expertai/LLaMAntino-3-SLIMER-IT: Zero-Shot NER for Italian

expertai/LLaMAntino-3-SLIMER-IT is an 8 billion parameter large language model (LLM) developed by expertai, based on Meta Llama 3. It is specifically instruction-tuned for zero-shot Named Entity Recognition (NER) on the Italian language, building upon the swap-uniba/LLaMAntino-3-ANITA-8B-Inst-DPO-ITA model.

Key Capabilities

• Zero-Shot NER: Designed to identify named entities without prior examples for specific tags.
• Prompt-Driven Entity Extraction: Leverages a unique prompting mechanism that includes a DEFINITION and GUIDELINES for the named entity to be extracted, enabling it to tackle novel entity types.
• Italian Language Focus: Optimized for performance on Italian text.
• Targeted Entity Recognition: While instructed on a reduced set of common tags (PER, ORG, LOC), its strength lies in its ability to generalize to new, user-defined entity categories.

Use Cases

This model is particularly well-suited for:

• Custom NER tasks in Italian: When standard NER models lack specific entity types required for a project.
• Rapid prototyping: Quickly extracting new entity types from Italian text without extensive re-training.
• Information extraction: Identifying specific, context-dependent information from unstructured Italian documents.

For more technical details and usage examples, refer to the GitHub repository and the associated arXiv paper.