Listen to this story
|
After the release of Meta’s Llama 3, Databricks has announced its partnership with Meta to make Llama 3 available to enterprises of all sizes through a fully managed API on the Databricks platform.
Llama 3 comes in two initial versions – an 8 billion parameter model and a 70 billion parameter model. These models outperform existing open-source LLMs such as Gemma and Mistral, and even match the performance of top proprietary models like GPT-4 and Gemini on many benchmarks.
Databricks Model Serving will provide instant access to Meta Llama 3 via Foundation Model APIs, allowing users to experiment with, switch between, and deploy foundation models across all cloud providers easily. Customers can try Meta Llama 3 directly from the Databricks AI Playground in the coming days.
Naveen Rao, VP of generative AI at Meta, highlighted the significance of this release. “We are leveraging the awesome innovations Meta baked into their latest models with Databricks. The capabilities are really game changing, and we provide the ability to customize, manage, and deploy at scale securely. This is also a great way to enable generative AI on your data with our RAG capabilities built on top of these models,” he said.
“Meta Llama 3, which will be rolling out regionally in the next few days, can be accessed through the same unified API on Databricks Model Serving that thousands of enterprises are already using to access other open and external models,” the company said through its blog post.
Apart from Databricks, Llama 3 models are now also rolling out on Amazon SageMaker, Google Cloud, Hugging Face, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM, and Snowflake. Additionally, the models will be compatible with hardware platforms provided by AMD, AWS, Dell, Intel, NVIDIA, and Qualcomm.
Last month, Databricks also announced the launch of DBRX. This new model outperforms SOTA open-source models like Llama 2 70B, Mixtral-8x7B and Grok-1 across various benchmarks, including language understanding (MMLU), programming (Human Eval) and Math (GSM 8K).