normster/RealGuardrails-Qwen2.5-7B-SFT

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Feb 17, 2025License:mitArchitecture:Transformer Open Weights Cold

The normster/RealGuardrails-Qwen2.5-7B-SFT is a 7.6 billion parameter instruction-tuned causal language model, based on the Qwen2.5 architecture, developed by normster. It is specifically fine-tuned on the RealGuardrails dataset to enhance system prompt adherence and precedence, making it particularly effective at following complex instructions. With a context length of 32768 tokens, this model excels in scenarios requiring strict adherence to guardrails and system-level directives.

Loading preview...

Model Overview

The normster/RealGuardrails-Qwen2.5-7B-SFT is a 7.6 billion parameter instruction-tuned language model built upon the Qwen2.5 architecture. Developed by normster, this model's primary differentiator is its specialized training on the RealGuardrails dataset. This dataset focuses on improving the model's ability to adhere to system prompts and prioritize instructions, a critical aspect for reliable AI applications.

Key Capabilities

  • Enhanced System Prompt Adherence: Specifically fine-tuned to follow system-level instructions and guardrails with high fidelity.
  • Instruction Following: Excels in scenarios where strict adherence to given directives is paramount.
  • Qwen2.5 Architecture: Leverages the robust capabilities of the Qwen2.5 base model.
  • Large Context Window: Supports a context length of 32768 tokens, allowing for processing extensive inputs while maintaining instruction adherence.

Training Details

The model was trained via Supervised Fine-Tuning (SFT) on the systemmix split of the RealGuardrails dataset, comprising approximately 150,000 examples. Training was conducted using the custom torchllms library, with a batch size of 128, a learning rate of 2e-5, and a maximum sequence length of 4096 during training. This focused training regimen ensures its specialized performance in guardrail enforcement.

Good for

  • Applications requiring strict adherence to predefined rules and system prompts.
  • Use cases where model outputs must consistently stay within specified boundaries.
  • Developing AI agents that need to follow complex, multi-layered instructions reliably.

Finetunes
3 models