Instructions to use A-Funakoshi/bert-finetuned-multilingual-sentiments-adafactor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use A-Funakoshi/bert-finetuned-multilingual-sentiments-adafactor with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="A-Funakoshi/bert-finetuned-multilingual-sentiments-adafactor")# Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("A-Funakoshi/bert-finetuned-multilingual-sentiments-adafactor") model = AutoModelForSequenceClassification.from_pretrained("A-Funakoshi/bert-finetuned-multilingual-sentiments-adafactor") - Notebooks
- Google Colab
- Kaggle
multilingual-sentimentsデータセットをbertベースのモデルでfinetuningしたもの
ベースモデル:cl-tohoku/bert-base-japanese-whole-word-masking
データセット:tyqiangz/multilingual-sentiments
バッチサイズ: 16固定
オプティマイザ: adafactor
Optunaでハイパーパラメータ探索
- 学習率スケジュールのタイプ(lr_scheduler_type): constant, linear, cosine
- 学習率(learning rate): 1e-6 ~ 1e-4
- 勾配累積ステップ(gradient_accumulation_steps): 1, 2, 4, 8, 16
- 正則化(weight_decay): 1e-6 ~ 1e-1
Optunaでの探索結果は以下
- 学習率スケジュールタイプ(lr_scheduler_type): cosine
- 学習率(learning rate): 2.64959745375728e-05
- 勾配累積ステップ(gradient_accumulation_steps): 4
- 正則化(weight_decay): 1.1217342612041105e-06
このハイパーパラメータを使って再度finetuningした.
- Downloads last month
- 6