GLiNER icon indicating copy to clipboard operation
GLiNER copied to clipboard

optimize training

Open AleKitsune98 opened this issue 1 year ago • 9 comments

How to specify parameters like warmup_ratio, dropout in finetuning?

AleKitsune98 avatar May 17 '24 16:05 AleKitsune98

Check this file for the details.

mkmohangb avatar May 18 '24 07:05 mkmohangb

thank you, I have tried the script but when I come to the val step it says "Object of type SimpleNameSpace is not JSONSerializable"

AleKitsune98 avatar May 18 '24 07:05 AleKitsune98

Hi @AleKitsune98, can you send me a screenshot of the error?

urchade avatar May 18 '24 11:05 urchade

I was able to reproduce the issue and created a pull request with the fix.

mkmohangb avatar May 20 '24 10:05 mkmohangb

Thank you very much, now I am getting division by zero during evaluation. I'm using training with huggingface model in config.yaml

AleKitsune98 avatar May 20 '24 17:05 AleKitsune98

Have you downloaded the correct dataset ? @AleKitsune98 Here is the link: # "NER_datasets": val data from the paper can be obtained from "https://drive.google.com/file/d/1T-5IbocGka35I7X3CE6yKe5N_Xg2lVKT/view"

urchade avatar May 20 '24 18:05 urchade

I created a custom dataset following the format of sample_data.json

AleKitsune98 avatar May 20 '24 19:05 AleKitsune98

The training evaluation is intended for benchmark comparison on the data I linked before, but you can modify the code by looking at this link if you have JSON evaluation data.

urchade avatar May 20 '24 19:05 urchade

Thank you very much. I will take a look at it. I appreciate your work!

AleKitsune98 avatar May 20 '24 20:05 AleKitsune98