pozapas/gemma-3-evacuation

VISIONPricing:Input $0.2 / Output $0.4Concurrent Unit Cost:1Model Size:4.3BQuant:BF16Context Size:32kPublished:Jun 16, 2025License:ccArchitecture:Transformer Featherless Exclusive Cold

The pozapas/gemma-3-evacuation model is a 4.3 billion parameter Gemma-3 variant, fine-tuned by pozapas for specialized question answering in the evacuation and fire safety domain. It excels at providing accurate responses to technical questions about building evacuation, fire safety regulations, and emergency planning. This model leverages a 32768 token context length and is optimized for domain-specific accuracy, achieving 83% source citation accuracy and 91% numerical value accuracy. It is intended to support emergency planning professionals, building designers, and code consultants with regulatory compliance and best practices.

Loading preview...

Overview

This model, pozapas/gemma-3-evacuation, is a fine-tuned version of Google's Gemma-3-4B-it, specifically adapted for the domain of evacuation and fire safety. Developed by pozapas, it utilizes Parameter-Efficient Fine-Tuning (PEFT) with LoRA on the pozapas/evacuation-safety-qa dataset, comprising over 23,000 question-answer pairs. The model has a context length of 2048 tokens and is available in Q8_0 GGUF format for efficient local inference.

Key Capabilities

  • Domain-Specific Q&A: Provides accurate answers to technical questions regarding building evacuation, fire safety regulations, and emergency planning.
  • High Accuracy: Demonstrates strong performance in domain-specific metrics, including 83% source citation accuracy and 91% numerical value accuracy.
  • Regulatory Support: Assists in applying safety regulations and understanding fire safety requirements.
  • Efficient Inference: Optimized for local deployment with a Q8_0 GGUF build for llama.cpp.

Intended Use Cases

  • Emergency Planning: Supports professionals in decision-making processes.
  • Building Design & Code Consulting: Aids designers and consultants in applying safety regulations.
  • Education: Educates stakeholders on fire safety best practices.

Limitations

  • Knowledge is restricted to the training dataset's scope and may not reflect post-June 2025 code changes.
  • Does not fully cover regional variations in building codes.
  • Should not replace professional engineering judgment or official code interpretation.

Finetunes
2 models