w2v-bert-2.0-luo_19_77h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the CLEAR-GLOBAL/LUO_19_77H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.2419
• Wer: 0.2906
• Cer: 0.0898

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: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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_ratio: 0.1
• training_steps: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.469	0.8446	1000	0.8816	0.6085	0.1993
0.1171	1.6892	2000	0.6389	0.4184	0.1531
0.0837	2.5338	3000	0.5226	0.3805	0.1395
0.1017	3.3784	4000	0.3857	0.3532	0.1133
0.036	4.2230	5000	0.3766	0.3457	0.1161
0.0509	5.0676	6000	0.3433	0.3408	0.1149
0.0444	5.9122	7000	0.2983	0.3082	0.0981
0.0704	6.7568	8000	0.2803	0.2972	0.0968
0.0516	7.6014	9000	0.3242	0.2932	0.1006
0.0362	8.4459	10000	0.2760	0.3047	0.0967
0.0253	9.2905	11000	0.2727	0.2782	0.0908
0.026	10.1351	12000	0.2789	0.2959	0.1049
0.0274	10.9797	13000	0.2542	0.2782	0.0922
0.0218	11.8243	14000	0.2694	0.2646	0.0904
0.0201	12.6689	15000	0.2575	0.3007	0.0922
0.0201	13.5135	16000	0.2419	0.2901	0.0896
0.0216	14.3581	17000	0.2478	0.2795	0.0933
0.0079	15.2027	18000	0.2974	0.2844	0.0890
0.0352	16.0473	19000	0.2596	0.2959	0.0930
0.0302	16.8919	20000	0.2831	0.2491	0.0849
0.0115	17.7365	21000	0.2966	0.2751	0.0920

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1