OdiaGenAI-LLM/odia-gemma-2b-base

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2.5BQuant:BF16Context Size:8kPublished:Apr 1, 2024License:gemmaArchitecture:Transformer Featherless Exclusive Cold

OdiaGenAI-LLM/odia-gemma-2b-base is a 2.5 billion parameter, pre-trained decoder-only large language model based on Google's Gemma 2B architecture, developed by OdiaGenAI. It is specifically pre-trained on the 49 million token Culturex-Odia dataset, a filtered version of CulturaX, making it proficient in both Odia and English. This model is designed for tasks requiring strong Odia language understanding and generation, serving as a foundational base for further fine-tuning in specific applications.

Loading preview...

Odia-Gemma-2B-Base: A Foundational Odia LLM

Odia-Gemma-2B-Base is a 2.5 billion parameter pre-trained large language model developed by OdiaGenAI, built upon Google's Gemma 2B architecture. This model is uniquely specialized for the Odia language, while also supporting English, making it a valuable resource for Odia-centric natural language processing tasks.

Key Capabilities

  • Bilingual Proficiency: Pre-trained for both Odia and English languages.
  • Specialized Training Data: Utilizes the 49 million token Culturex-Odia dataset, derived from mc4 and OSCAR corpora, ensuring strong performance in Odia.
  • Decoder-Only Architecture: A foundational model suitable for a wide range of generative tasks.
  • Base Model: Designed as a pre-trained base, it is ideal for further instruction fine-tuning to adapt to specific downstream applications.

Good For

  • Odia Language Research and Development: A strong starting point for projects focused on Odia text generation, understanding, and analysis.
  • Custom Application Development: Developers can fine-tune this model for specific Odia-language tasks such as chatbots, content creation, or translation.
  • Academic and Research Purposes: Useful for exploring large language model capabilities in low-resource languages like Odia.

Note: As a pre-trained base model, it is not instruction-tuned and may require fine-tuning for optimal performance on specific human-instruction following tasks. It lacks built-in moderation and may produce inappropriate responses without further fine-tuning.

Finetunes
3 models