Modelpublic

nilc-nlp/psst-model-4e-1s-difflib

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes6downloads

Model Card

psst-model-4e-1s-difflib

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.3447
—Wer: 0.1540
—Iu F1: 0.7087
—Iu Tp: 810
—Iu Fp: 478
—Iu Fn: 188

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: 1e-05
—trainbatchsize: 4
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 8
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 332
—training_steps: 4740

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Iu F1	Iu Tp	Iu Fp	Iu Fn
1.0729	0.4998	592	0.5149	0.2641	0.6233	402	312	174
0.9312	0.9996	1184	0.4770	0.2695	0.7009	457	271	119
0.5909	1.4989	1776	0.4784	0.2608	0.7393	380	72	196
0.5226	1.9987	2368	0.4648	0.2585	0.7708	454	148	122
0.2837	2.4981	2960	0.4956	0.2201	0.7619	448	152	128
0.2858	2.9979	3552	0.4899	0.2193	0.7742	468	165	108
0.1001	3.4973	4144	0.5498	0.2149	0.7788	456	139	120
0.0915	3.9970	4736	0.5480	0.2154	0.7853	461	137	115
0.0915	4.0	4740	0.5480	0.2153	0.7853	461	137	115

Framework versions

—Transformers 5.6.2
—Pytorch 2.6.0+cu124
—Datasets 2.21.0
—Tokenizers 0.22.2