DictaLM-3.0-1.7B-Instruct: A Hebrew-Optimized LLM

DictaLM-3.0-1.7B-Instruct is part of the Dicta-LM 3.0 collection, an open-weight series of large language models developed by Dicta. This specific model is a 1.7 billion parameter instruction-tuned variant, built upon the Qwen3-1.7B-Base architecture, and operates with full BF16 precision. It has been extensively trained on vast corpora of both Hebrew and English texts.

Key Capabilities & Differentiators

• State-of-the-Art Hebrew Performance: DictaLM 3.0 models establish new benchmarks for their weight class in Hebrew language processing, excelling as both base and chat models.
• Extensive Context Window: Features a substantial context length of 40960 tokens, allowing for processing longer and more complex inputs.
• Tool Calling Support: The model is equipped with tool-calling capabilities, enabling integration with external tools and APIs for enhanced functionality.
• Instruction Following: Optimized for instruction-following, leveraging chat templates for effective prompt rendering.

Intended Use Cases

• Hebrew Language Applications: Ideal for chatbots, content generation, and natural language understanding tasks specifically in Hebrew.
• Research and Development: Provides a strong foundation for further research and fine-tuning in Hebrew NLP.
• Conversational AI: Suitable for building interactive AI assistants that require robust Hebrew dialogue capabilities.