wikitext_roberta-base

This model is a fine-tuned version of roberta-base on the wikitext wikitext-2-raw-v1 dataset. It achieves the following results on the evaluation set:

• Loss: 1.2143
• Accuracy: 0.7371

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.4175	0.99	37	1.3355	0.7194
1.438	1.99	74	1.2953	0.7249
1.4363	2.99	111	1.2759	0.7276
1.3391	3.99	148	1.2904	0.7252
1.3741	4.99	185	1.2621	0.7290
1.2771	5.99	222	1.2312	0.7353
1.287	6.99	259	1.2542	0.7289
1.29	7.99	296	1.2290	0.7345
1.2948	8.99	333	1.2537	0.7286
1.2741	9.99	370	1.2199	0.7354
1.2342	10.99	407	1.2520	0.7309
1.2199	11.99	444	1.2738	0.7260
1.206	12.99	481	1.2286	0.7335
1.221	13.99	518	1.2421	0.7327
1.2062	14.99	555	1.2402	0.7328
1.2305	15.99	592	1.2473	0.7308
1.2426	16.99	629	1.2250	0.7318
1.2096	17.99	666	1.2186	0.7353
1.1961	18.99	703	1.2214	0.7361
1.2136	19.99	740	1.2506	0.7311

Framework versions

• Transformers 4.21.0.dev0
• Pytorch 1.11.0+cu113
• Datasets 2.3.3.dev0
• Tokenizers 0.12.1