Telugu LLM Labs has announced the release of Navarasa 2.0, a Gemma 7B/2B instruction-tuned model capable of processing content in 15 Indian languages along with English. This latest iteration marks a significant advancement following the recent introduction of fine-tuned Gemma models for nine Indian languages.

Check out the models on Hugging Face.

Navarasa 2.0 represents a substantial leap in capabilities, leveraging Gemma 7B/2B SFT models as its base. Notably, the model’s generative prowess has been expanded to encompass a diverse array of Indian languages, including Marathi, Urdu, Konkani, Assamese, Nepali, and Sindhi. 

The model was trained on a single with E2E Networks Limited using NVIDIA A100 GPUs which took approximately 44 hours for the 7 billion model and 18 hours for the 2 billion model. 

This achievement was realised through the translation of the alpaca-cleaned-filtered dataset into these additional languages, effectively broadening its linguistic repertoire.

The 15 languages supported by Navarasa 2.0 are as follows: Hindi, Telugu, Tamil, Kannada, Malayalam, Marathi, Gujarati, Bengali, Punjabi, Odia, Urdu, Konkani, Assamese, Nepali, Sindhi, and English.

Ravi Theja Desetty from LlamaIndex and Ramsri Goutham Golla, through their collaborative venture have introduced Telugu LLM Labs, initiating a significant stride towards enhancing the Telugu Natural Language Processing (NLP) space. 

Regarding datasets, Telugu LLM Labs undertook the task of translating the alpaca-cleaned-filtered dataset into six additional Indian languages. This effort led to the creation of the Indic Alpaca Datasets collection, consolidating all relevant datasets in one accessible repository. The model was fine-tuned using approximately 630,000 instruction samples.

Furthermore, the capabilities demonstrated by Navarasa 2.0 mirror those of its predecessor, allowing for seamless interaction across various language contexts. Noteworthy features include the ability to process instructions and input in native languages, respond in the same language, and handle multilingual interactions involving English.