whisper-large-v3-translate-zh-v0.1-lt

This model is a fine-tuned version of openai/whisper-large-v3.

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
2.1282	0.0739	1000	2.1852	1.9014	4.4904
1.8567	0.1478	2000	1.8366	1.7295	3.8716
1.6968	0.2217	3000	1.2615	1.6279	2.4825
1.6264	0.2956	4000	1.0536	1.5625	1.8101
1.5687	0.3695	5000	1.0932	1.5410	2.1218
1.531	0.4433	6000	1.5156	1.2533	2.3689
1.4875	0.5172	7000	1.4697	0.9560	1.5588
1.4518	0.5911	8000	1.4521	1.0170	1.6392
1.4472	0.6650	9000	1.4463	1.0084	1.6420
1.3991	0.7389	10000	1.4238	0.9266	1.6992
1.4266	0.8128	11000	1.4141	0.8365	1.3056
1.3755	0.8867	12000	1.4033	0.7904	1.3119
1.3833	0.9606	13000	1.4004	0.8600	1.3333
1.3224	1.0345	14000	1.3770	0.8243	1.4560
1.3295	1.1084	15000	1.3770	0.7852	1.4298
1.3136	1.1823	16000	1.3564	0.7176	1.1826
1.2832	1.2561	17000	1.3535	0.6767	1.1781
1.2917	1.3300	18000	1.3584	0.7255	1.1218
1.27	1.4039	19000	1.3330	0.6590	1.1242
1.2704	1.4778	20000	1.3379	0.6934	1.1944
1.2614	1.5517	21000	1.3330	0.6949	1.1820
1.2455	1.6256	22000	1.3350	0.6931	1.0892
1.2475	1.6995	23000	1.3154	0.6662	1.1576
1.2583	1.7734	24000	1.3164	0.6490	1.0705
1.2333	1.8473	25000	1.3184	0.6833	1.1480
1.2462	1.9212	26000	1.3125	0.6672	1.1612
1.2279	1.9950	27000	1.3047	0.6644	1.2179
1.1908	2.0689	28000	1.3047	0.6938	1.2221
1.1831	2.1428	29000	1.2998	0.6316	1.0717
1.1705	2.2167	30000	1.3018	0.6165	1.0958
1.171	2.2906	31000	1.3027	0.6109	1.0868
1.1567	2.3645	32000	1.3037	0.6485	1.1736
1.1705	2.4384	33000	1.2969	0.6078	1.0515
1.1819	2.5123	34000	1.2949	0.6158	1.0362
1.1447	2.5862	35000	1.2920	0.6365	1.0558
1.17	2.6601	36000	1.2881	0.6339	1.0868
1.1495	2.7340	37000	1.2949	0.6297	1.0437
1.1395	2.8078	38000	1.2900	0.6285	1.1221
1.15	2.8817	39000	1.2891	0.5997	1.0217
1.1623	2.9556	40000	1.2881	0.6085	1.0395

Framework versions

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