. A 32 GB card holds Q4_K_M at 32K comfortably. Google also publishes its own quantization-aware (QAT) Q4_0 build, a single 17.7 GB file: about 20.1 GB at 4K, comfortable on 24 GB up to 16K. Its 1.2 GB vision encoder is a separate file, and with it loaded even 4K is tight on 24 GB. The other levels are sized from bartowski's GGUF files, and no per-level quality loss has been published. That repository also carries two MTP draft-head files (349 MB at Q4_0, 515 MB at Q8_0) for speculative decoding; they are not the model itself and are not counted in these sizes.

Running Ternary Bonsai 2 27B locally #

At Q2_0 and 4K of context, Ternary Bonsai 2 27B needs about 8.2 GB — 7.2 GB of weights, 0.25 GB of KV cache and a 0.8 GB activation buffer. The smallest cards in this index that clear that comfortably are 10 GB ones — 2 of them, from the Arc B570 10G up — and 67 of the 78 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.

What longer context costs #

Going from 4K to 32K adds about 1.75 GB, taking the total to 10.2 GB. That is gentler than the parameter count suggests: only 16 of 64 layers keep a KV cache that grows with context, so the cache scales at a fraction of the usual rate. The model's native window is 256K; holding all of it at Q2_0 would need about 26 GB.

Which build to download #

This index only tracks Ternary Bonsai 2 27B in GGUF, across 1 level (Q2_0). No per-level perplexity sweep has been published for this model, so the quality column is empty rather than estimated — within one model, more bits per weight is the only ordering the data supports. No throughput has been measured for this model on this site's hardware. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 — CUDA only, and now archived, so local use rather than serving.

Common questions #

  • How much VRAM does Ternary Bonsai 2 27B need?
  • About 8.2 GB at Q2_0 with 4K of context and batch 1, rising to roughly 10.2 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.
  • Will Ternary Bonsai 2 27B run on a 10GB GPU?
  • Yes — at Q2_0 and 4K context it needs about 8.2 GB, which leaves 1.8 GB spare on a Arc B570 10G, or on any of the other 10 GB cards here (2 in all). That is the smallest size in this index that clears it comfortably; 67 of 78 do.
  • Which quantization of Ternary Bonsai 2 27B should I use?
  • Q2_0 is the only level in the index for this model, so there is nothing to choose between here.

Where this model fits #

Sized at Q2_0 with a 4K context window, smallest card first. Comfortable means the estimate uses at most 88% of the memory.

How to actually run this #

Deployment guides for this model and this class of hardware.

This page's figures change when the model or the runtime does.Last updated 2026-11-19 RSS → /feed.xml

── more in #large-language-models 4 stories · sorted by recency
── more on @prism ml 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/model-ternary-bonsai…] indexed:0 read:7min 2026-11-19 · —