taskydata/deberta-v3-base_10xp3_10xc4_128

Hyperparameters:

• learning rate: 2e-5
• weight decay: 0.01
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• gradient_accumulation_steps:1
• eval steps: 5000
• max_length: 128
• num_epochs: 3

Dataset version:

• “craffel/tasky_or_not”, “10xp3_10xc4”, “15f88c8”

Checkpoint:

• 10000 steps

Results on Validation set:

Step	Training Loss	Validation Loss	Accuracy	Precision	Recall	F1
5000	0.036400	0.266518	0.926913	0.999662	0.916934	0.956513
10000	0.022500	0.222881	0.952443	0.999494	0.946227	0.972132
15000	0.016600	0.634102	0.882638	0.999789	0.866301	0.928270
20000	0.011300	1.138026	0.849013	0.999796	0.827928	0.905781
25000	0.010300	0.623522	0.895619	0.999728	0.881166	0.936710
30000	0.006300	0.776632	0.879492	0.999804	0.862697	0.926204
35000	0.000500	0.704599	0.899149	0.999698	0.885220	0.938982