Skip to content
Other

MedGuard-ShieldGemma-2B

LoRA adapters for medication prescription checks

2 min readupdated

This release provides a set of LoRA (lightweight add-ons on top of a base model) adapters trained on the MedGuardBench dataset to verify prescription safety against a patient's profile. Given structured patient data and a proposed prescription, the model outputs a safety verdict and flags potential risk categories. Built on the google/shieldgemma-2b base model, it is designed for research into medical guardrails.

Key features

  • Adapters built on google/shieldgemma-2b for checking medication prescriptions.
  • Trained on the MedGuardBench dataset with evaluation across 17 risk categories.
  • Multiple training variants: SFT, Patient Profile Unrolling, and Guided GRPO.
  • Detailed precision, recall, and F1-score metrics for each variant included in the model card.

What you can do

  • Load the adapters via the peft library for AI safety research tasks.
  • Switch between different training branches (sft, cu, grpo) depending on your experiments.
  • Test safety verdict generation and verification tracing (available in the CU variant).

Example

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("google/shieldgemma-2b", device_map="auto")
tok = AutoTokenizer.from_pretrained("sarakhosravi/MedGuard-ShieldGemma-2B", subfolder="cu")
model = PeftModel.from_pretrained(base, "sarakhosravi/MedGuard-ShieldGemma-2B", subfolder="cu")
Read moreDetails
Benchmarks on the MedGuardBench test split (729 rows):
- SFT: Acc 90.1, Rec 95.3, FPR 15.4
- CU (Patient Profile Unrolling): Acc 90.1, Rec 96.6, FPR 16.9
- Guided GRPO: Acc 80.0, Rec 98.4, FPR 40.0
Limitations: These models are intended strictly for research on medication safety guardrails and are not medical devices.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.9 GB · ready files5.9 GB
  • FP8 / INT8recommendedweights 2.5 GB3.3 GB
  • 4-bitweights 1.4 GB2.1 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 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 shieldgemma-2b.

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.

    # sarakhosravi/MedGuard-ShieldGemma-2B
    # ===== 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 google/shieldgemma-2b --local-dir ~/models/shieldgemma-2b
    hf download sarakhosravi/MedGuard-ShieldGemma-2B --local-dir ~/models/MedGuard-ShieldGemma-2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to shieldgemma-2b
  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

google/shieldgemma-2b

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

New posts about versions of shieldgemma-2b (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 8All on Hugging Face
LoRAs and adapters 1All on Hugging Face

About the model

sarakhosravi/MedGuard-ShieldGemma-2B
Author
sarakhosravi
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