f1rdavs/tajik-banking-intent-classifier

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

tajik-banking-intent-classifier

This model is a fine-tuned version of xlm-roberta-base trained on a Tajik-translated version of the Banking77 dataset. The dataset contains customer service queries related to banking, classified into 77 different intent categories.

🧾 Model description

• —Base model: XLM-RoBERTa Base
• —Language: Tajik (tg)
• —Task: Text classification (intent recognition)
• —Number of classes: 77 <p>The model is designed to classify banking-related queries into one of 77 categories such as cardpayment, atmsupport, balance, lostorstolen_card, etc. It is useful for building customer support bots or virtual assistants that operate in the Tajik language.</p>

✅ Intended uses

• —Banking customer support chatbots for Tajik-speaking users
• —Voice or text-based virtual assistants in the finance domain
• —Automated ticket or query routing in Tajik financial services

⚠️ Limitations

• —The model may not generalize well to non-banking topics
• —Classification performance depends on the quality and accuracy of the dataset translation

📚 Training and evaluation data

• —Dataset: Banking77 dataset translated from English to Tajik
• —Size: ~13,000 examples across 77 intent classes
• —Source: Original banking77 English dataset, translated via machine translation

⚙️ Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 2e-05
• —trainbatchsize: 16
• —evalbatchsize: 16
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —num_epochs: 5
• —mixedprecisiontraining: Native AMP

Framework versions

• —Transformers 4.52.4
• —Pytorch 2.6.0+cu124
• —Datasets 3.6.0
• —Tokenizers 0.21.1