AI BriefingHuggingFace Blog news
The latest AI and developer news from HuggingFace Blog, with short summaries.
Visit website →Hugging Face Releases Optimum, an Optimization Toolkit
Hugging Face has released Optimum, an open-source library that supports hardware-specific performance optimization for transformer models.
HF-Graphcore IPU Optimization
Hugging Face and Graphcore have entered a partnership to provide Transformer models optimized for IPU.
DeDLOC: Internet-Based Collaborative Distributed Training
The DeDLOC algorithm has been released, which trains language models by leveraging the resources of multiple users even in slow internet network environments.
Hugging Face Adds Official Support for spaCy Models
spaCy NLP pipelines can now be shared, installed, and used via the Inference API on the Hugging Face Hub.
SageMaker-Hugging Face Deployment
A new inference solution was unveiled through the collaboration between Hugging Face and Amazon SageMaker, making it easy to deploy transformer models.
Sentence Transformers Hub Support
Hugging Face Hub now supports Sentence Transformers, offering embedding model sharing and inference features.
BERT CPU Inference Optimization and Scaling Guide
This explains optimization strategies and benchmarking methodology for maximizing BERT model inference performance in modern CPU environments.
Accelerate for PyTorch Distributed Training
It's a library that helps run PyTorch training scripts in a distributed environment with minimal code changes.
Distributed Training with Transformers and SageMaker
Through the collaboration between Hugging Face and Amazon SageMaker, distributed training of Transformers models is optimized.
How BigBird's Attention Works
This explains the working principle and efficiency of Block Sparse Attention, the core of the BigBird model.
Hugging Face and Amazon SageMaker Sign Partnership
Hugging Face and Amazon have launched optimized deep learning containers for SageMaker through a strategic partnership.
4 Key Techniques for Transformer Efficiency
This summary outlines four key approaches that reduce Transformer's computational complexity to handle long contexts.
Hugging Face RAG Doubles Retrieval Speed with Ray Integration
Hugging Face Transformers integrated Ray into its RAG models to improve document retrieval speed and fine-tuning scalability.
Hugging Face Improves TensorFlow Model Performance
The performance of TensorFlow models in Hugging Face Transformers has been improved, becoming up to 10% faster than Google's implementation.
Integrated Support for DeepSpeed and FairScale
It explains how to use the ZeRO features of DeepSpeed and FairScale in the Hugging Face Trainer to improve large-scale model training efficiency.
Hugging Face Unveils Technology to Accelerate Inference Speed by 100x
Hugging Face has unveiled a technical approach that improves Transformer inference speed by 100x through library optimization and hardware-specific compilation.
Encoder-Decoder Warm-starting Technique
This introduces a Warm-starting technique that leverages pretrained BERT or GPT2 to reduce the training cost of Encoder-Decoder models.
Hugging Face and Ray Tune Integration Supported
Ray Tune has been integrated into the Hugging Face Transformers library, enabling advanced hyperparameter optimization.
Hugging Face Releases Block Sparse Library
Hugging Face has released the pytorch_block_sparse library, which reduces model size and increases speed.
Reformer: A Transformer That Handles 500,000 Tokens
Reformer uses techniques like LSH attention to minimize memory usage while processing long sequences of up to 500,000 tokens.