whisper-large-v3FT-Dahnon-arabic

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: 1.3779
• Wer: 50.7533

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
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• gradient_accumulation_steps: 16
• total_train_batch_size: 48
• total_eval_batch_size: 24
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.9222	1.0	12	1.8887	119.0207
1.9246	2.0	24	1.8887	119.1149
1.9071	3.0	36	1.8311	118.6441
1.76	4.0	48	1.6348	108.4746
1.5037	5.0	60	1.4199	78.4369
1.2721	6.0	72	1.2490	77.1186
1.0506	7.0	84	1.1074	69.3974
0.8575	8.0	96	1.0098	67.2316
0.7006	9.0	108	0.9531	63.6535
0.5644	10.0	120	0.9194	61.0169
0.443	11.0	132	0.9248	61.2053
0.3326	12.0	144	0.9312	61.1111
0.2412	13.0	156	0.9585	59.4162
0.1673	14.0	168	1.0117	59.7928
0.1111	15.0	180	1.0518	61.4878
0.0737	16.0	192	1.125	58.8512
0.0476	17.0	204	1.1680	59.5104
0.0333	18.0	216	1.1963	62.0527
0.0256	19.0	228	1.2344	63.0885
0.021	20.0	240	1.2129	58.6629
0.0173	21.0	252	1.2812	58.4746
0.0119	22.0	264	1.2930	58.5687
0.0114	23.0	276	1.2695	59.3220
0.0125	24.0	288	1.2969	59.6045
0.0111	25.0	300	1.2812	58.3804
0.011	26.0	312	1.3184	59.0395
0.0112	27.0	324	1.25	58.3804
0.0125	28.0	336	1.2549	60.7345
0.0099	29.0	348	1.3096	59.6987
0.0111	30.0	360	1.2402	58.2863
0.0125	31.0	372	1.2598	58.9454
0.0093	32.0	384	1.3145	57.0621
0.0092	33.0	396	1.3096	60.0753
0.0098	34.0	408	1.2881	58.9454
0.0132	35.0	420	1.2666	60.4520
0.015	36.0	432	1.25	60.4520
0.0127	37.0	444	1.2148	57.1563
0.0141	38.0	456	1.2100	55.5556
0.0109	39.0	468	1.2617	58.2863
0.0139	40.0	480	1.1943	57.4388
0.0149	41.0	492	1.2402	57.6271
0.0093	42.0	504	1.2803	58.1921
0.0084	43.0	516	1.2373	56.1205
0.0085	44.0	528	1.2285	54.6139
0.0063	45.0	540	1.2598	58.6629
0.0076	46.0	552	1.2725	57.4388
0.0105	47.0	564	1.1836	56.4972
0.0092	48.0	576	1.2256	51.7891
0.0057	49.0	588	1.2197	55.5556
0.0045	50.0	600	1.2490	53.0132
0.0072	51.0	612	1.1709	53.0132
0.0043	52.0	624	1.2178	53.2015
0.003	53.0	636	1.2158	52.1657
0.002	54.0	648	1.2549	51.9774
0.0029	55.0	660	1.2344	54.3315
0.0021	56.0	672	1.2021	51.3183
0.0013	57.0	684	1.2520	53.2015
0.0008	58.0	696	1.2676	51.7891
0.0005	59.0	708	1.2910	51.9774
0.0005	60.0	720	1.2998	50.7533
0.0004	61.0	732	1.3115	51.3183
0.0003	62.0	744	1.3184	51.9774
0.0003	63.0	756	1.3223	51.6008
0.0003	64.0	768	1.3281	51.7891
0.0003	65.0	780	1.3330	51.6949
0.0003	66.0	792	1.3369	51.5066
0.0003	67.0	804	1.3418	51.2241
0.0002	68.0	816	1.3447	51.5066
0.0002	69.0	828	1.3477	51.3183
0.0002	70.0	840	1.3506	51.4124
0.0002	71.0	852	1.3535	51.1299
0.0002	72.0	864	1.3555	51.2241
0.0002	73.0	876	1.3584	51.2241
0.0002	74.0	888	1.3604	50.9416
0.0002	75.0	900	1.3623	50.4708
0.0002	76.0	912	1.3633	51.0358
0.0002	77.0	924	1.3652	51.1299
0.0002	78.0	936	1.3682	51.0358
0.0002	79.0	948	1.3682	51.0358
0.0002	80.0	960	1.3701	51.3183
0.0002	81.0	972	1.3711	51.2241
0.0002	82.0	984	1.3721	51.0358
0.0002	83.0	996	1.3730	51.1299
0.0002	84.0	1008	1.3740	50.8475
0.0002	85.0	1020	1.375	51.0358
0.0002	86.0	1032	1.3760	51.1299
0.0002	87.0	1044	1.3770	51.1299
0.0002	88.0	1056	1.3760	51.1299
0.0002	89.0	1068	1.3770	51.2241
0.0002	90.0	1080	1.3779	51.3183
0.0002	91.0	1092	1.3779	51.2241
0.0002	91.7111	1100	1.3779	50.7533

Framework versions

• Transformers 4.50.3
• Pytorch 2.4.1
• Datasets 3.0.0
• Tokenizers 0.21.1