README.md

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: ASR4-for-40-epochs
    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: 0.26843348202571504

ASR4-for-40-epochs

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: 0.4791
• Wer: 0.2684

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
5.1111	0.92	100	0.7687	0.4387
1.1201	1.83	200	0.6388	0.3767
0.9734	2.75	300	0.6319	0.3658
0.9297	3.67	400	0.5740	0.3373
0.9142	4.59	500	0.5591	0.3268
0.8462	5.5	600	0.5627	0.3227
0.8366	6.42	700	0.5491	0.3158
0.8272	7.34	800	0.5398	0.3243
0.8137	8.26	900	0.5363	0.3113
0.7643	9.17	1000	0.5528	0.3117
0.7738	10.09	1100	0.5194	0.3285
0.7622	11.01	1200	0.5348	0.3043
0.707	11.93	1300	0.5179	0.2909
0.7242	12.84	1400	0.5153	0.3138
0.7093	13.76	1500	0.5116	0.2951
0.673	14.68	1600	0.5002	0.2941
0.6877	15.6	1700	0.4958	0.3050
0.6665	16.51	1800	0.5032	0.2865
0.6507	17.43	1900	0.4871	0.2809
0.6308	18.35	2000	0.4953	0.2947
0.6507	19.27	2100	0.4998	0.2837
0.6027	20.18	2200	0.4963	0.2868
0.623	21.1	2300	0.4955	0.2953
0.6047	22.02	2400	0.5034	0.2852
0.5825	22.94	2500	0.4781	0.2795
0.585	23.85	2600	0.4851	0.2843
0.5838	24.77	2700	0.4957	0.2742
0.5718	25.69	2800	0.4885	0.2810
0.5646	26.61	2900	0.4778	0.2724
0.5476	27.52	3000	0.4914	0.2751
0.5333	28.44	3100	0.4879	0.2788
0.5533	29.36	3200	0.4820	0.2726
0.5321	30.28	3300	0.4816	0.2686
0.5161	31.19	3400	0.4865	0.2812
0.5326	32.11	3500	0.4818	0.2704
0.5188	33.03	3600	0.4816	0.2669
0.506	33.94	3700	0.4804	0.2755
0.5122	34.86	3800	0.4803	0.2667
0.506	35.78	3900	0.4785	0.2708
0.5064	36.7	4000	0.4755	0.2730
0.4997	37.61	4100	0.4804	0.2708
0.4904	38.53	4200	0.4772	0.2678
0.4774	39.45	4300	0.4791	0.2684

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0