Command A+

Overview

Cohere Labs Command A is an open weights research release of a 111 billion parameter model developed by Cohere and Cohere Labs. It is optimized for demanding enterprises that require fast, secure, and high-quality AI, and is deployable on just two GPUs. The model supports a context length of 256K, but it is configured in Hugging Face for 128K, a value that can be updated in the configuration if needed. The model is an auto-regressive language model that uses an optimized transformer architecture. After pretraining, it uses supervised fine-tuning (SFT) and preference training to align model behavior to human preferences for helpfulness and safety. Models input text only and generate text only. The model has been trained on 23 languages, including English, French, Spanish, Italian, German, Portuguese, Japanese, Korean, Arabic, and Chinese. By default, Command A is configured as a conversational model, replying in a conversational fashion and using Markdown as well as LaTeX where appropriate. It has been trained specifically for tasks like the final step of Retrieval Augmented Generation (RAG), and with conversational tool use capabilities that allow it to interact with external tools like APIs, databases, or search engines. It can be configured with two safety modes, contextual mode and strict mode, to set guardrails suitable to user needs.