metadata

base_model: Umong/wav2vec2-xls-r-300m-bengali
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: wav2vec2-xls-r-300m-bengali
    results: []

wav2vec2-xls-r-300m-bengali

This model is a fine-tuned version of Umong/wav2vec2-xls-r-300m-bengali on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.1636
Wer: 0.0883

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.3076	0.16	400	1.5883	0.9394
0.8841	0.33	800	0.5188	0.5337
0.5896	0.49	1200	0.4029	0.4340
0.4964	0.66	1600	0.3429	0.3766
0.4553	0.82	2000	0.3196	0.3642
0.4222	0.99	2400	0.3004	0.3436
0.3709	1.15	2800	0.2812	0.3225
0.352	1.32	3200	0.2753	0.3124
0.3283	1.48	3600	0.2616	0.2979
0.3235	1.65	4000	0.2573	0.2944
0.3129	1.81	4400	0.2458	0.2809
0.306	1.98	4800	0.2344	0.2771
0.2701	2.14	5200	0.2318	0.2661
0.2653	2.31	5600	0.2253	0.2629
0.2626	2.47	6000	0.2186	0.2542
0.2541	2.63	6400	0.2074	0.2474
0.2235	2.8	6800	0.2102	0.2442
0.2185	2.96	7200	0.2019	0.2327
0.2061	3.13	7600	0.1994	0.2308
0.2011	3.29	8000	0.1942	0.2260
0.1986	3.46	8400	0.1867	0.2187
0.197	3.62	8800	0.1825	0.2177
0.1931	3.79	9200	0.1856	0.2153
0.1879	3.95	9600	0.1777	0.2088
0.1599	4.12	10000	0.1781	0.0968
0.153	4.28	10400	0.1738	0.0944
0.1475	4.45	10800	0.1713	0.0905
0.1448	4.61	11200	0.1683	0.0907
0.1445	4.78	11600	0.1649	0.0897
0.1423	4.94	12000	0.1636	0.0883

Framework versions

Transformers 4.33.0
Pytorch 2.0.0
Datasets 2.14.5
Tokenizers 0.13.3