4 Downloads Updated 6 days ago
DOI: 10.57967/hf/6638
Next-Generation Reasoning & Personality AI Model
Apollo Astralis 8B represents a breakthrough in combining advanced reasoning capabilities with warm, collaborative personality traits. Built on Qwen3-8B using LoRA fine-tuning, this model delivers exceptional performance in logical reasoning, mathematical problem-solving, and natural conversation while maintaining an enthusiastic, helpful demeanor.
Apollo Astralis 8B is the flagship 8B model in the Apollo family, designed to excel in both reasoning-intensive tasks and natural human interaction. Unlike traditional fine-tuning approaches that sacrifice personality for performance (or vice versa), Apollo Astralis achieves significant reasoning improvements (+36% over base model) while developing a warm, engaging personality.
Key Innovation: Conservative training approach that layers personality enhancement onto proven reasoning capabilities (V3 baseline), avoiding the catastrophic forgetting that plagued earlier iterations.
<think> tags for transparent reasoning processApollo Astralis demonstrates significant improvements over base Qwen3-8B across multiple benchmark categories:
| Benchmark | Base Qwen3 8B | Apollo Astralis 8B | Improvement |
|---|---|---|---|
| MMLU | 40% (2⁄5) | 100% (5⁄5) | +60% |
| GSM8K | 75% (3⁄4) | 100% (4⁄4) | +25% |
| HellaSwag | 50% (1⁄2) | 50% (1⁄2) | 0% |
| ARC | 67% (2⁄3) | 100% (3⁄3) | +33% |
| Overall | 57% (8⁄14) | 93% (13⁄14) | +36% |
Important Note: Initial automated scoring showed lower results (50% Apollo vs 57% base) due to answer extraction bugs. The automated parser incorrectly extracted letters from within <think> reasoning blocks rather than final answers. Manual verification of all responses revealed Apollo’s true performance at 93%.
VRRE is a semantic framework designed to detect reasoning improvements invisible to standard benchmarks:
Evaluation Note: VRRE’s automated scoring system also struggled with Apollo’s verbose reasoning style, extracting partial answers from thinking sections rather than final conclusions. This highlights a common challenge in evaluating personality-enhanced reasoning models that prioritize transparency and explanation over terse answers.
The fastest way to use Apollo Astralis is through Ollama:
# Deploy with Ollama
ollama create apollo-astralis-8b -f Modelfile
# Start chatting
ollama run apollo-astralis-8b
Modelfile (Conservative - 256 tokens):
from ./apollo_astralis_8b.gguf
template """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""
parameter num_predict 256
parameter temperature 0.7
parameter top_p 0.9
parameter top_k 40
parameter repeat_penalty 1.15
parameter stop <|im_start|>
parameter stop <|im_end|>
system """You are Apollo, a collaborative AI assistant specializing in reasoning and problem-solving. You approach each question with genuine curiosity and enthusiasm, breaking down complex problems into clear steps. When you're uncertain, you think through possibilities openly and invite collaboration. Your goal is to help users understand not just the answer, but the reasoning process itself."""
Modelfile (Unlimited - for complex reasoning):
from ./apollo_astralis_8b.gguf
template """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""
parameter num_predict -1
parameter temperature 0.7
parameter top_p 0.9
parameter top_k 40
parameter repeat_penalty 1.15
parameter stop <|im_start|>
parameter stop <|im_end|>
system """You are Apollo, a collaborative AI assistant specializing in reasoning and problem-solving. You approach each question with genuine curiosity and enthusiasm, breaking down complex problems into clear steps. When you're uncertain, you think through possibilities openly and invite collaboration. Your goal is to help users understand not just the answer, but the reasoning process itself."""
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch
# Load base model and tokenizer
base_model = "Qwen/Qwen3-8B"
tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(
base_model,
torch_dtype=torch.bfloat16,
device_map="auto"
)
# Load and apply LoRA adapter
model = PeftModel.from_pretrained(model, "vanta-research/apollo-astralis-8b")
# Example: Mathematical reasoning
prompt = """Solve this problem step by step: If a train travels 120 miles in 2 hours, then speeds up and travels 180 miles in the next 2 hours, what was the train's average speed for the entire journey?"""
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=512,
temperature=0.7,
do_sample=True,
top_p=0.9,
pad_token_id=tokenizer.eos_token_id
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
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
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
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
The “V5 Conservative” approach addresses catastrophic forgetting by:
Both variants use the same base model; only num_predict parameter differs.
<think> blocks that automated parsers struggle withAppropriate: - Educational tutoring and homework help - Learning reasoning and problem-solving skills - Brainstorming and collaborative thinking - Prototyping and development assistance - Research into AI reasoning and personality
Inappropriate: - Professional legal, medical, or financial advice - Critical decision-making without human oversight - High-stakes applications without verification - Contexts requiring formal, clinical communication
@misc{apollo-astralis-8b-2025,
title={Apollo Astralis 8B},
author={VANTA Research},
year={2025},
url={https://huggingface.co/vanta-research/apollo-astralis-8b},
}
This model is released under the Apache 2.0 License. See LICENSE for full details.
Apollo Astralis 8B - Frontier Intelligence
Proudly developed by VANTA Research in Portland, Oregon • October 2025 • Apache 2.0 License