Skip to content
Other

Ripple-65B Preview DFlash2

драфтер для спекулятивного декодирования

2 min readupdated

Это BF16 draft-модель от ricursiveAI для спекулятивного декодирования с Ripple-65B Preview или Ripple-65B Preview NVFP4. Содержит 4.04B параметров и архитектуру DFlash2DraftModel, но не генерирует текст самостоятельно — нужна target-модель.

Ключевые особенности

  • Предлагает блоки по 16 токенов, target-модель их проверяет; корректная реализация speculative rejection sampling сохраняет распределение target.
  • DFlash2 объединяет параллельный block drafting, dynamic convolutions и candidate selector.
  • Пять слоёв: четыре sliding-attention с окном 4 096 токенов и один full-attention NoPE; hidden size 5 120, query heads / KV heads — 80 / 8.
  • 4.04B параметров в BF16; shared token embedding и output head берутся у target и в этом репозитории не хранятся.
  • 32 кандидата на позицию, selector rank-256, kernel size 4, group size 16; configured maximum positions — 131 072.

Пример

Флаги для SGLang-сервера с поддержкой Ripple; это дополнение к запуску, а не самостоятельная команда.

--speculative-algorithm DFLASH --speculative-draft-model-path ricursiveAI/Ripple-65B-Preview-DFlash2 --speculative-draft-model-quantization unquant --speculative-num-draft-tokens 16
ПодробнееВалидация экспорта: runtime ripple-sglang на восьми H100 с target-чекпоинтом RL-step-500.
Совместимость и производительность с другими ревизиями target или версиями runtime не заявлены.
Нужен runtime с поддержкой Ripple и DFlash2DraftModel, включая per-layer attention pattern, NoPE-слой, four-tap convolutions и target hidden-state taps.
Tokenizer, chat template, token embeddings и output head — от target-модели. Лицензия Apache 2.0.
Discuss Telegram 18 2.3K

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.

    # ricursiveAI/Ripple-65B-Preview-DFlash2
    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 ricursiveAI/Ripple-65B-Preview-DFlash2 --local-dir ~/models/Ripple-65B-Preview-DFlash2
    # 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 Ripple-65B-Preview-DFlash2 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

ricursiveAI/Ripple-65B-Preview-DFlash2
Author
ricursiveAI
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

2.3K18

Ask AI about this model

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

Comments