When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
In today’s roundup of the latest AI news, we discuss a new Stable LM 2 language model series and recent releases by Meta and Intel.
Stability AI Introduces Two New Versions of Language Models
Stability AI has unveiled its latest Stable LM 2 language model series additions. They come in two versions. The first is a base model with 12 billion parameters. The second is an instruction-tuned variant, trained on an extensive dataset of two trillion tokens across seven languages: English, German, Spanish, French, Italian, Dutch, and Portuguese. The developers emphasized that balancing high performance, efficiency, memory requirements, and speed was possible.
Furthermore, Stability AI has also released a new version of Stable LM 2 1.6B, enhancing its conversational capabilities across the same seven languages while maintaining its low system requirements.
Stable LM 2 12B is a pair of powerful 12 billion parameter language models trained on multilingual data in English, Spanish, German, Italian, French, Portuguese, and Dutch, featuring a base and instruction-tuned model.
You can now try the model here: https://t.co/6QGjTkjgOc…
— Stability AI (@StabilityAI) April 8, 2024
The models are designed for efficient multilingual tasks. According to Stability AI, the 12B model handling tasks are typically feasible only for larger models. The instruction-tuned version excels in tool usage and function calling and is suitable for various applications, including RAG systems. Both models are available for non-commercial and commercial use with a Stability AI Membership and can be tested on Hugging Face.
The company plans for a long-context variant are also in the pipeline, further expanding the model range.
Meta Plans to Tease Two Versions of Llama 3 Language Model Next Week
Meta is gearing up to tease two versions of its highly anticipated Llama 3 large language model (LLM) next week. According to The Information, a company representative stated this at an event in London on April 9. Meta intends to release the largest multimodal Llama 3 model this summer or within the next month, per TechCrunch.
The tech giant introduced Llama 2 as an open-source platform in July 2023. A similar approach is anticipated for the upcoming release. The company remained tight-lipped about the parameter size for Llama 3, yet expectations suggest it will boast approximately 140 billion parameters. It will handle various questions, including controversial topics.
While details on the teased versions remain scarce, the eventual release is expected to include multimodal support, allowing users to input text and images for responses. This move by Meta positions Llama 3 as its answer to competitors like OpenAI’s GPT-4 and Google’s Gemini, signaling Meta’s determination to compete in the AI landscape.
Intel Introduces Gaudi 3 AI Chip Outperforming Nvidia’s H100 Processor
Intel has unveiled its latest artificial intelligence chip, Gaudi 3, to challenge Nvidia’s dominance in the semiconductor market powering AI applications. At the Intel Vision event on April 9, the company explained that by leveraging Taiwan Semiconductor Manufacturing Co’s 5nm process, Gaudi 3 outperformed Nvidia’s prior generation H100 processor.
Intel conducted tests on models such as Meta’s open-source Llama and the Falcon, backed by Abu Dhabi. The company says the new chip offers 50% faster training for large language models and faster computing for generative AI responses. That’s why Gaudi 3 could be an alternative to Nvidia and will likely be able to meet the growing demand for AI chips, which are in short supply.
The #IntelGaudi 3 #AI accelerator offers a highly competitive alternative to NVIDIA’s H100 with higher performance, increased scalability, and PyTorch integration. Explore more key product benefits. https://t.co/sXdQKjYFw0 pic.twitter.com/iJSndBQkvT
— Intel AI (@IntelAI) April 9, 2024
Intel announced that the new Gaudi 3 chips will be available to customers in the third quarter. However, the company did not disclose a price range for the chips.
Earlier in the year, Nvidia introduced its B100 and B200 GPUs, successors to the H100, boasting improved performance. These chips are slated to begin shipping later this year.
