Skip to content
Other

MCApple Guard v1.0 Flash

source code and artifacts of prompt injection detector

2 min readupdated

This repository provides the source code and public model artifacts for detecting prompt injections, jailbreaks, and system instruction overrides in text. The tool classifies input text as either safe or an attack without executing the submitted string. It relies on a ready-to-use ONNX artifact for inference, a standalone native training checkpoint based on DeBERTa-v2, and an included FastAPI service. This solution is useful for developers of AI agents and system gateways to filter incoming requests before processing.

Key features

  • Includes a ready-to-use FastAPI service for deploying a local detection API
  • Supports fast local CPU ONNX inference via prediction scripts
  • Contains a portable trainer for fine-tuning and export with parity checking
  • Supports a sliding text processing window with a limit of up to 3584 tokens

What you can do

  • Launch your own local attack detection API service via Uvicorn
  • Screen text for prompt injection attempts using the ONNX prediction script
  • Perform local fine-tuning of the model on your own annotated JSONL datasets
  • Export trained checkpoints to ONNX format with result parity verification

Example

Running the local FastAPI service via the command line with API key generation:

export API_KEY="$(python -c 'import secrets; print(secrets.token_urlsafe(32))')"
python -m uvicorn app:app --host 127.0.0.1 --port 8000
Read moreDetails
Classification threshold: 0.067598
Architecture: DeBERTa-v2 with six encoder layers
Limitations: maximum text length of 32768 characters, up to 16 windows of 256 tokens
Requirements: Python 3.12, PyTorch 2.8+, Transformers 4.57.6
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.5 GB · ready files1.1 GB
  • FP8 / INT8recommendedweights 0.3 GB0.8 GB
  • 4-bitweights 0.1 GB0.7 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.

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.

    # gaozx/mcapple-guard-v1.0-flash-20261121
    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 gaozx/mcapple-guard-v1.0-flash-20261121 --local-dir ~/models/mcapple-guard-v1.0-flash-20261121
    # The model is in ~/models. How to run it is in the model card (link above).
  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 mcapple-guard-v1.0-flash-20261121 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

gaozx/mcapple-guard-v1.0-flash-20261121
Author
gaozx
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments