csml_word2vec

This model is a fine-tuned version of facebook/wav2vec2-base on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.2183
• Wer: 0.1227

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: 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
• lr_scheduler_warmup_steps: 500
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.2282	1.0	100	3.1130	1.0
2.9562	2.0	200	2.9543	1.0
2.9212	3.0	300	2.9317	1.0
2.9119	4.0	400	2.9110	1.0
2.9099	5.0	500	2.9099	1.0
2.9027	6.0	600	2.9106	1.0
2.4303	7.0	700	1.9136	0.9615
0.7427	8.0	800	0.5675	0.4321
0.4794	9.0	900	0.4161	0.3183
0.3372	10.0	1000	0.3228	0.2441
0.2685	11.0	1100	0.2762	0.1916
0.2092	12.0	1200	0.2522	0.1654
0.1848	13.0	1300	0.2361	0.1564
0.1678	14.0	1400	0.2246	0.1383
0.1377	15.0	1500	0.2183	0.1337
0.122	16.0	1600	0.2206	0.1310
0.1125	17.0	1700	0.2205	0.1285
0.0938	18.0	1800	0.2219	0.1273
0.0994	19.0	1900	0.2209	0.1239
0.1	20.0	2000	0.2183	0.1227

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1