miosipof/whisper_large_BB_and_EC_v1

Whisper Large v3

This model is a fine-tuned version of openai/whisper-large-v3 on the ASR_BB_and_EC dataset. It achieves the following results on the evaluation set:

• Loss: 0.2251
• Wer: 145.1895

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4667	0.1142	50	2.3898	126.5306
1.3136	0.2283	100	0.8863	65.3061
0.6296	0.3425	150	0.7403	55.9767
0.551	0.4566	200	0.6749	61.2245
0.4789	0.5708	250	0.6446	67.6385
0.4246	0.6849	300	0.5675	77.5510
0.3786	0.7991	350	0.5163	45.4810
0.3179	0.9132	400	0.4786	84.8397
0.3118	1.0274	450	0.4678	105.5394
0.2689	1.1416	500	0.4322	125.3644
0.2473	1.2557	550	0.3924	48.1050
0.2319	1.3699	600	0.3980	208.7464
0.2098	1.4840	650	0.3545	52.1866
0.2215	1.5982	700	0.3489	48.1050
0.1981	1.7123	750	0.3378	76.3848
0.1803	1.8265	800	0.3295	43.7318
0.1693	1.9406	850	0.3095	76.9679
0.1406	2.0548	900	0.2993	43.4402
0.1252	2.1689	950	0.2810	37.3178
0.111	2.2831	1000	0.2854	164.1399
0.1166	2.3973	1050	0.2752	124.4898
0.1183	2.5114	1100	0.2493	90.3790
0.1014	2.6256	1150	0.2441	210.2041
0.1076	2.7397	1200	0.2340	152.1866
0.0891	2.8539	1250	0.2312	214.5773
0.0841	2.9680	1300	0.2251	145.1895

Framework versions

• PEFT 0.13.2
• Transformers 4.45.2
• Pytorch 2.2.0
• Datasets 3.1.0
• Tokenizers 0.20.3