Qualcomm Technologies and Hugging Face share a common goal: make it easier for innovators to harness the power of machine intelligence. One step toward achieving this is Hugging Face will host optimized transformers models for Qualcomm Cloud AI 100 systems to deploy at production scale, with minimum coding complexity.

What is Qualcomm Cloud AI 100?

Qualcomm Cloud AI 100 is Qualcomm Technologies’ first-generation AI inference accelerator for datacenter and edge compute business, representing the company’s investments in machine learning and utilizing its expertise in the area from the consumer mobile SoC world as well as bringing it to the enterprise segment.

Qualcomm Technologies is planning for three different form factors in terms of commercial solutions:

• a full-blown PCIe form-factor accelerator card, which is meant to achieve up to an astounding 400TOPs inference performance at 75W TDP
• a DM.2 at 25W TDP, which is gaining popularity in the enterprise segment and is akin to two M.2 connectors next to each other
• a DM.2e at 15W TDP, representing a smaller and lower-power thermal envelope form-factor for the edge compute segment

Democratizing transformers for cloud and edge with Qualcomm Cloud AI 100

Hugging Face plays host to hundreds of transformers, which applies lessons learned in NLP to computer vision. The transformers library is downloaded an average of 2 million times every month and demand is growing. With a user base of more than 50,000 developers, Hugging Face has seen one of the fastest adoptions of an open-source project.

Now, with its extended collaboration with Qualcomm Technologies, Hugging Face is connecting the ultimate transformer toolset with today's most power-efficient AI hardware, enabling better TCO for inferencing in the cloud as well as the edge.

Hugging Face plans to integrate Optimum, a new open-source library and toolkit for Qualcomm Cloud AI 100, enabling developers access to hardware-optimize models certified by Hugging Face. As part of the collaboration, teams from Hugging Face and Qualcomm Technologies will optimize versions of BERT-Large, DistilBERT. RoBERTa, Albert, XML-RoBERTa, BART, T5, and GPT2.

This will also enable Qualcomm Cloud AI 100 devices to unlock performance advantages — through the Hugging Face platform — with its elegant simplicity and a superlative range of models that democratizes transformers and NLP at the cloud and the edge.

Qualcomm Cloud AI is a product of Qualcomm Technologies, Inc. and/or its subsidiaries.