emilios/whisper-sm-farsipal-e5

Whisper small Greek Farsioal and El Greco

This model is a fine-tuned version of emilios/whisper-sm-el-farsipal-e4 on the mozilla-foundation/common_voice_11_0,google/fleurs el,el_gr dataset. It achieves the following results on the evaluation set:

• Loss: 0.4871
• Wer: 17.1991

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-06
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1259	2.49	1000	0.4834	18.3692
0.1002	4.49	2000	0.4604	17.8027
0.1096	6.98	3000	0.4553	17.8770
0.0885	9.46	4000	0.4551	17.9606
0.0675	11.95	5000	0.4631	17.9049
0.0675	14.44	6000	0.4619	17.9049
0.0645	16.93	7000	0.4678	17.6727
0.0535	19.41	8000	0.4685	17.6634
0.039	21.49	9000	0.4746	17.6727
0.0447	23.98	10000	0.4761	17.6634
0.0393	26.46	11000	0.4792	17.7656
0.0308	28.95	12000	0.4851	17.8678
0.0301	31.44	13000	0.4846	17.4499
0.031	33.93	14000	0.4849	17.8306
0.0263	36.41	15000	0.4880	17.6170
0.0256	38.9	16000	0.4871	17.1991
0.0236	41.39	17000	0.4883	17.2641
0.0195	43.88	18000	0.4880	17.5706
0.0193	46.36	19000	0.4993	17.7285
0.0161	48.85	20000	0.4968	17.8306

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 2.0.0.dev20221216+cu116
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2