Closed zhanghanqi closed 2 years ago
Sorry for the late reply. Actually, a variation of about 0.3 in accuracy could easily appear due to randomness of code/difference of device and environment. Running the code multiple times will give you a similar accuracy as reported.
Thank you!
lr=0.0001, weight_decay=0.0001, cub embedding-size=512, optim=adamw The target accuracy rate should be 72.1. After running many experiments, it is only 71.8. Are there any tips to improve the accuracy rate to 72.1?