Skip to content
Other

Zils Mini Body 3

4,5B decision-тело для subnet 127

2 min read

zils-mini-body-3 от tensor-tailor — decision body для Zils d3-mini soup pool (Bittensor subnet 127). Это модель на 4,54B в шасси d3-mini на базе Qwen3.5-4B: языковые и визуальные тензоры в bf16, равномерное среднее двух d3-mini house body плюс merged-LoRA файнтюн.

Ключевые особенности

  • Не прошла собственный pre-screen: soup gain +0,63 на mini_combo50 и +0,86 на mini_dev_gold при пороге 0,8 на обоих наборах.
  • На своих dev-наборах уступает предыдущему zils-mini-body-2: 72,80 против 75,07 на mini_combo50 и 77,91 против 79,58 на mini_dev_gold.
  • LoRA rank 16, alpha 32, dropout 0,05, lr 1e-4; обучены проекции attention и Gated-DeltaNet во всех 32 языковых слоях — 152 матрицы, затем merged в bf16.
  • 34 687 входных строк: aimed и natural public data, строки из предыдущих body и regularisation; строки длиннее 2 048 токенов пропускались.
  • Официальные проверки Zils перед коммитом прошли: commit check, copy check и twin check; rate 1.0, base residual 0,81.

Пример

Сценарий из карточки: тело выставляется в раунде d3-mini как seat attempt и оценивается на собственных dev-наборах. Для этого чекпойнта mini_combo50 даёт 72,80, а mini_dev_gold — 77,91.

ПодробнееНа mini_combo50 — 72,80, на mini_dev_gold — 77,91; стартовое среднее {torchcast, d3-mini} — 71,83 и 76,02.
У предыдущего zils-mini-body-2 — 75,07 и 79,58.
Обучение: LoRA rank 16, 34 687 строк, чекпойнт 0,75 эпохи; строки длиннее 2 048 токенов пропускались.
Ограничение: модель не прошла pre-screen, это seat attempt; официальный раунд — единственный тест.
Discuss Telegram 0 15

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 GB

Tip: go 8-bit.FP8 / INT8 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: FP8 / INT8 — from RTX 2060 · BF16 — 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.

    # tensor-tailor/zils-mini-body-3
    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 tensor-tailor/zils-mini-body-3 --local-dir ~/models/zils-mini-body-3
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to zils-mini-body-3
  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

  1. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. torchcast-decision-4b
  4. zils-mini-body-3

New posts about versions of Qwen3.5-4B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 1All on Hugging Face

About the model

tensor-tailor/zils-mini-body-3
Author
tensor-tailor
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

150

Ask AI about this model

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

Comments