Canada AI Startups Cohere has released a new product called "Command A"of AI ModelsThe model focuses on lightweight applications and claims to require only two NVIDIA A100 or H100 GPUs for easy deployment, claiming that "performance is comparable to GPT-4o" and that it achieves "maximum performance with minimum hardware".
Cohere said Command A is designed specifically for small and medium-sized business environments.It supports 256k context lengths and 23 languages.For comparison, other competitors' "similar models" require 32 GPUs to deploy.
In the performance test, theCommand A can output up to 156 tokens per second.The Command A is also a very good performer in benchmarks of command tracing, SQL, agent and tool tasks. Command A also excels in benchmarks for command tracing, SQL, agent programs, and utility tasks.
Citing performance data, Cohere claims that the industry's large language models can have serious latency problems when outputting results if they are "oversized"; if you just want to get to the right answer quickly, Command A is a relatively good choice.
Cohere has now published the corresponding Command A on the Hugging Face platform (Click here to visit), open for use by academics, and will be available on other cloud service platforms in the future.