Cogito v1 is a suite of advanced large language models (LLMs) developed using Iterated Distillation and Amplification (IDA) to achieve superior performance in reasoning and intelligence.
Features
Iterated Distillation and Amplification
This is an alignment strategy that enhances model intelligence beyond its overseers' limitations through iterative self-improvement. It involves amplifying intelligence using subroutines and then distilling this into the model's parameters.
Open License
All model sizes, ranging from 3B to 70B parameters, are available under an open license, allowing users to download and use them freely from platforms like Huggingface and Ollama.
Multiple Model Sizes
Offers models in various sizes such as 3B, 8B, 14B, 32B, and 70B, with plans to release even larger models soon, catering to different computational needs and use cases.
Dual Functionality
Each model can operate in standard mode for direct tasks or in reasoning mode for tasks requiring deeper analysis and self-reflection, similar to reasoning models.