curiositytech/MARS
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kLicense:llama3Architecture:Transformer0.0K Cold
MARS is an 8 billion parameter instruction-tuned causal language model developed by Curiosity Technology, based on Meta Llama 3 8B Instruct. It is specifically fine-tuned using LoRA on an extensive in-house Turkish dataset and translated open-source datasets, making it highly optimized for Turkish language understanding and generation. This model offers strong performance for conversational AI applications requiring Turkish language proficiency.
Loading preview...
MARS: Turkish-Optimized Llama 3 8B
MARS is the inaugural large language model from Curiosity Technology, built upon the robust Meta Llama 3 8B Instruct architecture. This model distinguishes itself through its specialized training regimen, focusing heavily on the Turkish language.
Key Capabilities & Training:
- Turkish Language Proficiency: MARS has been fine-tuned using LoRA (Low-Rank Adaptation) on a unique combination of in-house Turkish datasets and Turkish translations of various open-source datasets. This targeted training makes it particularly adept at understanding and generating Turkish text.
- Base Model: Leverages the strong foundational capabilities of Llama 3 8B Instruct.
- Training Duration: The model underwent a focused training period of 3 days on 4xA100 GPUs, indicating a dedicated effort to imbue it with specific linguistic capabilities.
Use Cases:
- Turkish Conversational AI: Ideal for chatbots, virtual assistants, and other interactive applications requiring high-quality Turkish language interaction.
- Turkish Content Generation: Suitable for tasks involving generating text, summaries, or responses in Turkish.
- Research and Development: Provides a strong base for further research and development in Turkish natural language processing, with the promise of future release of its Turkish datasets to the community.