metadata

license: apache-2.0
tags:
  - automatic-speech-recognition
  - librispeech_asr
  - generated_from_trainer
model-index:
  - name: sew-mid-100k-librispeech-clean-100h-ft
    results: []

sew-mid-100k-librispeech-clean-100h-ft

This model is a fine-tuned version of asapp/sew-mid-100k on the LIBRISPEECH_ASR - CLEAN dataset. It achieves the following results on the evaluation set:

Loss: 3.9609
Wer: 1.0

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: 0.0003
train_batch_size: 4
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
total_train_batch_size: 32
total_eval_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 3.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.9558	0.11	100	2.9238	1.0
2.8978	0.22	200	2.9614	1.0
2.9267	0.34	300	3.3049	1.0
3.1351	0.45	400	2.9246	1.0
3.4365	0.56	500	4.2798	1.0
3.1861	0.67	600	4.0740	1.0
2.914	0.78	700	3.6876	1.0
3.0777	0.9	800	3.7421	1.0
2.8181	1.01	900	3.7825	1.0
2.8211	1.12	1000	3.9630	1.0
2.8209	1.23	1100	3.9605	1.0
2.8304	1.35	1200	3.7005	1.0
2.85	1.46	1300	3.5085	1.0
2.8509	1.57	1400	3.6157	1.0
2.8643	1.68	1500	3.5116	1.0
2.8265	1.79	1600	3.6322	1.0
2.8032	1.91	1700	4.0325	1.0
2.8053	2.02	1800	4.0125	1.0
2.819	2.13	1900	3.7971	1.0
2.8163	2.24	2000	3.9216	1.0
2.8214	2.35	2100	3.9178	1.0
2.8116	2.47	2200	3.9604	1.0
2.81	2.58	2300	3.9279	1.0
2.8051	2.69	2400	3.9737	1.0
2.8179	2.8	2500	3.9725	1.0
2.8098	2.91	2600	3.9591	1.0

Framework versions

Transformers 4.12.0.dev0
Pytorch 1.9.0+cu111
Datasets 1.13.4.dev0
Tokenizers 0.10.3