whisper-large-v2-translate-zh-v0.1-lt-ct2

This model is a fine-tuned version of openai/whisper-large-v2.

Model description

3500小时 (日语音频,中文字幕) 数据微调, 翻译模式直出中文

Usage

task='translate', language='ja'

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 4000
• dropout: 0.1
• mask_time_prob: 0.05
• mask_feature_prob: 0.2
• condition_on_previous_text_rate: 0.5

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.743	0.0740	1000	1.5631	0.8223	1.4517
1.6014	0.1479	2000	1.4808	0.6775	1.0950
1.5549	0.2219	3000	1.4381	0.6756	1.1158
1.5283	0.2958	4000	1.4174	0.6992	1.1137
1.474	0.3698	5000	1.3849	0.6570	1.1369
1.4193	0.4437	6000	1.3657	0.6544	1.1339
1.4148	0.5177	7000	1.3477	0.6386	1.1647
1.3754	0.5916	8000	1.3392	0.6228	1.0461
1.3441	0.6656	9000	1.3362	0.6196	1.0609
1.3545	0.7395	10000	1.3176	0.6354	1.2138
1.3498	0.8135	11000	1.3236	0.6631	1.2232
1.31	0.8874	12000	1.3020	0.6199	1.0018
1.3213	0.9614	13000	1.2966	0.5922	1.0021
1.2375	1.0353	14000	1.2900	0.6097	1.0639
1.2334	1.1093	15000	1.2963	0.6150	1.0920
1.2277	1.1832	16000	1.2888	0.6077	1.0929
1.2087	1.2572	17000	1.2779	0.5954	1.0012
1.2131	1.3311	18000	1.2722	0.5776	1.0075
1.2012	1.4051	19000	1.2716	0.5726	1.0211
1.1912	1.4790	20000	1.2707	0.6007	1.1538
1.2127	1.5530	21000	1.2749	0.6086	1.0742
1.1789	1.6269	22000	1.2797	0.5765	1.0072
1.1527	1.7009	23000	1.2761	0.5855	1.0588
1.1693	1.7748	24000	1.2701	0.5635	0.9928
1.1709	1.8488	25000	1.2662	0.5980	1.0697
1.1637	1.9227	26000	1.2749	0.5872	1.0392
1.1562	1.9967	27000	1.2587	0.5651	1.0121
1.0929	2.0706	28000	1.2668	0.5857	1.0139
1.1232	2.1446	29000	1.2710	0.5742	0.9997
1.1045	2.2185	30000	1.2656	0.5643	0.9897
1.0841	2.2925	31000	1.2695	0.5835	1.0181
1.0868	2.3664	32000	1.2707	0.5673	0.9964
1.0938	2.4404	33000	1.2644	0.5712	0.9928
1.0938	2.5143	34000	1.2662	0.5750	1.0109
1.0848	2.5883	35000	1.2677	0.5841	1.0832
1.0914	2.6622	36000	1.2638	0.5801	1.0299
1.0688	2.7362	37000	1.2587	0.5694	1.0072
1.0856	2.8101	38000	1.2581	0.5646	1.0057
1.1037	2.8841	39000	1.2557	0.5771	1.0262
1.0652	2.9580	40000	1.2566	0.5634	0.9979

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1