apollo-astralis-2

<div align="center">

<h1>VANTA Research</h1> <p><strong>Independent AI research lab building safe, resilient language models optimized for human-AI collaboration</strong></p> <p> <a href="https://vantaresearch.xyz"><img src="https://img.shields.io/badge/Website-vantaresearch.xyz-black" alt="Website"/></a> <a href="https://unmodeledtyler.com/work-with-vanta-research"><img src="https://img.shields.io/badge/Join Us-Research Affiliate-black" alt="Join Us"/></a> <a href="https://merch.vantaresearch.xyz"><img src="https://img.shields.io/badge/Merch-merch.vantaresearch.xyz-sage" alt="Merch"/></a> <a href="https://x.com/vanta_research"><img src="https://img.shields.io/badge/@vanta_research-1DA1F2?logo=x" alt="X"/></a> <a href="https://github.com/vanta-research"><img src="https://img.shields.io/badge/GitHub-vanta--research-181717?logo=github" alt="GitHub"/></a> </p> </div>

---

Apollo Astralis 2

Apollo Astralis 2 is a fine-tuned language model built on the new Ministral 3 8B Reasoning architecture, optimized for:

• Logical reasoning and inference
• Scientific and mathematical problem-solving
• Commonsense understanding
• Multi-step analytical thinking
• Collaborative problem-solving

This model represents a 10% performance improvement over it's previous iteration, with significant gains across reasoning benchmarks while maintaining strong general capabilities.

---

Model Details

• Model Name: Apollo Astralis 2
• Developer: VANTA Research
• Base Model: Ministral-3-8B-Reasoning-2512
• Training Method: Low-Rank Adaptation (LoRA)
• Parameters: 8B base + 70.5MB LoRA adapter
• Training Data: Custom in-house synthetic data generation containing ~26,000 examples across reasoning, logic, math, and science domains

Dataset Composition

• Logical Reasoning
• PIQA
• Mathematics
• Science & Commonsense
• CommonsenseQA
• WinoGrande
• Human-AI Collaboration
• Identity & Persona

---

Benchmark Results

---

Quick Start

import torch
from transformers import AutoTokenizer, BitsAndBytesConfig, Mistral3ForConditionalGeneration
# Note: PEFT not needed - this is the full merged model!

# Configure 4-bit quantization
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
)

# Load model
base_model = Mistral3ForConditionalGeneration.from_pretrained(
    "Ministral-3-8B-Reasoning-2512",
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
    torch_dtype=torch.float16,
)

model = Mistral3ForConditionalGeneration.from_pretrained(
    "vanta-research/apollo-astralis-2",
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
    torch_dtype=torch.float16,
)
tokenizer = AutoTokenizer.from_pretrained("vanta-research/apollo-astralis-2")
model.eval()

Examples

Logical Reasoning

prompt = """If all roses are flowers, and some flowers fade quickly, can we conclude that some roses fade quickly? Explain your reasoning."""

# Apollo's response includes:
# - Clear problem breakdown
# - Syllogistic structure analysis
# - Identification of logical fallacy
# - Final conclusion with explanation

Mathematical Problem Solving

prompt = """A store offers 25% off, then an additional 10% off the sale price. Is this the same as 35% off? Show your work."""

# Apollo's response includes:
# - Step-by-step calculation
# - Comparison of compound vs simple discounts
# - Clear final answer
# - Practical explanation of why they differ

Creative Problem Solving

prompt = """I have a 3-liter jug and a 5-liter jug. How can I measure exactly 4 liters?"""

# Apollo's response includes:
# - Systematic approach
# - Step-by-step solution
# - Explanation of mathematical principles
# - Enthusiastic encouragement

---

Technical Limitations

• Memory: Requires ~16GB for full precision inference (less with quantization)
• Speed: Response generation may be slower due to chain-of-thought reasoning
• Deployment: Best served via Ollama or HuggingFace; other formats may require conversion

Ethical Considerations

Responsible Use

• Educational Focus: Designed for learning and exploration, not professional advice
• Verification Required: Always verify critical information, especially in technical domains
• Personality Awareness: Warm tone should not be mistaken for emotional capacity or consciousness
• Bias Acknowledgment: May reflect biases from base model and training data

Intended Use Cases

Appropriate:

• Educational tutoring and homework help
• Learning reasoning and problem-solving skills
• Brainstorming and collaborative thinking
• Prototyping and development assistance
• Research into AI reasoning and persona stability

Inappropriate:

• Professional legal, medical, or financial advice
• Critical decision-making without human oversight
• High-stakes applications without verification
• Contexts requiring formal, clinical communication

Citation

@misc{apollo-astralis-2,
  title={Apollo Astralis 2},
  author={VANTA Research},
  year={2025},
  url={https://huggingface.co/vanta-research/apollo-astralis-2},
}

License

Apache 2.0

---

Contact

• Organization: [email protected]
• Engineering/Design: [email protected]

Proudly developed by VANTA Research in Portland, Oregon