Modelpublic

pszemraj/electra-small-discriminator-zeroshot-v1.1

sourceHugging Faceapache-2.0updated 9mo agoView on Hugging Face

1likes59downloads

Model Card

electra-small-discriminator-zeroshot-v1.1-none

This model is a fine-tuned version of google/electra-small-discriminator on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.3747
—F1 Macro: 0.4125
—F1 Micro: 0.4620
—Accuracy Balanced: 0.4701
—Accuracy: 0.4620
—Precision Macro: 0.5162
—Recall Macro: 0.4701
—Precision Micro: 0.4620
—Recall Micro: 0.4620

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Datasets	mnli_m	mnli_mm	fevernli	anli_r1	anli_r2	anli_r3	wanli	lingnli	wellformedquery	rottentomatoes	amazonpolarity	imdb	yelpreviews	hatexplain	massive	banking77	emotiondair	emocontext	empathetic	agnews	yahootopics	biasframes_sex	biasframes_offensive	biasframes_intent	financialphrasebank	appreviews	hateoffensive	trueteacher	spam	wikitoxic_toxicaggregated	wikitoxic_obscene	wikitoxic_identityhate	wikitoxic_threat	wikitoxic_insult	manifesto	capsotu
Accuracy	0.853	0.861	0.838	0.583	0.592	0.588	0.709	0.787	0.603	0.75	0.863	0.808	0.879	0.432	0.497	0.391	0.546	0.607	0.234	0.801	0.562	0.77	0.639	0.628	0.629	0.861	0.37	0.502	0.814	0.744	0.798	0.786	0.767	0.778	0.096	0.462
Inference text/sec (A100, batch=64)	4180.0	4161.0	2824.0	3233.0	3243.0	3239.0	4494.0	4288.0	5222.0	4396.0	2563.0	888.0	1035.0	4326.0	5447.0	5221.0	4871.0	4971.0	2852.0	3946.0	1585.0	4274.0	4097.0	4109.0	4229.0	3468.0	4476.0	1198.0	4514.0	1360.0	1267.0	1287.0	1232.0	1314.0	3936.0	4116.0

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 5e-05
—trainbatchsize: 64
—evalbatchsize: 64
—seed: 80085
—optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.04
—num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Micro	Accuracy Balanced	Accuracy	Precision Macro	Recall Macro	Precision Micro	Recall Micro
0.4765	0.32	5000	0.5300	0.7326	0.7528	0.7329	0.7528	0.7322	0.7329	0.7528	0.7528
0.4408	0.65	10000	0.5099	0.7402	0.765	0.7359	0.765	0.7463	0.7359	0.765	0.765
0.4169	0.97	15000	0.4976	0.7473	0.7702	0.7439	0.7702	0.7517	0.7439	0.7702	0.7702
0.387	1.3	20000	0.4943	0.7525	0.7742	0.7498	0.7742	0.7559	0.7498	0.7742	0.7742
0.3905	1.62	25000	0.4931	0.7522	0.775	0.7484	0.775	0.7572	0.7484	0.775	0.775
0.4001	1.95	30000	0.4924	0.7544	0.7752	0.7524	0.7752	0.7568	0.7524	0.7752	0.7752
0.3995	2.27	35000	0.4900	0.7543	0.7758	0.7517	0.7758	0.7576	0.7517	0.7758	0.7758
0.3981	2.6	40000	0.4906	0.7529	0.7742	0.7504	0.7742	0.7558	0.7504	0.7742	0.7742
0.4232	2.92	45000	0.4904	0.7544	0.776	0.7516	0.776	0.7579	0.7516	0.776	0.776

Framework versions

—Transformers 4.33.3
—Pytorch 2.1.2+cu121
—Datasets 2.16.1
—Tokenizers 0.13.3