Skip to content
Other

Ternary-Bonsai-2-9B

тернарная 9B VLM в GGUF для Colab

1 min readupdated

Это GGUF-сборка Ternary-Bonsai-2-9B — 9B vision-language модели, производной от Qwen3.5-9B с неизменной архитектурой. Веса хранятся как тернарные {-1, 0, +1}, поэтому основной вариант PQ2_0 занимает ~2,42 ГБ вместо ~17,9 ГБ в FP16. В репозитории есть готовый бинарник llama.cpp (PrismML fork) и проектор для изображений, чтобы запустить всё на Colab GPU.

Ключевые особенности

  • Поддержка vision: отдельный mmproj, рекомендуется Q8_0 (~624 МБ).
  • Контекст 262 144 токена — гибридное внимание, 24 linear + 8 full attention слоёв.
  • Варианты GGUF: PQ2_0 (2,42 ГБ), PTQ1_0 (2,00 ГБ), Q2_0 (2,56 ГБ), MTP и DFlash2 drafters.
  • Спекулятивное декодирование: MTP-голова и DFlash2; DFlash2 полезен на copy-heavy задачах, но на общем чате медленнее обычного декодирования.
  • Запуск на T4: пакет и модель скачиваются менее чем за минуту по карточке.

Пример

from huggingface_hub import hf_hub_download
repo = 'IamPradeep/Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU'
model = hf_hub_download(repo, 'Ternary-Bonsai-2-9B-PQ2_0.gguf')
mmproj = hf_hub_download(repo, 'Ternary-Bonsai-2-9B-mmproj-Q8_0.gguf')
# затем: ./llama_bin/llama-cli -m <model> --mmproj <mmproj> --image img.jpg -ngl 99 -fa on
ПодробнееБенчмарков в карточке нет. Для vision нужен Q8_0 mmproj; BF16 может падать в llama-server при параллельных запросах. PQ2_0, PTQ1_0, MTP и DFlash2 требуют PrismML fork; upstream llama.cpp b11556+ читает только Q2_0. Thinking mode лучше давать ≥8192 выходных токенов.
Discuss Telegram 0 103

Will it run?

Runs in GGUF Q2_0

The most precise of the ready quantised files fits.

  • GGUF Q2_0weights 7.1 GB · ready files8.3 GB
  • GGUF TQ1_0weights 1.9 GB · ready files2.5 GB
  • GGUF Q8_0recommendedweights 1.5 GB · ready files2.1 GB

Tip: go 8-bit.GGUF Q8_0 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q8_0 — from RTX 2060 · GGUF Q2_0 — from RTX 3080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # IamPradeep/Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download IamPradeep/Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU --local-dir ~/models/Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

IamPradeep/Ternary-Bonsai-2-9B-GGUF-Colab-Prebuilt-GPU
Author
IamPradeep
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1030

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments