chanind / frame-semantic-transformer

Frame Semantic Parser based on T5 and FrameNet
https://chanind.github.io/frame-semantic-transformer
MIT License
51 stars 10 forks source link

What's the performance of T5-small on FN 1.7 dev/test? #5

Closed Zce1112zslx closed 2 years ago

chanind commented 2 years ago

The T5-small performance on FN 1.7 dev/test as compared with open-sesame and t5-base is below:

Task Sesame F1 (dev/test) T5-small Model F1 (dev/test) T5-base Model F1 (dev/test)
Trigger identification 0.80 / 0.73 0.74 / 0.70 0.78 / 0.71
Frame classification 0.90 / 0.87 0.83 / 0.81 0.89 / 0.87
Argument extraction 0.61 / 0.61 0.68 / 0.70 0.74 / 0.72

This info is also in the readme

Zce1112zslx commented 2 years ago

thx