torgo_xlsr_finetune_M05

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.2781
• Wer: 0.2436

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5164	0.84	1000	3.3225	1.0
1.535	1.68	2000	1.5429	0.7317
0.8257	2.53	3000	1.3691	0.5730
0.6012	3.37	4000	1.2522	0.4660
0.491	4.21	5000	1.1413	0.4295
0.4401	5.05	6000	1.4117	0.3744
0.3768	5.89	7000	1.3632	0.3531
0.3746	6.73	8000	1.4208	0.3557
0.3127	7.58	9000	1.3584	0.3336
0.2595	8.42	10000	0.9971	0.2878
0.2458	9.26	11000	1.2794	0.3166
0.2638	10.1	12000	1.1851	0.2861
0.2384	10.94	13000	1.2144	0.2810
0.215	11.78	14000	1.1125	0.2623
0.2095	12.63	15000	1.5495	0.3081
0.1939	13.47	16000	1.4090	0.2818
0.1757	14.31	17000	1.3261	0.2564
0.1629	15.15	18000	1.3435	0.2453
0.152	15.99	19000	1.3273	0.2479
0.1516	16.84	20000	1.2215	0.2292
0.155	17.68	21000	1.3536	0.2453
0.135	18.52	22000	1.3292	0.2470
0.1206	19.36	23000	1.2781	0.2436

Framework versions

• Transformers 4.26.1
• Pytorch 2.2.1
• Datasets 2.18.0
• Tokenizers 0.13.3