Skip to content
Computer vision

Qwen3.5-9B-Base CPT-0.4coding

9B-база для дообучения

1 min read

В репозитории — CPT-вариант Qwen3.5-9B-Base, предобученной причинной LLM с визуальным энкодером от Qwen Team. Карточка описывает именно базовую Qwen3.5-9B-Base, а суффикс CPT-0.4coding в названии не раскрывает деталей дообучения. Модель предназначена для файнтюна, in-context learning и исследований, а не для прямого диалога.

Ключевые особенности

  • Единая vision-language база: early fusion на мультимодальных токенах; по заявлению авторов, превосходит Qwen3-VL в reasoning, коде, агентах и визуальном понимании.
  • Гибридная архитектура: Gated DeltaNet и sparse Mixture-of-Experts для высокой пропускной способности и низкой задержки.
  • Контекст 262 144 токена нативно, расширяется до 1 010 000.
  • Поддержка 201 языка и диалекта.
  • 9B параметров, 32 слоя, hidden size 4096, FFN 12 288.
  • Control tokens <|im_start|> и <|im_end|> обучены для LoRA-style PEFT с официальным chat template.

Пример

Сценарий из карточки: загрузить веса в Transformers, vLLM или SGLang и дообучить под свою задачу. Для PEFT используйте официальный chat template: control tokens уже обучены, поэтому эмбеддинги можно не трогать.

ПодробнееКонтекст: 262 144 токена нативно, до 1 010 000.
Языки: 201.
Архитектура: 32 слоя, 8 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN)); 9B параметров.
Ограничение: pre-trained only, не для прямого взаимодействия; требования к VRAM не указаны.
Совместимость: Transformers, vLLM, SGLang.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 17 GB · ready files19 GB
  • FP8 / INT8recommendedweights 8.3 GB9.7 GB
  • 4-bitweights 4.7 GB5.7 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 3080 · BF16 — from RTX 4090

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.

    # losercheers/Qwen3.5-9B-Base-CPT-0.4coding
    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 losercheers/Qwen3.5-9B-Base-CPT-0.4coding --local-dir ~/models/Qwen3.5-9B-Base-CPT-0.4coding
    # 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 Qwen3.5-9B-Base-CPT-0.4coding (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

losercheers/Qwen3.5-9B-Base-CPT-0.4coding
Author
losercheers
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Language models

Endexis

Endexis