Models

Overview

Our suite of AI models boasts an impressive array of capabilities, covering a wide range of language processing tasks. These models are sourced from leading AI research labs such as AI21 Labs, Anthropic, Cohere, OpenAI, and Replicate.

• The Pulze series, our very own home-grown suite of AI models, are tailored to align with our clients’ specific needs. The Pulze-v0 particularly stands out for its capacity to identify and utilize the most suitable pre-existing model for any given prompt. These models are renowned for their rapid processing speed and hold a special place among our enterprise clientele.

• AI21 Labs models cater to diverse needs. From the robust J2-Ultra, ideal for the most complex language tasks, to the efficient J2-Light designed for simpler, quicker responses.

• AlephAlpha offers the Luminous model series. The Luminous-Supreme shines in creative text writing, the Luminous-Supreme-Control is optimized for tasks like information extraction and language simplification, the Luminous-Base-Control for tasks like classification and labeling, and the Luminous-Extended-Control further enhances this capability.

• Anthropic’s Claude series shines in delivering sophisticated dialogue and creative content generation.

• Cohere’s Command models are optimized for instruction-following conversational tasks.

• OpenAI offers a line of GPT models, each fine-tuned for specific needs – the GPT-4 with its remarkable chat optimization, and the GPT-3.5 Turbo providing cost-effective solutions. The Text-Davinci and Text-Curie series also excel in comprehensive language tasks and are the best models under certain conditions.

• Replicate makes it easy to run machine learning models in the cloud from your own code. We currently support some of their open-source models.

• HuggingFace is the platform where the machine learning community collaborates on models, datasets, and applications.

• MosaicML offers two standout models: The MPT-7B-Instruct with 6.7B parameters, trained on Databricks Dolly-15k and Anthropic datasets; and the larger MPT-30B-Instruct, boasting 30B parameters, expanded with datasets from CompetitionMath to Spider, both delivering a swift 5 RPS. Meanwhile, Meta introduces the Llama2-70B-Chat, a 70B parameter chat-optimized model with a 4096-token context, licensed under the LLAMA 2 Community License.

• Finally, GooseAI’s GPT-J and GPT-Neo series serve as powerful open-source alternatives for a variety of use-cases.

Each model is specifically designed and tailored to perform at optimal levels, keeping in mind various factors such as task complexity, processing power, response speed, cost efficiency, and output quality.