README.md

---
license: cc-by-4.0
tags:
- llama-3.2
- nuclear-physics
- lora
---
# Vers3Dynamics Nuclear-Expert

**From 26 Million kg of Ore to Mushroom Cloud** — A Llama-3.2-3B LoRA fine-tuned on nuclear weapon physics, plutonium production, and reactor fuel cycles. Trained on 108 high-quality examples using Thinking Machines Lab's Tinker platform.

## Capabilities
- **Yield Calculations**: "What's the yield for a 15 kg Pu pit?" → "59 kt TNT, fireball ~80 m radius."
- **Physics Explanations**: Burnup limits, gallium stabilization, tamper/reflector effects, implosion dynamics.
- **Dramatic & Educational**: Responses blend awe with responsibility — e.g., "The pit compresses in microseconds... but this is simulation only."

**Warning**: Educational/research use only. No classified info or weapon instructions. Based on declassified IAEA/DOE sources.

## Usage
```python
from peft import PeftModel
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch

# Load base + LoRA
base = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.2-3B",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base, "ciaochris/Nuclear-Expert-LoRA-3B")

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B")

# Pipeline
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Query
messages = [{"role": "user", "content": "Yield for a 12 kg plutonium pit?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
output = pipe(prompt, max_new_tokens=200, temperature=0.7)
print(output[0]["generated_text"])