wav2vec2-large-xlsr-khmer-compressed

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:

Model description

More information needed

More information needed

More information needed

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 32
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
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: 410
num_epochs: 100
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
14.3656	2.7397	200	4.4510	1.0
3.8382	5.4795	400	3.5467	1.0
3.5549	8.2192	600	3.5465	1.0
3.5274	10.9589	800	3.5229	1.0
3.5195	13.6986	1000	3.5110	1.0
3.5069	16.4384	1200	3.5079	1.0
3.497	19.1781	1400	3.4951	1.0
3.4445	21.9178	1600	3.1051	1.0
2.6018	24.6575	1800	1.6434	0.9640
1.6356	27.3973	2000	1.1678	0.8733
1.2884	30.1370	2200	0.9586	0.8061
1.129	32.8767	2400	0.8261	0.7539
0.9722	35.6164	2600	0.7473	0.7066
0.91	38.3562	2800	0.6862	0.6566
0.8078	41.0959	3000	0.6475	0.6374
0.7463	43.8356	3200	0.6087	0.6088
0.6935	46.5753	3400	0.5811	0.5921
0.6615	49.3151	3600	0.5571	0.5790
0.6561	52.0548	3800	0.5429	0.5595
0.589	54.7945	4000	0.5249	0.5487
0.5497	57.5342	4200	0.5140	0.5364
0.5325	60.2740	4400	0.4973	0.5314
0.518	63.0137	4600	0.4871	0.5209
0.5026	65.7534	4800	0.4842	0.5205
0.4896	68.4932	5000	0.4770	0.5111
0.4445	71.2329	5200	0.4726	0.5028
0.4378	73.9726	5400	0.4621	0.4943
0.4222	76.7123	5600	0.4603	0.4886
0.4207	79.4521	5800	0.4611	0.4849
0.4209	82.1918	6000	0.4613	0.4915
0.4128	84.9315	6200	0.4553	0.4808
0.3836	87.6712	6400	0.4557	0.4742
0.3811	90.4110	6600	0.4582	0.4777
0.3757	93.1507	6800	0.4545	0.4730
0.3627	95.8904	7000	0.4531	0.4697
0.3676	98.6301	7200	0.4521	0.4699