 | tinygrad is a lightweight deep learning framework that aims to be the easiest to add new accelerators to support both inference and training. tinygrad supports simple basic operations, including SOTA vision models/efficientnet.py and language model models/transformer.py. tinygrad is working on supporting Apple Neural Engine and Google TPU in accel/ folder. Ultimately custom hardware will be built for tinygrad to reach the ultimate goal: an extremely fast inference/training framework. (but it’s pretty slow now) Install git clone https://github.com/g… |
#lightweight #deep #learning #framework #tinygrad