Training Data & Citations
The datasets below are used to train Snortle AI. Unless noted as generated locally, datasets are from Hugging Face.
Pre-training
SmolLM-Corpus
Author: HuggingFaceTB
Source: Hugging Face dataset page
License: ODC-By v1.0
Use: Pre-training. Subsets used: cosmopedia-v2 and fineweb-edu-dedup.
GSM8K
Author: OpenAI
Source: Hugging Face dataset page
License: MIT
Use: Pre-training (math word problems, main split).
Python-Edu
Author: Avelina
Source: Hugging Face dataset page
License: ODC-By v1.0
Use: Pre-training (Python code, for the code portion of the mix).
Snortle Arithmetic (generated)
Author: Hiram Jiang
Source: Generated locally with Python code.
Use: Pre-training.
Fine-tuning
Snortle Fine-Tuning Data (generated)
Author: Hiram Jiang
Source: Generated locally with Python code.
Use: Fine-tuning.
SmolTalk (smol-smoltalk)
Author: HuggingFaceTB
Source: Hugging Face dataset page
License: Apache 2.0
Use: Fine-tuning (chat conversations).
Dataset names, authors, sources, licenses, and use notes are provided for attribution and transparency. Please refer to each linked source and license for their applicable terms.