Cloud and open source AI provider H2O.ai on Thursday introduced a small language model that the vendor says offers users tighter privacy and data controls than large language models.
H2O-Danube-1.8B is an open source natural language model with 1.8 billion parameters.
The small language model can run on portable devices such as smartphones, laptops and desktop computers.
It was trained on 1 trillion tokens from different web sources and techniques refined from models such as Llama 2 and Mistral, H20.ai reported. It is also released under the Apache 2.0 Open Source License.
H20.ai also introduced H20-Danube-1.8B-Chat, a chatbot version of the model refined for conversational applications, as well as the new small language model.
The language model and chat version are now available on Cuddly face.
H2O-Danube-1.8B comes after the supplier introduced H2OGPTe in January. H2OGPTe is an enterprise generative AI platform that can retrieve internal data and host LLMs privately.
Small language models
This also comes as the generative AI market is seeing increasing small language models (SLM) emerging.
More recently, tech giant Google introduced its open source Gemma models featuring 2 billion parameters. Additionally, French startup Mistral AI is popular with cloud providers like AWS and Microsoft who have integrated the provider’s 7B model into their generative AI stack.
The growth of SLMs is due to several factors, according to Kashyap Kompella, an analyst at RPA2AI Research.
On the one hand, they are more affordable. Businesses can run them locally and on consumer devices without expensive GPUs.
Additionally, open source SLMs give companies greater control over how data is processed, Kashyap said.
For example, an LLM from a vendor like OpenAI can be used via an API, and the company’s data will go to OpenAI.
But a company-hosted open source model allows for greater control.
“The tradeoff here is better performance from commercial providers versus greater flexibility and control from open source models,” Kompella said. “SLMs may not match the performance of LLMs, but there are scenarios in which they are necessary. Small language models like H2O-Danube can be a useful addition to a company’s AI toolkit. “
Show what is possible
For H2O, the Danube is an example of what is possible for others.
“We demonstrate that you can own… your own LLM on your own content”, CEO and co-founder Sri Ambati said.
Small language models like H2O-Danube can be a useful addition to a company’s AI toolkit.
Kashyap KompellaAnalyst, RPA2AI Research
Even if a smaller model will not be as efficient as OpenAI’s GPT-4for example, it can serve as a safety guardrail for the larger model and can help users extend LLM training.
H2O-Danube was formed by a few people and cost between $30,000 and $40,000 to build, Ambati said.
“We’re demonstrating that you don’t have to be a big boys club to build this model by raising tens of hundreds of millions of dollars,” he said.
However, the cost of building a model is relative, said Arun Chandrasekaran, an analyst at Gartner.
Sometimes suppliers refine an existing model to create a new model, so the cost can be lower than creating a new model from scratch, he added.
Too many models
The emergence of new models like the H20-Danube can also be disruptive for businesses, Chandrasekaran continued. “Today’s customers are overwhelmed with model choices because there is literally a new model coming out every week, if not every day.”
This led to a new category of tools called model routers. Template routers help businesses automate template selection based on applications and key user outcomes.
The profusion of different models makes it difficult for a provider like H2O.ai, Chandrasekaran said.
Even though H2O.ai has gained popularity among open source users, it still risks being drowned out by vendors such as Google.
“How they can, for example, gain new customers by improving their branding and explaining a clear value proposition for their models – I think that’s probably going to be a challenge,” he said.
Vendors like H2O.ai must then try to compete with much larger AI vendors like OpenAI, Anthropic, and Google by ensuring that H2O.ai models have sufficient support and adoption by the community, he added.
Esther Ajao is a news editor at TechTarget and host of podcasts covering artificial intelligence software and systems.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.Ok
