english-marathi-colloquial-translator
This model is a fine-tuned version of
unsloth/tinyllama-chat-bnb-4bit on the RutujaPatil29/marathi-model-dataset dataset.
It achieves the following results on the evaluation set:
Model description
This model is designed to translate English sentences into Marathi with a focus on colloquial language, informal phrases, and slang. Unlike traditional translation models, it understands the nuances of everyday spoken Marathi.
Intended uses & limitations
✅ Handles casual and slang Marathi well
✅ Understands sentence structure and context
⚠️ May struggle with highly formal translations
⚠️ Does not support code-mixing (Hinglish/Marathenglish) yet
Training and evaluation data
Dataset Sources:
The model was fine-tuned on a custom dataset created from:
1️⃣ Scraped articles & text from Marathi news websites (Loksatta, Sakal, Maharashtra Times)
2️⃣ Manually annotated informal translations for better conversational accuracy
3️⃣ Commonly used Marathi slang and spoken phrases
Data Cleaning & Preprocessing:
Removed formal literary phrases to retain conversational style
Tokenized text using SentencePiece/BPE
Aligned informal Marathi with English inputs
Evaluation Metrics:
BLEU Score – Measures translation accuracy
Perplexity (PPL) – Checks model fluency
Human Evaluation – Native Marathi speakers verified outputs
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|
| 4.1568 | 2.0 | 2 | 8.7847 |
| 4.1558 | 4.0 | 4 | 8.7847 |
| 4.1567 | 6.0 | 6 | 8.7847 |
| 4.1563 | 8.0 | 8 | 8.7837 |
| 3.8778 | 10.0 | 10 | 8.7797 |
Framework versions
- PEFT 0.14.0
- Transformers 4.48.3
- Pytorch 2.6.0+cu124
- Datasets 3.3.1
- Tokenizers 0.21.0
