Modelpublic

Tnaot/khmer-multispeaker-whisper-v1

sourceHugging Faceapache-2.0updated 11mo agoView on Hugging Face

0likes7downloads

Model Card

khmer-multispeaker-whisper-v1

This model is a fine-tuned version of Tnaot/bopha-stt-1.3.1 on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 1.8672
—Wer: 100.0

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: 1
—evalbatchsize: 1
—seed: 3407
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 4
—optimizer: Use OptimizerNames.ADAMW8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.05
—num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.7422	2.8	5	2.2969	100.0
0.7227	5.0	10	1.9062	100.0
0.8887	7.8	15	1.7812	100.0
0.8164	10.0	20	1.7578	100.0
0.4126	12.8	25	1.7969	100.0
0.3809	15.0	30	1.8359	100.0
0.2589	17.8	35	1.8594	100.0
0.2139	20.0	40	1.875	100.0
0.1386	22.8	45	1.8672	100.0
0.1357	25.0	50	1.8438	100.0
0.1028	27.8	55	1.8672	100.0
0.1299	30.0	60	1.8672	100.0
0.0765	32.8	65	1.8672	100.0
0.042	35.0	70	1.8672	100.0
0.0871	37.8	75	1.8594	100.0
0.1426	40.0	80	1.8672	100.0
0.0594	42.8	85	1.8672	100.0
0.1416	45.0	90	1.8672	100.0
0.0844	47.8	95	1.8672	100.0
0.0391	50.0	100	1.8672	100.0

Framework versions

—Transformers 4.57.1
—Pytorch 2.8.0+cu128
—Datasets 3.6.0
—Tokenizers 0.22.1