Skip to content
Other

dspark_grug

драфтер для Grug-67B

1 min readupdated

Это DSpark-драфтер для спекулятивного декодирования целевой open-athena/Grug-67B-A2B-Datakit-SFT-262K-2026.09.21-dr-doom от atutej. Архитектура: 6 слоёв Grug, пять tapped-слоёв, block size 8 и уменьшенный draft-словарь на 32 000 токенов. 669.1M обучаемых параметров, 1,161.3M всего с замороженными эмбеддингами и головами.

Ключевые особенности

  • DSpark: block size 8, 6 Grug-слоёв, 5 tapped-слоёв целевой модели.
  • 669.1M trainable-параметров; 1,161.3M всего.
  • Обучен на собственных ответах Datakit SFT mixture без регенерации; 75 834 шага, около 0.20 эпохи.
  • Нужен Marin vLLM fork с поддержкой method: dspark; для клиента спекулятивное декодирование прозрачно.
  • Чекпоинт не annealed: learning rate ещё 0.81 от пика, это снапшот продолжающегося запуска.

Пример

vllm serve open-athena/Grug-67B-A2B-Datakit-SFT-262K-2026.09.21-dr-doom \
  --speculative-config '{"model": "atutej/dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108", "num_speculative_tokens": 8, "method": "dspark"}'
ПодробнееНа шаге 75 834: speedup 2.04× на Datakit held-out и 2.64× на MATH-500; acceptance length 3.289 и 4.122.
669.1M обучаемых параметров, 1,161.3M всего; обучение — 0.20 эпохи, LR ещё 0.81 от пика.
Для запуска нужен Marin vLLM fork с поддержкой method: dspark; чекпоинт — снапшот незавершённого прогона.
Discuss Telegram 0 18

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.4 GB · ready files2.0 GB
  • FP8 / INT8recommendedweights 0.7 GB1.3 GB
  • 4-bitweights 0.4 GB0.9 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 2060

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.

    # atutej/dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108
    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 atutej/dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108 --local-dir ~/models/dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108
    # 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

  1. Grug-67B-A2B-Datakit-SFT-262K-2026.09.21family root
  2. Grug-67B-A2B-Datakit-SFT-262K-2026.09.21-dr-doom
  3. dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108

New posts about versions of Grug-67B-A2B-Datakit-SFT-262K-2026.09.21 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 6All on Hugging Face

About the model

atutej/dspark_grug-layers6-taps5-grug-67b-a2b-datakit-sft-dr-doom_datakit-sft-mix_itw-20261108
Author
atutej
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments