This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

Loss: 1.4836
Wer: 0.5834

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: 0.0003
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
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: 50
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
14.6942	1.2061	100	6.5141	1.0
4.1303	2.4121	200	3.3721	1.0
3.2905	3.6182	300	3.2445	0.9991
2.5549	4.8242	400	1.6377	0.8576
1.0703	6.0242	500	1.2979	0.7720
0.7126	7.2303	600	1.2322	0.7346
0.5567	8.4364	700	1.1741	0.6869
0.4418	9.6424	800	1.1064	0.6652
0.3591	10.8485	900	1.1531	0.6607
0.3191	12.0485	1000	1.2107	0.6573
0.2572	13.2545	1100	1.2592	0.6415
0.2301	14.4606	1200	1.2445	0.6526
0.2083	15.8485	1300	1.2553	0.6483
0.176	17.0485	1400	1.2505	0.6196
0.1616	18.2545	1500	1.2371	0.6166
0.1424	19.4606	1600	1.2504	0.6157
0.1477	20.7273	1700	1.3091	0.6486
0.1558	21.9333	1800	1.4144	0.6398
0.1377	23.1333	1900	1.3280	0.6299
0.1365	24.3394	2000	1.2693	0.6254
0.1235	25.5455	2100	1.3243	0.6320
0.1208	26.7515	2200	1.3589	0.6366
0.1115	28.0364	2300	1.3636	0.6186
0.1053	29.2424	2400	1.3775	0.6262
0.1003	30.4485	2500	1.4475	0.6311
0.091	31.6545	2600	1.4514	0.6150
0.0844	32.8606	2700	1.4349	0.6028
0.0802	34.0606	2800	1.4325	0.6003
0.0781	35.2667	2900	1.4654	0.6061
0.0695	36.4727	3000	1.4644	0.6014
0.0666	37.6788	3100	1.4893	0.5918
0.0695	38.8848	3200	1.4353	0.5999
0.0631	40.0848	3300	1.4604	0.5912
0.0622	41.2909	3400	1.4656	0.5925
0.0547	42.4970	3500	1.4575	0.5901
0.0543	43.7030	3600	1.4594	0.5796
0.0535	44.9091	3700	1.4599	0.5831
0.0514	46.3394	3800	1.4609	0.5822
0.0484	47.5455	3900	1.4737	0.5806
0.044	48.7515	4000	1.4808	0.5848
0.0442	49.9576	4100	1.4836	0.5834

Framework versions

Transformers 4.48.3
Pytorch 2.5.1+cu124
Datasets 3.4.1
Tokenizers 0.21.1

sam422001/siraiki

siraiki

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

sam422001/
siraiki