wav2vec2-large-xlsr-facebook-1b-words-phoneme-exp-1-v17

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

• Loss: 0.4330
• Per Hf Metric: 0.0660
• Per Avg: 0.0887

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per Hf Metric	Per Avg
4.214	1.0	102	2.6404	1.0	1.0
2.2959	2.0	204	1.5080	0.6936	0.6893
1.4046	3.0	306	0.8916	0.2267	0.2498
0.8131	4.0	408	0.6238	0.1059	0.1375
0.7166	5.0	510	0.7134	0.0997	0.1269
0.6686	6.0	612	0.6974	0.0792	0.1089
0.5977	7.0	714	0.4408	0.0804	0.1065
0.6328	8.0	816	0.8702	0.0740	0.1031
0.5317	9.0	918	0.7305	0.0780	0.1073
0.4816	10.0	1020	0.7411	0.0732	0.0994
0.5379	11.0	1122	0.6265	0.0687	0.0930
0.4539	12.0	1224	0.8386	0.0827	0.1100
0.4151	13.0	1326	0.5374	0.0644	0.0873
0.3461	14.0	1428	0.7327	0.0687	0.0928
0.5253	15.0	1530	0.4329	0.0671	0.0900
0.3975	16.0	1632	1.0215	0.0683	0.0950
0.3331	17.0	1734	0.9336	0.0649	0.0954
0.4153	18.0	1836	1.0176	0.0718	0.1009
0.3062	19.0	1938	0.8331	0.0668	0.0939
0.3671	20.0	2040	0.6999	0.0723	0.1033
0.2515	21.0	2142	0.7128	0.0687	0.0975
0.2466	22.0	2244	0.8877	0.0756	0.1049
0.3599	23.0	2346	0.8683	0.0724	0.1034
0.2695	24.0	2448	0.9040	0.0676	0.0978
0.2305	25.0	2550	0.8124	0.0692	0.0978

Framework versions

• Transformers 4.55.2
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4