You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

wav2vec2-darija-tamazigh-test-1000-rows

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

More information needed

More information needed

More information needed

The following hyperparameters were used during training:

learning_rate: 1e-06
train_batch_size: 4
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 2
total_train_batch_size: 64
total_eval_batch_size: 64
optimizer: Use 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: 15

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
13.9881	1.0	15	14.7541	1.0	1.2728
14.0785	2.0	30	14.7518	1.0	1.2690
13.6211	3.0	45	14.7476	1.0	1.2686
13.6767	4.0	60	14.7417	1.0	1.2724
14.0693	5.0	75	14.7340	1.0	1.2737
14.0154	6.0	90	14.7245	1.0	1.2745
13.5417	7.0	105	14.7133	1.0	1.2639
14.1507	8.0	120	14.7003	1.0	1.2546
13.7723	9.0	135	14.6852	1.0	1.2499
13.517	10.0	150	14.6685	1.0	1.2333
13.5245	11.0	165	14.6496	1.0	1.2410
13.6348	12.0	180	14.6287	1.0	1.2367
13.5503	13.0	195	14.6060	1.0	1.2486
13.4275	14.0	210	14.5804	1.0	1.2643

Safetensors

Model size

0.3B params

Tensor type

F32

Base model

Finetuned

(796)

this model