RakshithFury/Qwen2.5-7b-en-kn-translate

Model Overview

This model, RakshithFury/Qwen2.5-7b-en-kn-translate, is a specialized translation model built upon the powerful Qwen2.5-7B-Instruct architecture. Developed by Rakshith Rao, it focuses exclusively on high-quality translation between English and Kannada (ಕನ್ನಡ), a Dravidian language spoken in Karnataka, India.

Key Capabilities & Training

• Specialized Translation: Fine-tuned for accurate English-to-Kannada translation, addressing a specific linguistic need.
• Base Model: Leverages the robust capabilities of the Qwen2.5-7B-Instruct model, providing a strong foundation.
• Extensive Training Data: Trained on 500,000 English-Kannada translation samples, totaling over 64 million tokens, ensuring comprehensive coverage of common phrases and contexts.
• Performance: Achieved a token accuracy of 87% during training, indicating strong translation fidelity.
• Efficient Training: Utilized 4x NVIDIA A100-SXM4-40GB GPUs for approximately 6 hours and 48 minutes, employing LoRA and distributed training techniques.

Use Cases & Limitations

This model is ideal for applications requiring reliable and contextually appropriate translation of English text into Kannada. It provides significantly improved translations compared to generic models, as demonstrated by the provided examples.

Good for:

• Translating general English sentences to Kannada.
• Applications needing a dedicated English-Kannada translation component.

Limitations:

• May struggle with highly complex sentences or unusual vocabulary.
• Performance might degrade with very long input sentences.