Overview
The rudrashah/RLM-hinglish-translator is a specialized language model designed to translate Hinglish (a common informal blend of Hindi and English) into standard English. Built upon the Gemma-2B architecture, this 2.6 billion parameter model has been fine-tuned using Parameter-efficient Fine Tuning (PEFT) with the LoRA method.
Key Capabilities
- Hinglish to English Translation: Specifically engineered to understand and translate the unique linguistic characteristics of Hinglish.
- PEFT (LoRA) Fine-Tuning: Utilizes Low-Rank Adaptation to efficiently adapt the base Gemma-2B model, preserving its core strengths while specializing in Hinglish translation.
- Dataset Diversity: Fine-tuned on a combination of the
cmu_hinglish_dog dataset and real-world conversational data, including personal chats and Uber messages, to capture authentic Hinglish usage.
Usage
This model is particularly useful for applications requiring accurate translation of informal Hinglish communication. An example of its input/output is provided, demonstrating its ability to convert Hinglish phrases like "aapka name kya hai?" to English. Developers can integrate it using the transformers library, ensuring the correct template is applied for optimal output.
For a detailed usage example, refer to the Kaggle usage notebook.