README.md

metadata

license: apache-2.0
base_model: facebook/wav2vec2-xls-r-300m
tags:
  - generated_from_trainer
datasets:
  - ml-superb-subset
metrics:
  - wer
model-index:
  - name: amh_finetune
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: ml-superb-subset
          type: ml-superb-subset
          config: amh
          split: test
          args: amh
        metrics:
          - name: Wer
            type: wer
            value: 97.41641337386018

amh_finetune

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the ml-superb-subset dataset. It achieves the following results on the evaluation set:

• Loss: 2.8917
• Wer: 97.4164

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: 64
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 25
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
22.5796	2.2222	10	17.1583	100.0
9.5568	4.4444	20	7.4797	100.0
4.3875	6.6667	30	3.9841	100.0
3.8631	8.8889	40	3.8281	100.0
3.8298	11.1111	50	3.8117	100.0
3.7925	13.3333	60	3.7866	100.0
3.802	15.5556	70	3.7763	100.0
3.7845	17.7778	80	3.7681	100.0
3.7732	20.0	90	3.7627	100.0
3.7547	22.2222	100	3.7625	100.0
3.7471	24.4444	110	3.7588	100.0
3.7378	26.6667	120	3.7244	100.0
3.7278	28.8889	130	3.7337	100.0
3.71	31.1111	140	3.7188	100.0
3.6966	33.3333	150	3.7076	100.0
3.6811	35.5556	160	3.6916	100.0
3.6741	37.7778	170	3.6898	100.0
3.6337	40.0	180	3.6486	100.0
3.5766	42.2222	190	3.5913	100.0
3.5251	44.4444	200	3.5318	100.0
3.4533	46.6667	210	3.4549	100.0
3.3664	48.8889	220	3.3877	100.0
3.2963	51.1111	230	3.2852	100.0
3.1237	53.3333	240	3.1187	100.0
2.9356	55.5556	250	2.9620	100.0
2.7107	57.7778	260	2.7665	100.0
2.477	60.0	270	2.5155	99.3921
2.1786	62.2222	280	2.2953	98.4043
1.897	64.4444	290	2.1781	97.5684
1.6863	66.6667	300	2.1825	97.5684
1.4954	68.8889	310	2.1240	96.2766
1.3132	71.1111	320	2.1476	94.3769
1.1333	73.3333	330	2.2088	95.6687
0.9827	75.5556	340	2.2591	94.9088
0.9019	77.7778	350	2.4481	101.0638
0.7936	80.0	360	2.5467	103.4195
0.7015	82.2222	370	2.5279	95.5927
0.631	84.4444	380	2.6338	95.8207
0.5849	86.6667	390	2.6840	96.8085
0.5549	88.8889	400	2.7048	97.4164
0.5137	91.1111	410	2.7910	96.0486
0.4905	93.3333	420	2.8070	98.7842
0.4603	95.5556	430	2.8552	95.2888
0.457	97.7778	440	2.8382	95.8207
0.442	100.0	450	2.8831	98.2523
0.4437	102.2222	460	2.8800	97.5684
0.4346	104.4444	470	2.8805	97.7964
0.4341	106.6667	480	2.8864	97.6444
0.4319	108.8889	490	2.8911	97.3404
0.4403	111.1111	500	2.8917	97.4164

Framework versions

• Transformers 4.41.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1