zenlm/zen-guard

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 17, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

zenlm/zen-guard is a 4 billion parameter content-safety and guardrail model, repackaged from Alibaba Qwen's Qwen3Guard-Gen-4B. It is designed for commercial use under an Apache-2.0 license, providing robust content moderation capabilities. This model is specifically rebased from a prior Qwen2.5-3B base to ensure clean commercial applicability, making it suitable for filtering and safeguarding AI outputs.

Loading preview...

zenlm/zen-guard Overview

zenlm/zen-guard is a 4 billion parameter content-safety and guardrail model, primarily designed for filtering and moderating AI-generated content. It is a repackaged version of Alibaba Qwen's Qwen3Guard-Gen-4B, ensuring its availability under an Apache-2.0 license for commercial applications.

Key Characteristics

  • Content Safety: Functions as a dedicated guardrail model to ensure AI outputs adhere to safety guidelines.
  • Commercial Use: Rebased from a non-commercial Qwen2.5-3B base onto the Apache-2.0 licensed Qwen3Guard, making it suitable for commercial deployment.
  • Origin: Not trained from scratch; it leverages the robust capabilities of the Alibaba Qwen3Guard architecture.

Use Cases

  • Content Moderation: Ideal for applications requiring automated filtering of inappropriate or unsafe text.
  • AI Guardrails: Can be integrated into larger LLM systems to enforce content policies and prevent undesirable outputs.

Note: This model has been superseded by zenlm/zen3-guard, which is the canonical name for the updated version. However, zen-guard remains available for reproducibility purposes.

Finetunes
5 models