dalyaa/phi2-QA-darebah-new-2400-phitrain

phi-2

This model is a fine-tuned version of microsoftl on the dalyaa/darebah2400 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8341

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2.5e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• training_steps: 2500

Training results

Training Loss	Epoch	Step	Validation Loss
1.0567	0.4	100	1.0257
0.9463	0.8	200	0.9571
0.8397	1.2	300	0.9297
0.7876	1.6	400	0.9042
0.7484	2.0	500	0.8973
0.7301	2.4	600	0.8804
0.7023	2.8	700	0.8712
0.6604	3.2	800	0.8652
0.6727	3.6	900	0.8578
0.6542	4.0	1000	0.8549
0.6474	4.4	1100	0.8533
0.6208	4.8	1200	0.8503
0.6022	5.2	1300	0.8429
0.5997	5.6	1400	0.8488
0.6399	6.0	1500	0.8389
0.6273	6.4	1600	0.8410
0.5854	6.8	1700	0.8422
0.6062	7.2	1800	0.8360
0.5958	7.6	1900	0.8363
0.5933	8.0	2000	0.8403
0.5905	8.4	2100	0.8388
0.5604	8.8	2200	0.8366
0.572	9.2	2300	0.8349
0.5764	9.6	2400	0.8365
0.5926	10.0	2500	0.8341

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1