erg_assist

Gemma-3N-ERG-Hazmat-Assistant

Model Description

A specialized fine-tune of Gemma-3N focused on hazardous materials emergency response. This model combines ERG (Emergency Response Guidebook) protocols with atmospheric dispersion modeling capabilities similar to MARPLOT/ALOHA software.

• this is the 3rd iteration/attempt addindg vision and speech to the Gemma 3n model

Intended Use

Primary Applications:

• Emergency response planning and training
• Hazmat incident initial assessment
• Chemical safety education
• Plume dispersion estimation
• UN dangerous goods classification assistance

NOT intended for:

• Replacement of professional emergency response
• Final decision-making in actual emergencies
• Legal or regulatory compliance determinations

Training Data

Fine-tuned on:

• ERG 2024 emergency response protocols
• MARPLOT atmospheric dispersion models
• UN dangerous goods classifications (UN1000-UN3500)
• Chemical safety data sheets
• Plume modeling scenarios

Dataset: Johnohhh1/erg_chemical_dataset (36,421 examples)

Capabilities

The model demonstrates knowledge in:

• Chemical Identification: UN numbers, chemical properties, hazard classes
• Emergency Response: Isolation distances, evacuation zones, PPE requirements
• Atmospheric Modeling: Plume dispersion, dense gas behavior, stability classes
• Complex Scenarios: BLEVEs, water-reactive chemicals, multi-chemical incidents
• Decontamination: Procedures for various chemical exposures

Technical Specifications

• Base Model: unsloth/gemma-3n-E4B-it (7.87B parameters)
• Fine-tuned Parameters: 19.2M (0.24% of total)
• Quantization: Q8_0 GGUF format (7.3GB)
• Training: 1 epoch on 36,421 examples
• Hardware: Trained on A100 GPU via Google Colab

Example Usage

With Ollama:

ollama run gemma-erg

Sample Prompts:

"What are the initial isolation distances for UN1005 (anhydrous ammonia)?"
"Calculate plume dispersion for 1000 gallons of chlorine in D stability conditions"
"What PPE is required for sulfuric acid decontamination?"

Performance

The model shows strong performance in:

• Providing structured emergency response guidance
• Calculating atmospheric dispersion estimates
• Identifying chemical-specific hazards
• Recommending appropriate PPE and decontamination

Limitations

• Not a replacement for professional hazmat response
• Atmospheric calculations are estimates requiring field validation
• May not reflect latest regulatory changes post-training
• Regional response protocols may vary from model recommendations

Safety & Ethics

⚠️ Critical Safety Notice: This model provides educational estimates only. Always defer to:

• Local emergency response authorities
• Professional hazmat teams
• Official ERG documentation
• Site-specific assessments

Citation

If you use this model, please cite:

@misc{gemma-erg-hazmat,
  author = {Johnohhh1},
  title = {Gemma-3N-ERG-Hazmat-Assistant},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/Johnohhh1/gemma-3N-finetune-gguf}
}

Acknowledgments

• Trained using Unsloth for efficient fine-tuning
• Based on Google's Gemma architecture
• ERG data from public emergency response resources