You can deploy this model and start earning money today!

KAERI-MLP/
AtomicGPT-qwen2.5-7B

Name: AtomicGPT-qwen2.5-7B
Rating: 2.5 (15 reviews)

continual-pretraining

Views

No views yet

AtomicGPT-Qwen2.5-7B

AtomicGPT is a large language model (LLM) specialized in the nuclear engineering domain, developed at the Korea Atomic Energy Research Institute (KAERI). Based on Qwen2.5-7B, AtomicGPT deeply understands various nuclear technologies, theories, and terminology, including reactor design, radiation shielding, the nuclear fuel cycle, and nuclear safety and regulations. With this expertise, AtomicGPT delivers precise answers to technical and specialized questions in the nuclear domain.

Model Details


Base Model	Qwen/Qwen2.5-7B
Training	Continual Pre-training (CPT) + Instruction Tuning (IT)
Domain	Nuclear Engineering
Languages	Korean, English
License	Apache 2.0

Performance

Model	Multiple-Choice (EM)	Short-Answer (F1, %)	Descriptive (1–10)
Qwen2.5-7B (base)	28	15.37	3.67
AtomicGPT-Qwen2.5-7B (ours)	37	18.08	3.94
GPT-4	48	31.29	7.70

All evaluations were conducted under a zero-shot setting.

How to Use

python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "KAERI-MLP/AtomicGPT-Qwen2.5-7B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
)
model.eval()

input_text = "Query about Nuclear (Atomic Energy)"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**input_ids, max_new_tokens=1024)
print(tokenizer.decode(outputs[0]))

Citation

If you use this model in your research, please cite:

bibtex

@article{atomicgpt2026,
  title={AtomicGPT: A Domain-Adapted Large Language Model for Secure On-Premise Applications in Nuclear Engineering},
  journal={Nuclear Engineering and Technology},
  year={2026}
}

About

Developed by the MLP(Multimodal Language Processing) team at the Korea Atomic Energy Research Institute (KAERI).

KAERI-MLP/AtomicGPT-qwen2.5-7B