Skip to content
Language models

SEL Qwen 9B

accurate retrieval of relevant snippets in text and images

2 min readupdated

SEL Qwen 9B is an adapter for the multimodal Qwen3.5-9B model by spock-oss built for targeted information extraction. You feed the model a question along with text, an image, or both, and it outputs relevance scores for every input text token and image patch. Instead of generating an answer token by token, the model runs in a single forward pass through a dedicated dense selection head. The release includes a rank-32 LoRA adapter (a lightweight add-on on top of the base model) and a trained selection head, while the language backbone itself is quantized to NF4 on load. It is a solid fit for developers building RAG systems and pipelines that need to pinpoint exact answers across documents and visual data.

Key features

  • Single forward pass: scores the relevance of every snippet at once without slow autoregressive text generation.
  • Multimodal analysis: works with text files, images, or both together, scoring individual visual patches.
  • Hybrid attention: 8 full-attention layers in the model are converted to bidirectional, while the remaining 24 gated delta layers stay directional.
  • Lower VRAM footprint on load: the frozen language backbone is quantized to NF4, while the vision encoder stays in BF16 precision.

What you can do

  • Upgrade RAG and document search: extract exact sentences and clauses from contracts or knowledge bases tailored to a specific query.
  • Pinpoint regions in images: locate specific visual areas in screenshots, scans, or photos based on a text question.
  • Tune recall to fit your task: adjust the cutoff via the --threshold flag (defaulting to 0.4 for text and 0.3 for images) to pull narrower or broader context.

Example

hf download spock-oss/sel-qwen-9b --local-dir ./sel-qwen-9b
CUDA_VISIBLE_DEVICES=0 sel predict --weights ./sel-qwen-9b \
  --text 'This agreement is governed by the laws of Oregon.' \
  --query 'Which state governs the agreement?'
Read moreDetails
Benchmarks: 77.6% Mean F1 (task-macro) on the Run09 test set (3,136 samples), 75.3% across a 26-task gallery, and 79.9% on additional dev sources.
Hardware and limits: requires a single NVIDIA GPU with BF16 support; maximum sequence length is 2,048 tokens (including the question and visual patches).
Training: 88.65M trainable parameters (LoRA + dense head), trained for 2 epochs on 43.09M input tokens.
Model family: 0.8B, 4B, and 27B parameter versions are also available in the release.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 18 GB · ready files20 GB
  • FP8 / INT8recommendedweights 9.3 GB11 GB
  • 4-bitweights 5.4 GB7.0 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 2080 Ti · BF16 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Qwen3.5-9B.

Language models: with a 8192-token context.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # spock-oss/sel-qwen-9b — A30, FP8 / INT8
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve Qwen/Qwen3.5-9B --max-model-len 8192 --quantization fp8 --enable-lora --max-lora-rank 64 --lora-modules sel-qwen-9b=spock-oss/sel-qwen-9b
    Add more LoRAs to Qwen3.5-9B
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

Requires the base model

Qwen/Qwen3.5-9B

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. Qwen3.5-9B-Basefamily root
  2. Qwen3.5-9B
  3. sel-qwen-9b

New posts about versions of Qwen3.5-9B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 6All on Hugging Face
Merges 1All on Hugging Face

About the model

spock-oss/sel-qwen-9b
Author
spock-oss
Category
Language models
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