saravanakarthikeyan/GuardShield-Qwen2.5-3B-16bit

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 1, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

saravanakarthikeyan/GuardShield-Qwen2.5-3B-16bit is a 3.09 billion parameter Qwen 2.5-based language model, specifically designed as an AI content moderation guardrail. This FP16 merged model excels at classifying content for safety, achieving 86.35% safety recall and 92.95% benign classification precision. It is optimized for production microservices and high-throughput GPU serving engines like vLLM and TGI, ensuring sub-30ms time-to-first-token latency.

Loading preview...

GuardShield-Qwen2.5-3B-16bit: AI Content Moderation Guardrail

GuardShield-Qwen2.5-3B-16bit is a 3.09 billion parameter model built on the Qwen 2.5 architecture, provided as a fully merged FP16 Safetensors. It is engineered to function as an AI content moderation guardrail, specializing in analyzing prompts and classifying them into safety categories using a JSON output schema.

Key Capabilities & Performance

  • Content Moderation: Designed to detect and classify unsafe content with high accuracy.
  • High Safety Recall: Achieves a Safety Recall (Detection Rate) of 86.35%.
  • High Benign Precision: Demonstrates Benign Classification Precision of 92.95%.
  • JSON Adherence: Guarantees 100.0% JSON Syntax Adherence for reliable structured outputs.
  • Optimized for Throughput: Built for production microservices and high-throughput serving engines like vLLM and Hugging Face TGI, offering sub-30ms time-to-first-token latency on modern GPU hardware.

Ideal Use Cases

  • API Gateways: Integrating content moderation into OpenAI-compatible API gateways.
  • Production Microservices: Deploying as a fast, reliable guardrail in production environments.
  • Real-time Content Filtering: Applications requiring immediate classification of user inputs for safety.
  • Structured Output: Scenarios where consistent JSON output for content classification is critical.