Llama-electronic-radiology-TR: Domain-Adapted Turkish Radiology LLM

This model, developed by khazarai, is a specialized 1 billion parameter version of Llama-3.2-1B, uniquely adapted for the Turkish electronic radiology domain. It underwent continued pretraining on a dataset of Turkish-language electronic radiology PhD theses, focusing on enhancing its fluency, vocabulary, and semantic consistency within highly technical medical and radiological contexts. Unlike instruction-tuned models, its strength lies in foundational language modeling for a specific, complex domain.

Key Capabilities

• Domain-Specific Text Generation: Produces fluent and semantically rich Turkish text relevant to radiology, such as imaging protocols, research summaries, or academic abstracts.
• Medical Document Summarization: Capable of summarizing lengthy Turkish radiological texts, including reports or thesis chapters.
• Base for Downstream Tasks: Serves as an excellent foundation for further fine-tuning into instruction-tuned clinical models or Question-Answering systems in Turkish radiology.
• Research Applications: Supports the development of Turkish-language models for clinical Natural Language Processing (NLP), particularly in low-resource and domain-specific medical contexts.

Good For

• Researchers and developers working on Turkish medical NLP, especially in radiology.
• Generating academic or technical content in Turkish related to diagnostic imaging.
• Creating specialized summarization tools for Turkish radiological reports.
• As a pre-trained base for building more complex, instruction-tuned clinical AI applications in Turkish radiology.

Note: This model is not instruction-tuned and is not designed for direct prompt-based Q&A or dialogue without additional supervised fine-tuning. It is also not suitable for clinical decision-making due to its lack of factual grounding and real-time clinical judgment.