Skip to content
Computer vision

Qwen3.6-35B-A3B

открытые веса для agentic coding

1 min readupdated

В репозитории — веса и конфиги пост-обученной Qwen3.6-35B-A3B в формате Hugging Face Transformers. Это причинно-следственная языковая модель с vision-энкодером: 35B параметров всего, из них 3B активных. Совместима с Transformers, vLLM, SGLang и KTransformers.

Ключевые особенности

  • Agentic Coding: лучше справляется с frontend-задачами и рассуждениями на уровне репозитория.
  • Thinking Preservation: новая опция сохранять reasoning-контекст из истории сообщений.
  • MoE-архитектура: 256 экспертов, 8 routed + 1 shared, активируются 3B параметров.
  • Контекст 262 144 токена нативно, расширяется до 1 010 000 токенов.
  • Vision Encoder: модель работает с изображениями и текстом.

Пример

Сценарий из карточки: попросить модель разобрать frontend-задачу и предложить правки на уровне репозитория, сохраняя reasoning-контекст между итерациями. Например: «Посмотри на репозиторий, найди причину падения фронтенд-тестов и предложи патч».

ПодробнееSWE-bench Verified — 73.4, SWE-bench Multilingual — 67.2, Terminal-Bench 2.0 — 51.5.
Архитектура: 40 слоёв, 256 экспертов MoE, 8 routed + 1 shared, MTP с multi-steps.
Обучение: pre-training и post-training; нативный контекст 262 144 токена, до 1 010 000.
Требования к VRAM в карточке не указаны.
Discuss Telegram 0 29

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • F8_E4M3recommendedweights 35 GB · ready files39 GB
  • 4-bitweights 18 GB20 GB

Tip: go 8-bit.F8_E4M3 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: Won't fit any home GPU — renting is cheaper.

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.

    # savethisok/qwen36-fp8row-995ad96
    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 savethisok/qwen36-fp8row-995ad96 --local-dir ~/models/qwen36-fp8row-995ad96
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to qwen36-fp8row-995ad96
  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 qwen36-fp8row-995ad96 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

savethisok/qwen36-fp8row-995ad96
Author
savethisok
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

290

Ask AI about this model

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

Comments