emotion_xlmr3

This model is a fine-tuned version of xlm-roberta-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0471
• Accuracy: 0.9875
• F1: 0.9850

Model description

Hinglish Emotion Detection Model (XLM-RoBERTa Base — Fine-Tuned)

Version: 1.0 Author: Amaan Shaikh (amaan00z) Languages: Hinglish (Hindi–English code-mixed) Labels: anger (0), fear (1), disgust (2), sadness (3), disapproval (4), surprise (5), admiration (6), joy (7), love (8), neutral (9)

---

✨ Model Summary

This model is a Transformer-based Hinglish Emotion Classifier, fine-tuned on a high-quality 25,689-sample dataset designed specifically for real-world Indian social-media text.

It is built using XLM-RoBERTa Base, which performs strongly on multilingual and code-mixed text.

🎯 Labels

The model predicts 10 emotions:

ID Label

0 anger 1 fear 2 disgust 3 sadness 4 disapproval 5 surprise 6 admiration 7 joy 8 love 9 neutral

Training and evaluation data

This model was trained on a curated, cleaned, balanced Hinglish emotion dataset, including:

✅ 24,500 manually cleaned Hinglish emotion samples ✅ Additional 3,000 refined sarcastic emotion-enhanced Hinglish samples ✅ Removal of duplicates & noisy samples ✅ Emoji-heavy and Gen-Z slang patterns included

Dataset Source: (Add your Kaggle link)

---

🎯 Labels

The model predicts 10 emotions:

ID Label

0 anger 1 fear 2 disgust 3 sadness 4 disapproval 5 surprise 6 admiration 7 joy 8 love 9 neutral

citation

Shaikh, Amaan. (2025). Hinglish Emotion Detection Model (XLM-R). https://huggingface.co/amaan00z/emotion-hinglish-xlmr

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.1204	1.0	1445	0.0718	0.9833	0.9801
0.0788	2.0	2890	0.0521	0.9840	0.9809
0.045	3.0	4335	0.0481	0.9875	0.9849
0.0205	4.0	5780	0.0471	0.9875	0.9850

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1