Intel, Habana Labs, and Hugging Face have been trying to make an innovative society, not certainly with a floating vehicle; rather, it knurls a lot of uncertainty with possibilities.
Over the past few years, both Innovators have passed through various setbacks and massive outbreaks to improve efficiencies and tune up the key for adopting artificial intelligence with the help of open-source projects, integrated developer experiences, and scientific research. The mission keeps them going to gain a not stunted growth and resulted in crucial advantages for building and practicing high-quality transformer models.
Transformer models put through an advanced performance on a wide range of machine and deep learning tasks such as natural language processing (NLP), computer vision (CV), speech, and others.
Distributed Fine-Tuning on Intel Xeon Platform:
Data scientists are befuddled about the distribution of the training database when pieces of training on a single node CPU become slow. It makes them reply on distributed training where clustered servers keep a spare of the model, tune I with a subset of the training dataset, and exchange results across nodes through the Intel® with one API Collective Communications Library to revamp it to a final model faster.
Optimum Developer Experience
On the other hand, here comes the open-source library dubbed Optimum, which simplifies the transformer acceleration across a growing range of training and inference communication devices. It was Developed by Hugging Face. With the in-built optimization techniques and self-made scripts, baby step takers can use Optimum as a solution to leading sources.
Accelerated Training with Habana Gaudi
Habana Labs and Hugging Face are teaming up to make it much easier and speedier to train large-scale, high-quality transformer models. The amalgamation of Habana’s SynapseAI® software suite with the Hugging Face Optimum-Habana open source library switches on the data scientists and machine learning artificer to boost the transformer deep learning training with Habana processors – Gaudi and Gaudi2 – with an available line of code.
Few-shot Learning in Production
Intel Labs, Hugging Face, and UKP Lab jointly wrapped off SetFit, an ultra-simple framework for few-shot fine-tuning of Sentence Transformers. Few-shot learning with multi-core trained language models has upsurged as a most prominent solution to a real data scientist challenge: dealing with data with few to no labels.
Open source projects, integrated developer experiences, and scientific research are the few available ways for Intel to engage with the ecosystem and contribute to reducing the cost of AI.
