Simplify inference and evaluation for multi-task models

[Ingvarstep]

Classification (GLiNERClassifier)

• Supports single-label and multi-label text classification.
• Includes evaluation metrics (F1 score) for dataset evaluation.

Question-Answering (GLiNERQuestionAnswerer)

• Performs question-answering tasks by extracting answers from input texts.
• Includes evaluation capabilities using the GLiNERSquadEvaluator on datasets like SQuAD.

Relation Extraction (GLiNERRelationExtractor)

• Combines zero-shot named entity recognition (NER) and relation extraction.
• Supports user-defined entity types and relations with configurable thresholds and filtering options.
• Added support of evaluation on DocRED dataset;

Open Information Extraction (GLiNEROpenExtractor)

• Extracts structured information from text based on user queries.
• Supports customizable prompts for better alignment with task-specific needs.

Summarization (GLiNERSummarizer)

• Implements extractive summarization by identifying key information from text.

[urchade]

Thanks @Ingvarstep, will review it asap