Smaller, faster R-series version of Cohere wins over RAG, guesses in 23 languages

Confirming its intention to support a wide range of business use cases – including those that do not require expensive, intensive applications. Major languages (LLMs) – Introduction to AI Here has released the Command R7B, the smallest and fastest in its R series.

The R7B command is designed to facilitate fast recording and playback and uses retrieval-augmented generation (RAG) for accuracy. The model is 128K long and supports 23 languages. It outperforms other heavy-duty models in its category — Google’s Gemma, Meta’s Llama, Mistral’s Ministral — in tasks including math and text, Cohere says.

“This product was created for manufacturers and businesses that need to optimize the speed, cost and inventory of the equipment they use,” said Cohere co-founder and CEO Aidan Gomez. he writes in a blog post announcing a new model.

Best in maths, writing, RAG

Cohere has been focusing on businesses and their special events. The company launched Command-R in March they are strong Rule R+ in April, and it has changed throughout the year support speed and power. It has teased the Command R7B as the “last” model in its R series, and says it will release heavy prototypes to the AI ​​research team.

Cohere noted that the most important part of developing Command R7B was improving math, reasoning, coding and interpretation. The company seems to have done well in those areas, with a new sub-brand on top of them HuggingFace Open LLM Leaderboard against similar open heavy varieties including Gemma 2 9B, Ministral 8B and Llama 3.1 8B.

In addition, the smallest model in the R group performs better than competing models in places including AI assistants, using tools and RAG, which helps to improve accuracy when setting outputs in external data. Cohere says the Command R7B excels in the areas of negotiation including on-the-job technology and enterprise risk management (ERM) support; current policy; press office and customer service; HR questions; and in short. Cohere also said the model is “excellent” in retrieving and manipulating statistical information in the economy.

All told, the R7B Rule ranked first, on average, in key benchmarks including instructional compliance evaluation (IFeval); big bench (BBH); Q&A (GPQA) for graduates; Different soft thinking (MuSR); and a broad understanding of many languages (MMLU).

Removing unnecessary call functions

The R7B command can use tools including search engines, APIs and vector databases to extend its functionality. Cohere explains that using this modeling tool performs strongly against competitors in the Berkeley Function-Calling Leaderboard, which measures model accuracy in calling functions (interacting with data and external systems).

Gomez points out that this ensures his work in the “real world, a diverse and dynamic environment” and removes the need for unnecessary work. This would be a good choice for building “fast and capable” AI agents. For example, Cohere points out, by working as an additional search engine on the Internet, the Command R7B can break down complex queries into smaller ones, and perform better with high-level concepts and returns.

Because it is small, the R7B Command can be deployed on low-end and consumer CPUs, GPUs and MacBooks, allowing for on-device control. The model is now available on the Cohere and HuggingFace platform. Prices are $0.0375 for 1 million input tokens and $0.15 for 1 million output tokens.

“It’s a great choice for businesses looking for a cost-effective way to manage their internal documents and data,” writes Gomez.