diego-florez/Qwen2.5-7B-Comment-Sexuality-Reasoner

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 18, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The diego-florez/Qwen2.5-7B-Comment-Sexuality-Reasoner is a 7.6 billion parameter Spanish-language model, fine-tuned from Qwen/Qwen2.5-7B-Instruct, designed for context-aware analysis of social media comments. It classifies comments as non-sexualized, sexualized, or neutral, providing a Spanish reasoning and probability scores for each classification. This model specializes in evaluating comment intent relative to image context, topic, and author information, making it suitable for research into social media moderation and content analysis.

Loading preview...

Model Overview

The diego-florez/Qwen2.5-7B-Comment-Sexuality-Reasoner is a 7.6 billion parameter model, fine-tuned from Qwen/Qwen2.5-7B-Instruct using DoRA, specifically for analyzing Spanish social media comments. Its core function is to classify comments as non-sexualized, sexualized, or neutral, while also generating a detailed Spanish reasoning and probability scores for each category. A key differentiator is its context-aware approach, evaluating comments based on associated image context, topic, and author metadata, rather than just isolated words.

Key Capabilities

  • Context-aware classification: Analyzes comments in relation to image_context, topic, and author_gender.
  • Reasoning generation: Provides a short Spanish explanation for each classification.
  • Probability scores: Outputs confidence scores for each of the three labels.
  • Performance: Achieves 70.54% accuracy and 70.53% Macro F1 on its specific task, significantly outperforming a larger 32B model, and offers much faster inference (35s vs 4m 30s for 448 samples).
  • Reduced contradictions: LLM-as-a-judge evaluation indicates fewer contradictory reasonings compared to the 32B model.

Intended Use Cases

  • Research on Spanish social media comments and content moderation.
  • Context-aware classification of potentially sexualized content.
  • Dataset annotation and semi-supervised labeling.
  • Experimental research into sexuality/sexualization detection in social media.

This model is primarily a research model and not intended as a production-ready moderation system without further validation.

Finetunes
1 models