Zum Inhalt springen
ModelAufgelistetExtern
Herunterladen

Whisper Large v2 Translate Zh v0.2 St

This model is a fine-tuned version of openai/whisper-large-v2. It achieves the following results on the evaluation set: Loss: 1.2328 Ter: 60.4828 Bleu: 29.3658 Cer: 0.6602

Quellehuggingface.co/dfsADDADA/whisper-large-v2-translate-zh-v0.2-st/tree/646624a89440ca90e9f5b44371a23a27986febe0
ArtText modelBasismodellopenai/whisper-large-v2VersionmodelCBewertung

Modellquelle

Art
Text model
Dateigröße
3,00 GB
Basismodell
openai/whisper-large-v2
Version
vmodel
Aktualisiert
27. Sept. 2026
Downloads
7
Quelle
Hugging Face

Quellenbeschreibung

whisper-large-v2-translate-zh-v0.2-st

This model is a fine-tuned version of openai/whisper-large-v2. It achieves the following results on the evaluation set:

  • Loss: 1.2328
  • Ter: 60.4828
  • Bleu: 29.3658
  • Cer: 0.6602
ModelTERBLEUCER
v2-zh-v0.2-st60.482829.36580.6602
v2-zh-v0.1-lt65.056630.00340.7102
v3-zh-v0.1-lt67.197127.65720.7328

Model description

5000小时 (日语音频,中文带轴字幕) 数据微调, 翻译模式直出中文

Usage

task='translate', language='ja'

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: 1.5e-05
  • seed: 1145141919
  • total_train_batch_size: 256
  • optimizer: adamw with betas=(0.9,0.98) and epsilon=1e-06
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 7.0

Training results

Training LossEpochStepTerBleuCerValidation Loss
1.55210.1777100081.742021.52480.89511.5182
1.4390.3554200080.722922.34510.90501.4391
1.36550.5330300087.850421.81060.96371.3942
1.33370.7107400072.361925.03720.80211.3678
1.280.8884500070.369426.08190.78841.3418
1.23171.0661600067.953926.45080.75251.3320
1.20911.2438700072.668325.67450.79291.3200
1.19521.4215800072.651125.81940.80571.3073
1.18571.5991900067.111227.25540.73991.2972
1.15551.77681000068.643526.65950.75071.2923
1.13651.95451100072.072825.89300.80091.2863
1.09422.13221200073.629025.72070.83421.2930
1.09132.30991300067.569227.27510.75031.2845
1.06922.48761400067.293527.56900.74621.2808
1.06342.66521500067.118727.22630.73451.2874
1.06492.84291600070.414226.67930.81151.2705
1.01563.02061700067.484027.27070.74971.2726
1.01333.19831800061.576628.87710.67531.2796
0.99183.37601900063.936728.24970.70481.2773
0.98733.55372000064.116828.40870.70271.2751
0.98553.73132100063.690228.30990.70121.2770
0.97863.90902200065.141128.05470.71371.2705
0.95034.08672300066.055527.55990.73281.2769
0.95454.26442400063.573628.37900.69661.2777
0.95134.44212500062.539628.65780.68871.2747
0.93534.61982600065.224027.92740.71891.2741
0.93674.79742700062.804828.51890.68911.2757
0.93684.97512800065.272527.89430.72311.2715
0.91465.15282900063.413728.41510.69551.2780
0.90635.33053000062.291628.89330.68991.2770
0.89825.50823100060.721129.19590.67311.2819
0.9025.68593200061.149229.26060.67081.2786
0.90785.86353300061.958328.94410.67961.2799
0.89726.04123400061.306129.05000.67261.2826
0.90126.21893500061.956828.90050.68061.2848
0.89546.39663600061.061829.16540.66721.2822
0.88966.57433700060.482829.36580.66021.2812

Framework versions

  • Transformers 4.52.4
  • Pytorch 2.7.1+cu128
  • Datasets 4.0.0
  • Tokenizers 0.21.2

Quellen

1 Quelle
Hugging FacePrimärVersion 646624a89440ca90e9f5b44371a23a27986febe0
huggingface.co/dfsADDADA/whisper-large-v2-translate-zh-v0.2-st/tree/646624a89440ca90e9f5b44371a23a27986febe0

Verifiziert 30. Aug.

Modellartefakte

1 Artefakt

model.safetensors

safetensors · 3,00 GB · SHA-256 0abdde18518f…8ad0 · Hugging Face

Download file

Quellenauszüge

2 Auszüge
Readme30. Aug.huggingface.co/dfsADDADA/whisper-large-v2-translate-zh-v0.2-st

--- library_name: transformers license: apache-2.0 base_model: openai/whisper-large-v2 tags: - audio - automatic-speech-recognition - translate - generated_from_trainer language: - ja - zh metrics: - bleu - ter model-index: - name: whisper-large-v2-translate-zh-v0.2-st results: [] --- # whisper-large-v2-translate-zh-v0.2-st This model is a fine-tuned version of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2). It achieves the following results on the evaluation set: - Loss: 1.2328 - Ter: 60.4828 - Bleu: 29.3658 - Cer: 0.6602 | Model | TER | BLEU | CER | |:-------------:|:-----------:|:-----------:|:----------:| | v2-zh-v0.2-st | **60.4828** | 29.3658 | **0.6602** | | v2-zh-v0.1-lt | 65.0566 | **30.0034** | 0.7102 | | v3-zh-v0.1-lt | 67.1971 | 27.6572 | 0.7328 | ## Model description 5000小时 (日语音频,中文带轴字幕) 数据微调, 翻译模式直出中文 ## Usage task='translate', language='ja' ## 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: 1.5e-05 - seed: 1145141919 - total_train_batch_size: 256 - optimizer: adamw with betas=(0.9,0.98) and epsilon=1e-06 - lr_scheduler_type: linear - lr_scheduler_warmup_ratio: 0.1 - num_epochs: 7.0 ### Training results | Training Loss | Epoch | Step | Ter | Bleu | Cer | Validation Loss | |:-------------:|:------:|:-----:|:-------:|:-------:|:------:|:---------------:| | 1.5521 | 0.1777 | 1000 | 81.7420 | 21.5248 | 0.8951 | 1.5182 | | 1.439 | 0.3554 | 2000 | 80.7229 | 22.3451 | 0.9050 | 1.4391 | | 1.3655 | 0.5330 | 3000 | 87.8504 | 21.8106 | 0.9637 | 1.3942 | | 1.3337 | 0.7107 | 4000 | 72.3619 | 25.0372 | 0.8021 | 1.3678 | | 1.28 | 0.8884 | 5000 | 70.3694 | 26.0819 | 0.7884 | 1.3418 | | 1.2317 | 1.0661 | 6000 | 67.9539 | 26.4508 | 0.7525 | 1.3320 | | 1.2091 | 1.2438 | 7000 | 72.6683 | 25.6745 | 0.7929 | 1.3200 | | 1.1952 | 1.4215 | 8000 | 72.6511 | 25.8194 | 0.8057 | 1.3073 | | 1.1857 | 1.5991 | 9000 | 67.1112 | 27.2554 | 0.7399 | 1.2972 | | 1.1555 | 1.7768 | 10000 | 68.6435 | 26.6595 | 0.7507 | 1.2923 | | 1.1365 | 1.9545 | 11000 | 72.0728 | 25.8930 | 0.8009 | 1.2863 | | 1.0942 | 2.1322 | 12000 | 73.6290 | 25.7207 | 0.8342 | 1.2930 | | 1.0913 | 2.3099 | 13000 | 67.5692 | 27.2751 | 0.7503 | 1.2845 | | 1.0692 | 2.4876 | 14000 | 67.2935...

Note30. Aug.huggingface.co/dfsADDADA/whisper-large-v2-translate-zh-v0.2-st

Source context: 7 downloads · 0 likes · Pipeline automatic-speech-recognition · Library transformers · Repo dfsADDADA/whisper-large-v2-translate-zh-v0.2-st