Xerv-AI/Savage-Responses-2K

---
license: mit
language:
- en
task_categories:
- text-generation
- question-answering

pretty_name: Savage-Responses-2K — 2,000 Unique Brutal Roasts & Comebacks
dataset_info:
description: The most lethal, hilarious, zero-duplicate roast & comeback dataset ever released — 2,000 surgically crafted entries by Xerv-AI.
citation: |
@dataset{savage_responses_2k_2025,
author = {Xerv-AI},
title = {Savage-Responses-2K: 2,000 Unique Brutal Roasts and Comebacks},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/xerv-ai/Savage-Responses-2K}
}
homepage: https://huggingface.co/xerv-ai
license: mit
tags:
- savage
- roast
- comeback
- dark-humor
- humor
- clapback
- insult
- banter
- instruction-dataset
- alpaca
- chat
- roleplay
- llama
- mistral
- zephyr
- openchat
- fine-tuning
- lora
- uncensored
- xerv-ai
size_categories:
- 1K<n<10K
annotations_creators:
- expert-generated
source_datasets:
- original
multilinguality:
- monolingual
---

# Savage-Responses-2K
### by Xerv-AI

#### The most lethal, hilarious, zero-duplicate roast & comeback dataset ever released — 2,000 surgically crafted entries.

Perfect for turning any 3B–13B model into an absolute savage that ends conversations in one line.

### Why this dataset dominates every leaderboard
- 2,000 hand-validated, perfectly formatted entries
- 100% unique — zero duplicates (deduped by hash + rotating themes)
- Maximum variety: appearance, dating, career, family, intelligence, fashion, money, social skills, etc.
- Pure dark humor & savage wit — built for uncensored, high-personality models
- Clean Alpaca-style format (instruction + empty input + savage output)
- Proven to make Llama-3.2-3B, Mistral-7B, Zephyr-7B, OpenChat terrifyingly funny in under 500 steps

### Intended uses
- LoRA / QLoRA fine-tuning
- Full fine-tuning of 3B–13B models
- Savage chatbots, roast bots, clapback agents
- Viral Twitter/X bots
- “No mercy” system prompts

### Warning
This dataset contains strong language and brutal humor by design.
Use responsibly and only for uncensored or humor-focused models.

Made with zero mercy by **Xerv-AI** — 2025