xls-r-1b-bigcgen-combined-30hrs-model

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the BIGCGEN - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.6444
• Wer: 0.6496

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: 3e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: 30.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.0703	100	3.5624	1.0
No log	0.1406	200	2.7976	1.0
No log	0.2110	300	1.7477	1.0
No log	0.2813	400	0.9556	0.9632
5.9145	0.3516	500	0.8887	0.8956
5.9145	0.4219	600	0.8703	0.9244
5.9145	0.4923	700	0.6985	0.7486
5.9145	0.5626	800	0.8053	0.8899
5.9145	0.6329	900	0.6571	0.6398
1.3395	0.7032	1000	0.6413	0.6394
1.3395	0.7736	1100	0.6008	0.5699
1.3395	0.8439	1200	0.6438	0.6576
1.3395	0.9142	1300	0.6003	0.6022
1.3395	0.9845	1400	0.7025	0.7080
1.1915	1.0549	1500	0.8043	0.8568
1.1915	1.1252	1600	0.6106	0.6685
1.1915	1.1955	1700	0.6003	0.6195
1.1915	1.2658	1800	0.6331	0.6743
1.1915	1.3361	1900	0.7777	0.8180
1.0769	1.4065	2000	0.5906	0.6095
1.0769	1.4768	2100	0.6689	0.7242
1.0769	1.5471	2200	0.6466	0.6489
1.0769	1.6174	2300	0.5825	0.6113
1.0769	1.6878	2400	0.7114	0.7517
1.0964	1.7581	2500	0.8009	0.7965
1.0964	1.8284	2600	0.6260	0.6364
1.0964	1.8987	2700	0.5530	0.5519
1.0964	1.9691	2800	0.6974	0.7149
1.0964	2.0394	2900	0.5647	0.5955
1.0268	2.1097	3000	0.5556	0.5488
1.0268	2.1800	3100	0.6444	0.6502

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0