Dataset · Axion1-0
Training data for agentic models with reasoning and tool use
Axion1-0/axion-training-data
SFT trajectories and RL training tasks for open models on scientific reasoning and external tool usage.
What's inside
Contains between 10,000 and 100,000 English text examples used during the creation of the Axion 1.0 model (specifically for the 122B R3 run). The dataset is split into two parts featuring SFT trajectories (reasoning and retrieving missing information via tools) and tasks for subsequent reinforcement learning (RL).
What you can do with it
- SFT fine-tuning of LLMs for complex scientific reasoning and API or tool-calling skills
- Training agentic pipelines using reinforcement learning (RL)
- Analysis and distillation of chains of thought from the open Axion 1.0 project
- Tasks
- Text generation, Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 3.4K
- Likes
- 33
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("Axion1-0/axion-training-data")