-
Parameters: 354.8M Parameters
-
Base: GPT-2 Medium (Decoder)
-
Tokenizer: AutoTokenizer
-
Training: 2-Phase Full SFT
-
Purpose: Research Model -- Proof of Concept
-
Strengths: Short factual responses, small stories, basic reasoning
-
Limitations: Hard-limit at 1-2 Sentences, tends to misunderstand, no safety filter, prone to hallucinate
Citation
Kharazmi, A. (2025). Curiosity-16: A 354.8M Parameter Large Language Model (v1.0). Zenodo.
https://doi.org/10.5281/zenodo.17871084