Inverse Square Root is a learning rate schedule 1 / $\sqrt{\max\left(n, k\right)}$ where $n$ is the current training iteration and $k$ is the number of warm-up steps. This sets a constant learning rate for the first $k$ steps, then exponentially decays the learning rate until pre-training is over.
Paper | Code | Results | Date | Stars |
---|
Task | Papers | Share |
---|---|---|
Language Modelling | 95 | 9.01% |
Question Answering | 66 | 6.26% |
Decoder | 54 | 5.12% |
Text Generation | 46 | 4.36% |
Sentence | 43 | 4.08% |
Retrieval | 31 | 2.94% |
Translation | 30 | 2.85% |
Machine Translation | 25 | 2.37% |
Natural Language Understanding | 22 | 2.09% |
Component | Type |
|
---|---|---|
🤖 No Components Found | You can add them if they exist; e.g. Mask R-CNN uses RoIAlign |