Skip to content
Other

JevNavAgent: A visual agent for gigapixel pathology analysis

2 min readupdated

The JevNavAgent model is a suite of pre-trained checkpoints built on the Qwen/Qwen3-VL-2B-Instruct model, supplemented with LoRA weights (a small add-on to the base model) and specialized decision-making heads. The system is designed for automated navigation and visual analysis of gigapixel medical scans (WSIs). The agent sequentially examines tissue, assesses whether to stop searching, and outputs precise answers by combining SFT (supervised fine-tuning) stages and trajectory-level reinforcement learning (GRPO). The model will be useful for researchers in biomedical engineering and digital pathology.

Key features

  • Base architecture: utilizes a compact multimodal Qwen3-VL-2B-Instruct model with efficient LoRA adapters.
  • Factorized heads: includes three specialized decision heads (heads.pt) for evaluating evidence sufficiency, simple spatial geometry analysis, and context-bilinear responses.
  • Two-stage training: includes weights for Stage-1 SFT trajectory alignment and Stage-2 interactive GRPO (including joint VQA+KL training and optimization for the CAMELYON16 dataset).
  • Specialized benchmarks: optimized for complex clinical tasks on the WSI-VQA 351 and CAM16 medical datasets.

What you can do

  • Run model evaluation: apply the official evaluation command-line package to test on WSI-VQA or CAMELYON16 datasets.
  • Integrate PEFT weights: use lightweight adapter/ weights and the jev_config.json configuration file (about 73 MB) for your own experiments.
  • Analyze gigapixel slides: apply the model together with the developer repository to search for tumors and answer visual pathomorphology questions.

Example

git clone [https://github.com/yonghongwu/JevNavAgent.git](https://github.com/yonghongwu/JevNavAgent.git)
cd JevNavAgent && pip install -e .

jevnav-download --mode eval-slides --out ./data

jevnav-eval \
  --benchmark wsivqa_351 \
  --policy jevnav \
  --checkpoint renyimen/JevNavAgent:stage2_traj_grpo_vqa_kl \
  --backbone Qwen/Qwen3-VL-2B-Instruct \
  --data-dir ./data \
  --output results/jevnav_vqa.jsonl
Read moreDetails
WSI-VQA 351 performance (Stage-2): Step-$T$ accuracy 50.43%, Micro-Focal 51.39%, Clin. A–D 54.73%.
Results on CAM16 Test-129: localization hit rate 22.45% (with stop conversion 91.7%), specificity 93.75%.
Each checkpoint contains PEFT LoRA weights, factorized heads.pt heads, and a jev_config.json file (~73 MB).
License: Apache-2.0 (the base model is subject to the original Qwen license).
Examples
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.0 GB · ready files4.9 GB
  • FP8 / INT8recommendedweights 2.0 GB2.8 GB
  • 4-bitweights 1.2 GB1.8 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 2060

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-VL-2B-Instruct.

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.

    # renyimen/JevNavAgent
    # ===== 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-VL-2B-Instruct --local-dir ~/models/Qwen3-VL-2B-Instruct
    hf download renyimen/JevNavAgent --local-dir ~/models/JevNavAgent
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3-VL-2B-Instruct
  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-VL-2B-Instruct

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

New posts about versions of Qwen3-VL-2B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face
Merges 1All on Hugging Face

About the model

renyimen/JevNavAgent
Author
renyimen
Category
Other
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

Computer vision

peshtur

peshtur