Intel Announces Day-One Support for Meta's New Model
It has been announced that hardware acceleration support is provided from day one on Intel processor and graphics platforms for Meta's massive language and coding model.
Intel has announced that it has begun offering comprehensive day-one hardware support for the 816-billion-parameter open-weight Muse Spark Open model developed by Meta.
Key Features of the New Model
Officially released by Meta, Muse Spark Open stands out as a multimodal model with 816 billion parameters, specifically trained for coding-oriented workflows. The system is distinguished by its one-million-token context window and advanced comprehension capabilities.
Comprehensive Integration on Intel Hardware
Intel announced that it is providing day-one support for the new model on Intel Xeon processors and the upcoming Intel GPU D-series data center graphics platforms. This integration aims to enable users to make the most efficient use of hardware-based solutions.
Hardware Acceleration and Optimization
Hardware-accelerated inference operations performed on Intel systems are supported by targeted low-bit model quantization technologies. MXFP4 quantization methods are used for Intel GPUs, and INT4 weight-focused quantization methods for Xeon processors.