taskydata/deberta-v3-base_10xp3_10xc4_512

Hyperparameters:

• learning rate: 2e-5
• weight decay: 0.01
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 8
• gradient_accumulation_steps:1
• eval steps: 6000
• max_length: 512
• num_epochs: 2

Dataset version:

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

Checkpoint:

• 48000 steps

Results on Validation set:

Step	Training Loss	Validation Loss	Accuracy	Precision	Recall	F1
6000	0.031900	0.163412	0.982194	0.999211	0.980462	0.989748
12000	0.014700	0.106132	0.976666	0.999639	0.973733	0.986516
18000	0.010700	0.043012	0.995743	0.999223	0.995918	0.997568
24000	0.007400	0.095047	0.984724	0.999857	0.982714	0.991211
30000	0.004100	0.087274	0.990400	0.999829	0.989217	0.994495
36000	0.003100	0.162909	0.981972	1.000000	0.979434	0.989610
42000	0.002200	0.148721	0.980454	0.999986	0.977717	0.988726
48000	0.001000	0.094455	0.990437	0.999943	0.989147	0.994516