wav2vec2_300m

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

• Loss: -1.7265
• Cer: 2.8268
• Wer: 1.0

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: 5e-05
• 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: 100
• num_epochs: 300
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
-2.2058	16.6667	100	-1.7563	2.7979	1.0
-2.2553	33.3333	200	-1.7904	2.8052	1.0
-2.3332	50.0	300	-1.8135	2.8289	1.0
-2.3625	66.6667	400	-1.7726	2.7791	1.0
-2.3912	83.3333	500	-1.7941	2.7467	1.0
-2.4162	100.0	600	-1.7979	2.7745	1.0
-2.4141	116.6667	700	-1.7517	2.7428	1.0
-2.4383	133.3333	800	-1.8364	2.9398	1.0
-2.4441	150.0	900	-1.7691	2.8856	1.0
-2.4518	166.6667	1000	-1.7509	2.7699	1.0
-2.4653	183.3333	1100	-1.7286	2.8175	1.0
-2.4651	200.0	1200	-1.7521	2.7321	1.0
-2.4619	216.6667	1300	-1.6761	2.7655	1.0
-2.4649	233.3333	1400	-1.6887	2.7880	1.0
-2.4863	250.0	1500	-1.7591	2.7745	1.0
-2.472	266.6667	1600	-1.7198	2.7464	1.0
-2.477	283.3333	1700	-1.7385	2.8616	1.0
-2.4708	300.0	1800	-1.7265	2.8268	1.0

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1