Skip to content
Computer vision

Signpost-Jev-4B

a model for structured decisions and benchmark evaluation

2 min readupdated

Signpost-Jev-4B is a specialized model built on Qwen3.5-4B, trained to make structured decisions following the JevBench standard. It takes a state (as plain text or a JSON record) along with typed questions, and outputs probabilities for each answer option (true/false, multiple choice, or scales). Instead of generating free-form text, the model computes each decision in a single pass through the answer tokens. It is optimized to run with vLLM and is built for developers working on automated decision-making and data evaluation systems.

Key features

  • Purpose-built: fine-tuned for JevBench-style decision-making without generating unnecessary text.
  • Low latency: delivers results with a median latency of 44 ms per request on an L40S GPU.
  • Architecture: based on Qwen3.5-4B using LoRA (a lightweight adapter on top of the base model) at rank 16 in fp16, with the weights merged afterward.
  • Accuracy: achieves high accuracy on public JevBench datasets (up to 0.979 on easy tasks).

What you can do

  • Deploy a service: self-host the model via vLLM to handle decision-making requests.
  • Analyze structured data: feed in JSON records or text to score them against defined criteria and scales.
  • Test hypotheses: use it as an automated judge to verify structured outputs.

Example

Running the model via a vLLM server:

vllm serve skamalj/signpost-jev-4b --max-model-len 32768 --max-num-seqs 8 --logprobs-mode raw_logprobs
Read moreDetails
JevBench benchmark accuracy: easy — 0.979 (48 items), original — 0.875 (72 items), hard — 0.730 (111 items), overall — 0.827.
Median latency: 44 ms per decision (vLLM on a single L40S).
Training: 2 epochs on the full data mix, followed by 1 short epoch on hard records.
License: Apache-2.0.
Discuss Telegram 0 27

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 17 GB · ready files19 GB
  • FP8 / INT8recommendedweights 13 GB15 GB
  • 4-bitweights 11 GB12 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 5080 · 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 Qwen3.5-4B.

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.

    # skamalj/signpost-jev-4b
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Qwen/Qwen3.5-4B --local-dir ~/models/Qwen3.5-4B
    hf download skamalj/signpost-jev-4b --local-dir ~/models/signpost-jev-4b
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.5-4B
  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

Requires the base model

Qwen/Qwen3.5-4B

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

  1. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. signpost-jev-4b

New posts about versions of Qwen3.5-4B-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 12All on Hugging Face
Merges 1All on Hugging Face

About the model

skamalj/signpost-jev-4b
Author
skamalj
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

270

Ask AI about this model

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

Comments