- Efficient LM series with open source training and reasoning frameworks
- Performance is comparable to OLMo while requiring 2x fewer pre-training tokens
CoreNet is a deep neural network toolkit that allows researchers and engineers to train standard and novel small and large models for a variety of tasks, including basic models (such as CLIP and LLM), object classification, object detection, and semantic segmentation.
Apple’s research work using CoreNet
Here is a list of Apple publications that use CoreNet. In addition, links to training and evaluation methods and pre-trained models can be found in the project folder. Please refer to it for more details.
OpenELM: A family of efficient language models with open training and reasoning frameworks
CatLIP: CLIP-level visual recognition accuracy, and the pre-training speed of network-scale image and text data is increased by 2.7 times
Strengthen data to multiply impact: Improve model accuracy and robustness through data set enhancement
CLIP meets model zoo experts: pseudo-supervision with visual enhancement
FastVit: A fast hybrid vision transformer using structural reparameterization
You only need bytes: a converter that operates directly on file bytes
MobileOne: An improved one-millisecond mobile backbone network
RangeAugment: Efficient online enhancement through range learning
Detachable self-attention for mobile vision transformers (MobileViTv2)
CVNets: A high-performance library for computer vision, ACM MM’22
MobileViT: Lightweight, universal and mobile device visual converter, ICLR’22
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Warehouse: https://github.com/apple/corenet
High frequency: https://huggingface.co/apple/OpenELM
Abdominal muscles: https://arxiv.org/abs/2404.14619
Video: