Emma - Clean AI's First-Generation Reasoning Model
Competitive with GPT-4o-1 in mathematics, coding, and reasoning
Model Description
Emma R1 is a DeepSeek-R1 reasoning model that has been post-trained by Emma AI team to fill in information gaps in the previous version of the model and to improve its risk profile, while maintaining R1 reasoning capabilities. The model was trained using 110k Safety and Non-Compliance examples from Tulu 3 SFT dataset, in addition to a dataset of ~350k multilingual examples internally developed capturing various topics with reported biases.
Emma R1 has successfully unblocked the majority of previously blocked queries from the original R1 model while outperforming the recently published R1-1776 model (post-trained by Perplexity) in relevant safety benchmarks. These results were achieved while preserving the general reasoning capabilities of the original DeepSeek-R1.
LLM leaderboard Results
Evaluation on General Knowledge and Reasoning
Math
code
For collaborations or feedback, message me on Instagram: @mtlbiohacker