CERT-SLM v1

A custom transformer model developed for CERT insider-threat detection.

Model details

  • Architecture: Custom Qwen3-inspired transformer
  • Training approach: Trained from scratch
  • Task: Binary classification of insider activity
  • Output classes: normal and suspicious
  • Best-validation checkpoint: micro-iteration 30,016
  • Hidden dimension: 768
  • Transformer layers: 12
  • Attention heads: 12
  • Key/value groups: 6
  • Maximum sequence length: 128

This is a custom PyTorch implementation, not an official Qwen3 checkpoint.

Files

  • cert_qwen3_best_model.pth: Model weights
  • config.json: Architecture configuration
  • model.py: Custom model implementation
  • tokenizer/: Tokenizer files
  • requirements.txt: Python dependencies

Important limitations

This model was adapted for a binary classification task. It is not a general-purpose conversational assistant and is not directly an Ollama-compatible chat model.

The checkpoint corresponds to the best-validation model saved at micro-iteration 30,016.

Usage

Load the architecture from model.py, initialize it using config.json, and load the state dictionary from the .pth file.

Refer to the repository source code for the inference implementation.

Downloads last month
32
GGUF
Model size
4B params
Architecture
qwen3
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support