SurvivalAI v2 β€” Offline Emergency Preparedness Assistant

A fine-tuned Qwen2.5-0.5B model trained on 133 survival and emergency preparedness guides across 15 categories. Designed to run fully offline on mobile devices (iPhone SE via Private LLM or PocketPal).

⚠️ Required App Settings (IMPORTANT)

These settings must be applied or the model will repeat itself:

Setting Value
Repetition Penalty 1.15
Max Tokens / Max Length 512
Temperature 0.7

PocketPal: Tap model β†’ Settings β†’ Repeat Penalty β†’ 1.15
Private LLM: Chat settings β†’ Repetition Penalty β†’ 1.15

What it knows

  • First aid & medical emergencies
  • Water purification & sourcing
  • Food foraging & preservation
  • Shelter & fire building
  • Power outage preparedness
  • Navigation & orienteering
  • Communications (ham radio, comms plans)
  • Tools & repair
  • Military field manuals (FM series)
  • FEMA emergency guides
  • Sanitation & hygiene
  • Medicinal plants
  • And more

Which file to download?

File Size Recommended?
survival-qwen-0.5b-q5_k_m.gguf 401MB βœ… Yes β€” best quality
survival-qwen-0.5b-q4_k_m.gguf 379MB If storage is tight

Usage on iPhone SE

  1. Install PocketPal or Private LLM from the App Store
  2. PocketPal: Models β†’ Add Model β†’ search wattsonandme/survival-ai-0.5b β†’ pick q5_k_m
  3. Private LLM: + β†’ Download from URL β†’ paste: https://huggingface.co/wattsonandme/survival-ai-0.5b/resolve/main/survival-qwen-0.5b-q5_k_m.gguf
  4. Set Repetition Penalty to 1.15 before chatting
  5. Enable Airplane Mode β€” works fully offline

System Prompt

You are SurvivalAI, an offline emergency preparedness assistant. Answer clearly and concisely based on survival best practices.

Model Details

Property Value
Base model Qwen2.5-0.5B-Instruct
Fine-tuning LoRA (rank 16)
Training data 8,735 Q&A pairs from 133 PDFs
Format GGUF (Q4_K_M and Q5_K_M)
Downloads last month
11
GGUF
Model size
0.5B params
Architecture
qwen2
Hardware compatibility
Log In to add your hardware

4-bit

5-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support