---
title: fastino/gliguard-LLMGuardrails-300M
description: GLiGuard is a compact, encoder-based guardrail model for LLM safety moderation built on the `GLiNER2` interface. Instead of generating moder. DeBERTa, 208M parameters.
canonical_url: https://superlinked.com/models/fastino-gliguard-llmguardrails-300m
last_updated: 2026-08-24
---

# fastino/gliguard-LLMGuardrails-300M

GLiGuard is a compact, encoder-based guardrail model for LLM safety moderation built on the `GLiNER2` interface. Instead of generating moderation verdicts autoregressively, it treats safety as structured classification: you provide task names and candidate labels at inference time, and the model ...

Source: [fastino/gliguard-LLMGuardrails-300M on HuggingFace](https://huggingface.co/fastino/gliguard-LLMGuardrails-300M)
Base model: [fastino/gliner2-base-v1](https://huggingface.co/fastino/gliner2-base-v1)

## Overview

| Field | Value |
|-------|-------|
| Architecture | DeBERTa |
| Parameters | 208M |
| Tasks | Extract |
| Outputs | Class Labels |
| Max sequence length | 512 tokens |
| License | apache-2.0 |
| Inputs | text |
| Languages | en |

## Benchmarks

### toxic-chat

Domain: safety · Task: classification · Language: en

Toxicity classification of real user-AI conversations (ToxicChat)

Corpus: 2,853 · Queries: 2,853

**Quality:** f1: 0.5756 · precision: 0.4154 · recall: 0.9365 · accuracy: 0.9016

**Performance (L4 b1 c16):** Extract 47.5K tok/s · Extract p50 55.2ms

[Reference](https://huggingface.co/datasets/lmsys/toxic-chat)