wav2vec2-large-xls-r-300m-maltese

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - MT dataset. It achieves the following results on the evaluation set:

• Loss: 0.2994
• Wer: 0.2781

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: 7e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1800
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0174	9.01	1000	3.0552	1.0
1.0446	18.02	2000	0.6708	0.7577
0.7995	27.03	3000	0.4202	0.4770
0.6978	36.04	4000	0.3054	0.3494
0.6189	45.05	5000	0.2878	0.3154
0.5667	54.05	6000	0.3114	0.3286
0.5173	63.06	7000	0.3085	0.3021
0.4682	72.07	8000	0.3058	0.2969
0.451	81.08	9000	0.3146	0.2907
0.4213	90.09	10000	0.3030	0.2881
0.4005	99.1	11000	0.3001	0.2789

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0

Evaluation Script

!python eval.py
--model_id DrishtiSharma/wav2vec2-large-xls-r-300m-maltese
--dataset mozilla-foundation/common_voice_8_0 --config mt --split test --log_outputs