metadata

language:
  - vi
license: cc-by-nc-4.0
base_model: facebook/mms-1b-all
tags:
  - generated_from_trainer
datasets:
  - ducha07/audio_HTV_thoisu
metrics:
  - wer
model-index:
  - name: ASR-test
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: HTV news
          type: ducha07/audio_HTV_thoisu
        metrics:
          - name: Wer
            type: wer
            value: 0.2882930019620667

ASR-test-1

This model is a fine-tuned version of facebook/mms-1b-all on the HTV news dataset. It achieves the following results on the evaluation set:

Loss: 0.5663
Wer: 0.2883

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: 0.001
train_batch_size: 32
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
num_epochs: 40
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.772	0.92	100	0.8456	0.4411
1.1042	1.83	200	0.7041	0.4076
0.9814	2.75	300	0.7243	0.3782
0.9096	3.67	400	0.6771	0.3655
0.8823	4.59	500	0.6265	0.3627
0.8435	5.5	600	0.6200	0.3543
0.8157	6.42	700	0.6414	0.3417
0.822	7.34	800	0.5872	0.3431
0.7852	8.26	900	0.6012	0.3387
0.7533	9.17	1000	0.6023	0.3256
0.7609	10.09	1100	0.5837	0.3444
0.7568	11.01	1200	0.5791	0.3311
0.7091	11.93	1300	0.6227	0.3206
0.7098	12.84	1400	0.5766	0.3266
0.7006	13.76	1500	0.6084	0.3117
0.6673	14.68	1600	0.5857	0.3120
0.6832	15.6	1700	0.5754	0.3338
0.6646	16.51	1800	0.5963	0.3117
0.6524	17.43	1900	0.5816	0.3137
0.6385	18.35	2000	0.5691	0.3257
0.6433	19.27	2100	0.5929	0.3105
0.6129	20.18	2200	0.5709	0.3067
0.624	21.1	2300	0.5686	0.3168
0.6128	22.02	2400	0.5867	0.3080
0.584	22.94	2500	0.5680	0.3101
0.5956	23.85	2600	0.5611	0.3023
0.5825	24.77	2700	0.5821	0.2999
0.56	25.69	2800	0.5622	0.3012
0.56	26.61	2900	0.5590	0.3053
0.5523	27.52	3000	0.5758	0.2967
0.5335	28.44	3100	0.5649	0.3090
0.5686	29.36	3200	0.5703	0.2931
0.5488	30.28	3300	0.5709	0.2921
0.5249	31.19	3400	0.5646	0.2973
0.5278	32.11	3500	0.5628	0.2933
0.5252	33.03	3600	0.5663	0.2927
0.5092	33.94	3700	0.5618	0.2922
0.5099	34.86	3800	0.5616	0.2954
0.5031	35.78	3900	0.5670	0.2913
0.4959	36.7	4000	0.5679	0.2923
0.4936	37.61	4100	0.5675	0.2912
0.5012	38.53	4200	0.5661	0.2897
0.4819	39.45	4300	0.5663	0.2883

Framework versions

Transformers 4.37.0.dev0
Pytorch 2.1.0+cu121
Datasets 2.15.0
Tokenizers 0.15.0