mega-small-2048-C1024-MR50-sw_minipile-tk_ema32

This model is a fine-tuned version of pszemraj/mega-small-2048-C1024-simplewiki-MR50-tk_ema32 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.7559
• Accuracy: 0.4177

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 8
• seed: 3208
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-07
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
5.0539	0.05	100	5.0404	0.2907
4.8869	0.1	200	4.6659	0.3216
4.6364	0.15	300	4.4565	0.3416
4.8682	0.2	400	4.3119	0.3557
4.3904	0.25	500	4.2410	0.3664
4.3191	0.3	600	4.1880	0.3701
4.5587	0.35	700	4.0996	0.3789
4.1517	0.4	800	4.0724	0.3839
4.1427	0.45	900	4.0177	0.3892
3.8845	0.5	1000	3.9725	0.3928
4.1478	0.55	1100	3.9080	0.4007
4.0271	0.6	1200	3.8979	0.4002
4.0132	0.65	1300	3.8647	0.4057
3.7284	0.7	1400	3.8518	0.4063
3.9346	0.75	1500	3.8178	0.4100
4.0403	0.8	1600	3.8015	0.4126
3.9726	0.85	1700	3.7916	0.4138
3.8489	0.9	1800	3.7630	0.4162
3.7117	0.95	1900	3.7745	0.4162
3.654	1.0	2000	3.7559	0.4177

Framework versions

• Transformers 4.33.1
• Pytorch 2.1.0.dev20230809+cu121
• Datasets 2.14.5
• Tokenizers 0.13.3