NLLB-200 Fine-Tuned for Colloquial Telugu

Model Description

This model is a fine-tuned version of the NLLB-200 (Distilled 600M) designed for translating English sentences into colloquial Telugu. It has been optimized to better capture informal and conversational nuances.

Model Details

  • Model Name: anithasoma/nllb-finetuned-telugu
  • Base Model: facebook/nllb-200-distilled-600M
  • Fine-Tuned By: anithasoma
  • Languages: English β†’ Telugu (colloquial)
  • Framework: Transformers (πŸ€— Hugging Face)

πŸš€ Run the Model on Google Colab

Open in Colab

Click the badge above to launch the model in Google Colab!

Training Details

  • Dataset: anithasoma/refined_en_te
  • Training Environment: Google Colab with NVIDIA GPU.
  • Fine-Tuning Method: LoRA + PEFT (Parameter Efficient Fine-Tuning)
  • Epochs: Adjusted based on validation loss.
  • Metrics: BLEU Score, SacreBLEU Score Perplexity, Human Evaluation.

Evaluation Metrics

The model was evaluated using the BLEU and SacreBLEU metrics:

  • BLEU Score: 43.12
  • SacreBLEU Score: 43.12

How to Use

You can use this model in Python with the transformers library:

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("anithasoma/nllb-finetuned-telugu")
model = AutoModelForSeq2SeqLM.from_pretrained("anithasoma/nllb-finetuned-telugu")

def translate(text):
    inputs = tokenizer(text, return_tensors="pt")
    outputs = model.generate(**inputs)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

print(translate("Hello, how are you?"))

Model Card

Intended Use

This model is intended for generating colloquial Telugu translations from English text, improving conversational AI, and enhancing informal communication applications.

Limitations

  • May not perform well on formal or domain-specific text.
  • Can sometimes produce literal rather than context-aware translations.

License

This model is licensed under the Apache 2.0 License.

Contributors

Developed by anithasoma as part of the SAWiT AI Hackathon.


For feedback or collaboration, reach out via Hugging Face! πŸš€

Downloads last month
101
Safetensors
Model size
615M params
Tensor type
F32
Β·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for anithasoma/nllb-finetuned-telugu

Finetuned
(108)
this model

Dataset used to train anithasoma/nllb-finetuned-telugu