Skip to content
Other

Qwen3-8B-speculator.dspark-swa

опорный драфтер для Qwen3-8B

1 min read

Infomaniak-AI выложили Qwen3-8B-speculator.dspark-swa — baseline-драфтер для спекулятивного декодирования Qwen/Qwen3-8B. Это 5-слойная модель на стандартном DSpark Sliding-Window Attention с окном 128 и блоком до 8 токенов. Активных параметров — 1,072.20M; модель обучалась как часть контролируемой ablation-серии Infomaniak.

Ключевые особенности

  • Драфтер для спекулятивного декодирования Qwen3-8B: до 8 токенов за проход.
  • Архитектура KvNDSparkDraftModel: 5 слоёв sliding-window attention, окно 128.
  • Markov-голова gated low-rank с rank 256 и draft vocab 32,000.
  • Использует скрытые состояния Qwen3-8B из слоёв 2, 18, 33.
  • Кэш 10.49 MB на последовательность, активных параметров 1,072.20M.
  • Baseline в ablation-семействе Infomaniak: опорная точка для сравнения с KvN-RFF-вариантами.

Пример

import torch
from speculators.models.kvn_dspark import KvNDSparkDraftModel
model = KvNDSparkDraftModel.from_pretrained(
    "Infomaniak-AI/Qwen3-8B-speculator.dspark-swa",
    torch_dtype=torch.bfloat16,
    device_map="cuda:0",
)
model.eval()
ПодробнееВалидация: loss 0.7900, accept length 1.9020, greedy EAL 1.9780; Pos-0 — 51.8%.
Кэш 10.49 MB на последовательность, активных параметров 1,072.20M.
Задержка драфта при B=128 — 12.223 ms; в таблице ускорения это опорная точка 1.00×.
Нужна ветка speculators kvn-rff-rnn-mtp и метод kvn_dspark в vLLM; обучение — 17m 48s на NVIDIA H200.
Discuss Telegram 0 25

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # Infomaniak-AI/Qwen3-8B-speculator.dspark-swa
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Infomaniak-AI/Qwen3-8B-speculator.dspark-swa --local-dir ~/models/Qwen3-8B-speculator.dspark-swa
    # 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 Qwen3-8B-speculator.dspark-swa (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Infomaniak-AI/Qwen3-8B-speculator.dspark-swa
Author
Infomaniak-AI
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

250

Ask AI about this model

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

Comments