Supra-Nexus/supra-nexus-o1-training

Supra Nexus O1 Training Datasets

Overview

Comprehensive training datasets for Supra Nexus O1 models, including:

• —Identity training
• —Chain-of-thought reasoning
• —Self-improvement examples (O1.5)
• —Instruction following

Datasets Included

1. Identity Dataset (supra_identity.jsonl)

• —Model identity and alignment
• —Organization information
• —Capability descriptions

2. Instruction Dataset (supra_instruct_*.jsonl)

• —Direct instruction following
• —Various task completions
• —Code generation examples

3. Thinking Dataset (supra_thinking_*.jsonl)

• —Chain-of-thought reasoning with <thinking> tags
• —Mathematical reasoning
• —Complex problem solving

4. Self-Improvement Dataset (nexus-o1-upgrade-1.jsonl)

• —Recursive self-improvement examples
• —Clean code organization patterns
• —Meticulous vs sloppy implementation comparisons
• —Learning from mistakes

Usage

from datasets import load_dataset

# Load all datasets
dataset = load_dataset("Supra-Nexus/supra-nexus-o1-training")

# Access specific splits
identity_data = dataset["identity"]
thinking_data = dataset["thinking"]
upgrade_data = dataset["upgrade"]

Training with Zoo Gym

# Install Zoo Gym
pip install git+https://github.com/zooai/gym

# Train with identity dataset
gym train \
  --model_name_or_path "Supra-Nexus/supra-nexus-o1-instruct" \
  --dataset "Supra-Nexus/supra-nexus-o1-training" \
  --dataset_split "identity"

Dataset Statistics

Model Links

• —supra-nexus-o1-instruct
• —supra-nexus-o1-thinking

Citation

@dataset{supranexus2025training,
  title={Supra Nexus O1 Training Datasets},
  author={Supra Foundation LLC},
  year={2025},
  publisher={HuggingFace}
}

License

Apache 2.0 - Free for research and commercial use.

© 2025 Supra Foundation LLC