Skip to content
Other

DeFT-RAG-DM-4B

4B-диспетчер решений для multi-hop RAG

2 min read

DeFT-RAG-DM-4B — дообученная decision-модель из статьи DeFT-RAG (UT Dallas). Это 4,54B-модель на базе StartLux-Decision-4B: она выдаёт вероятности для yes/no-решений и оценку глубины вопроса в пайплайне multi-hop RAG. Язык — английский.

Ключевые особенности

  • Relevance: нужен ли найденный пассаж для ответа на вопрос или подвопрос; порог p(yes) ≥ 0.5.
  • Sufficiency: хватает ли top-10 пассажей, чтобы ответить; порог p(yes) ≥ 0.7.
  • Question depth: оценивает число hops по самому вопросу — none, one, two или three+ — как score от 0 до 3.
  • Direct answer: отвечает напрямую по пассажам, если они достаточны и глубина ≤ 2; иначе запускает цикл подвопросов.
  • Enough facts: отвечают ли уже найденные пары вопрос-ответ на исходный вопрос; порог ≥ 0.5.
  • Same question: в QA-кэше проверяет, не запрашивают ли два подвопроса одну и ту же информацию; порог ≥ 0.8.

Пример

huggingface-cli download startlux-models/StartLux-Decision-4B --local-dir StartLux-Decision-4B
huggingface-cli download Dingy-Kang/DeFT-RAG-DM-4B --local-dir DeFT-RAG-DM-4B
cd StartLux-Decision-4B && pip install -r requirements.txt
python -m startlux_decision.server --model ../DeFT-RAG-DM-4B --port 8090 --no-images --max-length 16384 --name deftrag-dm
ПодробнееБаза — StartLux-Decision-4B; LoRA rank 16, alpha 32, dropout 0.05, 1 эпоха, LR 1e-4.
Обучение: 48 036 примеров и 7 806 dev, только train-сплиты MuSiQue, 2WikiMultiHopQA и HotpotQA; тестовые данные не использовались.
Для сервера хватит одной GPU с ~20 GB свободной памяти; язык — английский.
Лицензия CC BY-NC 4.0: только исследовательское и некоммерческое использование с атрибуцией.
Discuss Telegram 1 7

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 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 3080

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.

    # Dingy-Kang/DeFT-RAG-DM-4B
    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 Dingy-Kang/DeFT-RAG-DM-4B --local-dir ~/models/DeFT-RAG-DM-4B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to DeFT-RAG-DM-4B
  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 StartLux-Decision-4B (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Dingy-Kang/DeFT-RAG-DM-4B
Author
Dingy-Kang
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

71

Ask AI about this model

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

Comments