llama3-space-reasoning

---

license: apache-2.0 base_model: meta-llama/Meta-Llama-3-8B tags: - llama-3 - peft - lora - space - physics - orbital-mechanics

Llama 3 Space Reasoning

Fine-tuned Llama 3 8B for orbital mechanics reasoning using LoRA.

Training Details

• Base Model: meta-llama/Meta-Llama-3-8B
• Method: LoRA (r=32, alpha=64)
• Dataset: 945 synthetic orbital mechanics problems with step-by-step solutions
• Training: 5 epochs on Kaggle P100 GPU
• Topics: Hohmann transfers, orbital periods, escape velocity, inclination changes

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B")

# Load LoRA adapters
model = PeftModel.from_pretrained(base_model, "YOUR_USERNAME/llama3-space-reasoning")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")

# Generate
prompt = "Calculate delta-v for Hohmann transfer from 400 km to 800 km altitude"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))

Example Output

Prompt: Calculate delta-v for a Hohmann transfer from LEO (400 km) to GEO (35,786 km)

Output:
Step 1: Current orbital velocity
v₁ = √(μ/r₁) = √(398600/(6371+400)) = 7.67 km/s

Step 2: Transfer orbit semi-major axis
a = (r₁ + r₂)/2 = (6771 + 42164)/2 = 24467.5 km
...

Training Code

See https://github.com/isaacasamoah/voyager for full training pipeline.