HIPPO
Collection
All the materials from our work: From Specialization to Generalization: Instruction-tuned LLMs for Robust Harmful Content Mitigation • 4 items • Updated
HIPPO 🦛
This repository contains our 4B model for HIPPO: Harmful Input, Positive and Productive Output. This is part of our work, From Specialization to Generalization: Instruction-tuned LLMs for Robust Harmful Content Mitigation, where we show that instruction-tuning an LLM for hate speech can dramatically improve its performance on hate speech-related tasks. We've collected 36 different hate speech datasets and converted them into a conversational template. We used it to train our HIPPO 4B and 32B models, which outperform several task-specific models.
- Our code: https://github.com/TUM-NLP/HIPPO
- Our 32B model: leukas/hippo-32b
- Our dataset: leukas/hippo_superset
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for leukas/hippo-4b
Base model
Qwen/Qwen3-4B-Instruct-2507