rotating-head-gp-gpt2-medium-wikitext

This model is a fine-tuned version of on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.1692
• Accuracy: 0.4228
• Perplexity: 23.7877
• Bleu: 0.1332

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 7

Training results

Training Loss	Epoch	Step	Accuracy	Bleu	Validation Loss	Perplexity
5.9062	0.2806	500	0.2234	0.0493	5.7470	313.2463
4.8598	0.5612	1000	0.2811	0.0698	4.7428	114.7554
4.3025	0.8418	1500	0.3170	0.0834	4.2329	68.9191
3.9635	1.1223	2000	0.3454	0.0932	3.9291	50.8590
3.7769	1.4029	2500	0.3636	0.1020	3.7427	42.2098
3.6738	1.6835	3000	0.3754	0.1066	3.6225	37.4295
3.5744	1.9641	3500	0.3845	0.1118	3.5325	34.2102
3.456	2.2447	4000	0.3902	0.1139	3.4704	32.1497
3.3972	2.5253	4500	0.3955	0.1230	3.4190	30.5384
3.3654	2.8058	5000	0.4007	0.1230	3.3686	29.0392
3.247	3.0864	5500	0.4043	0.1247	3.3328	28.0168
3.2403	3.3670	6000	0.4083	0.1298	3.2985	27.0714
3.2167	3.6476	6500	0.4112	0.1288	3.2693	26.2922
3.1903	3.9282	7000	0.4134	0.1305	3.2456	25.6768
3.1212	4.2088	7500	0.4161	0.1325	3.2262	25.1831
3.0816	4.4893	8000	0.4176	0.1307	3.2128	24.8480
3.0917	4.7699	8500	0.4196	0.1339	3.1985	24.4954
3.0562	5.0505	9000	0.4185	0.1326	3.2049	24.6521
3.0683	5.3311	9500	0.4195	0.1307	3.1970	24.4597
3.0502	5.6117	10000	0.4209	0.1331	3.1857	24.1847
3.0469	5.8923	10500	0.4217	0.1309	3.1790	24.0231
3.0245	6.1728	11000	3.1863	0.4205	24.1979	0.1294
3.0203	6.4534	11500	3.1783	0.4218	24.0068	0.1331
3.0265	6.7340	12000	3.1692	0.4228	23.7877	0.1332

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0