Yuu-Xie/distilbert-base-multilingual-cased-sentiment

Model Card for Model ID

Model Details

Model Description

This model is a fine-tuned version of distilbert-base-multilingual-cased on the clapAI/MultiLingualSentiment dataset (approx. 3.15 million samples). It is designed to provide a high-throughput, low-latency sentiment analysis baseline for multilingual environments.

The model supports three-class classification: Positive, Neutral, and Negative.

• Developed by: Yuu-Xie
• Model type: Transformer-based text classification (DistilBERT)
• Language(s) (NLP): Multilingual
• License: Apache-2.0
• Finetuned from model: distilbert-base-multilingual-cased

Uses

Direct Use

This model can be directly used for social media comment analysis, product review monitoring, and multilingual public opinion tracking. It is particularly suitable for deployment scenarios with high real-time requirements or limited computing resources.

How to Get Started with the Model

from transformers import pipeline

classifier = pipeline(
    task="text-classification", 
    model="Yuu-Xie/distilbert-base-multilingual-cased-sentiment"
)

texts = [
    "A good environment with good food. Price is reasonable.",
    "这个产品质量很一般，不建议购买。",
    "コードレス設計で車内の掃除もできます。"
]

predictions = classifier(texts)
print(predictions)

Training Details

Training Data

The training set is sourced from clapAI/MultiLingualSentiment, containing roughly 3.15 million labeled multilingual text samples.

Training Procedure

Training Hyperparameters

• Training regime: bf16 mixed precision
• Batch Size: 128 (Train), 256 (Eval)
• Max Steps: 62,500
• Learning Rate: 2e-5 (warm-up)

Speeds, Sizes, Times

• Hardware: NVIDIA A10 (24GB)
• Training Time: 3 hours 41 minutes
• Throughput: ~6.5 it/s (Training)
• Model Size: 519 MiB (134M parameters)

Evaluation

Results

Final evaluation results on 393,436 independent test samples:

Metric	Score
Accuracy	0.7989
Macro Avg F1	0.7891
Weighted Avg F1	0.7988

Classification Report

Class	Precision	Recall	F1-score
Positive	0.8500	0.8386	0.8443
Negative	0.8183	0.8359	0.8270
Neutral	0.7001	0.6923	0.6961

Citation

@misc{yuu-xie2026distilbert-sentiment,
  author = {Yuu-Xie},
  title = {DistilBERT-base-multilingual-cased-sentiment},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Yuu-Xie/distilbert-base-multilingual-cased-sentiment}}
}