sunkencity/survival-expert-qwen-32b

Survival Expert Qwen 32B

Survival Expert Qwen 32B is a specialized large language model fine-tuned for preparation, survival, and emergency response scenarios.

This model is a fine-tuned version of Qwen/Qwen2.5-32B-Instruct using Supervised Fine-Tuning (SFT) with TRL.

Model Description

Unlike general-purpose models, this model has been specifically trained to excel in:

• Disaster Preparedness: Comprehensive planning for urban and wilderness emergencies.
• Resource Management: Efficient rationing, purification, and procurement of water, food, and energy.
• Risk Assessment: Analyzing situational hazards and providing mitigation strategies.
• Tactical & Practical Advice: Clear, actionable steps for high-stakes situations.

The fine-tuning process significantly enhances its focus on survival-specific logic, reducing refusal rates for sensitive survival topics while maintaining safety, and improving the practical applicability of its advice compared to the base Qwen-2.5-32B model.

Quick start

from transformers import pipeline

question = "What are the first three priorities in a sudden wilderness stranding scenario?"
generator = pipeline("text-generation", model="sunkencity/survival-expert-qwen-32b", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=512, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

This model was trained with SFT on Hugging Face Jobs infrastructure.

Framework versions

• TRL: 0.27.0
• Transformers: 4.57.6
• Pytorch: 2.9.1
• Datasets: 4.5.0
• Tokenizers: 0.22.2

Citations

Cite TRL as:

@misc{vonwerra2022trl,
        title        = {{TRL: Transformer Reinforcement Learning}},
        author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
        year         = 2020,
        journal      = {GitHub repository},
        publisher    = {GitHub},
        howpublished = {\url{https://github.com/huggingface/trl}}
}