README.md

metadata

language:
  - ga
  - en
license: apache-2.0
base_model: openai/whisper-small
tags:
  - generated_from_trainer
datasets:
  - ymoslem/IWSLT2023-GA-EN
  - ymoslem/FLEURS-GA-EN
  - ymoslem/BitesizeIrish-GA-EN
  - ymoslem/SpokenWords-GA-EN-MTed
metrics:
  - bleu
  - wer
model-index:
  - name: Whisper Small GA-EN Speech Translation Raw + warmup_ratio=0.01
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: IWSLT-2023, FLEURS, BiteSize, and SpokenWords
          type: ymoslem/IWSLT2023-GA-EN
        metrics:
          - name: Bleu
            type: bleu
            value: 30.07
          - name: Wer
            type: wer
            value: 66.32147681224674

Whisper Small GA-EN Speech Translation Raw + warmup_ratio=0.01

This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, and SpokenWords dataset. It achieves the following results on the evaluation set:

• Loss: 1.7281
• Bleu: 30.07
• Chrf: 46.7
• Wer: 66.3215

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.03
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
2.1013	0.2155	100	1.8575	8.31	24.75	116.5241
1.5495	0.4310	200	1.5721	14.29	33.25	105.0428
1.3385	0.6466	300	1.5329	20.71	38.26	86.9428
1.071	0.8621	400	1.4540	20.37	39.28	85.1418
0.4771	1.0776	500	1.4936	18.05	39.17	93.8316
0.4685	1.2931	600	1.5303	24.36	39.36	75.6866
0.4477	1.5086	700	1.5242	22.93	42.01	80.3242
0.4238	1.7241	800	1.5052	26.32	43.01	68.8879
0.3802	1.9397	900	1.5171	25.94	41.44	73.7956
0.1429	2.1552	1000	1.5741	28.83	43.83	65.4660
0.1607	2.3707	1100	1.6029	27.67	43.2	64.9257
0.1513	2.5862	1200	1.6130	28.61	44.28	66.1864
0.137	2.8017	1300	1.6087	21.97	40.99	89.4642
0.112	3.0172	1400	1.6146	28.74	44.01	65.9613
0.0717	3.2328	1500	1.6156	27.3	42.78	70.0585
0.0596	3.4483	1600	1.6381	27.31	45.58	69.6983
0.064	3.6638	1700	1.6262	29.73	45.88	65.9163
0.0642	3.8793	1800	1.6798	30.78	46.13	68.2575
0.0335	4.0948	1900	1.6854	29.55	45.06	67.8523
0.0366	4.3103	2000	1.6963	28.83	44.42	68.8879
0.036	4.5259	2100	1.7062	28.05	43.79	69.6983
0.0259	4.7414	2200	1.7279	28.75	45.25	68.3926
0.0353	4.9569	2300	1.7084	29.7	46.13	66.3665
0.0138	5.1724	2400	1.6906	30.81	46.26	64.1603
0.0156	5.3879	2500	1.7135	29.09	45.94	67.4471
0.0133	5.6034	2600	1.7311	29.86	45.61	65.5110
0.0161	5.8190	2700	1.7067	29.5	45.22	67.0869
0.0098	6.0345	2800	1.7038	30.32	46.6	65.3309
0.008	6.25	2900	1.7261	29.88	46.41	66.8167
0.0045	6.4655	3000	1.7281	30.07	46.7	66.3215

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1