distilbert_sa_GLUE_Experiment_logit_kd_pretrain_stsb

This model is a fine-tuned version of gokuls/distilbert_sa_pre-training-complete on the GLUE STSB dataset. It achieves the following results on the evaluation set:

• Loss: 0.2956
• Pearson: 0.8628
• Spearmanr: 0.8598
• Combined Score: 0.8613

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: 256
• eval_batch_size: 256
• seed: 10
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Pearson	Spearmanr	Combined Score
1.8052	1.0	23	1.2355	0.2355	0.2368	0.2361
0.8213	2.0	46	0.5382	0.7641	0.7680	0.7660
0.4216	3.0	69	0.3781	0.8401	0.8371	0.8386
0.2818	4.0	92	0.3205	0.8486	0.8448	0.8467
0.1988	5.0	115	0.3463	0.8489	0.8498	0.8494
0.1583	6.0	138	0.3100	0.8574	0.8539	0.8557
0.1249	7.0	161	0.3252	0.8556	0.8527	0.8542
0.111	8.0	184	0.3495	0.8529	0.8497	0.8513
0.099	9.0	207	0.2956	0.8628	0.8598	0.8613
0.0825	10.0	230	0.3060	0.8587	0.8555	0.8571
0.0682	11.0	253	0.2985	0.8584	0.8564	0.8574
0.0671	12.0	276	0.3001	0.8568	0.8538	0.8553
0.0555	13.0	299	0.3107	0.8600	0.8568	0.8584
0.0575	14.0	322	0.3221	0.8592	0.8560	0.8576

Framework versions

• Transformers 4.26.0
• Pytorch 1.14.0a0+410ce96
• Datasets 2.9.0
• Tokenizers 0.13.2