Datasets:

minas2025
/

warm-chat-12k

language:
  - en
license: other
task_categories:
  - text-generation
tags:
  - chat
  - conversational
  - sft
  - warm-tone
pretty_name: warm-chat-12k
size_categories:
  - 10K<n<100K

warm-chat-12k

11,400 warm conversational SFT rows + 600 held-out eval rows, messages format (role/content). Built because every modern public chat set is either a technical manual or a roleplay convention, and neither teaches a model to just... talk like a person.

Filtered for genuine chat: no code, no math, no fiction, no roleplay, no refusals, no "as an AI" disclaimers, no markdown lectures. What's left is greetings, advice, everyday questions, and casual back-and-forth — with feeling.

Composition

Source	Rows	Teacher
Verdugie/opus-candid-training-data	1,136	Claude Opus 4.6
nvidia/Nemotron-SFT-Instruction-Following-Chat-v2 (reasoning_off)	10,264	Kimi-K2 / GLM-4.6 / Qwen3-235B-Instruct

Every row carries a src field with its origin. License note: Nemotron data is ODC-By (attribution required — this card is that attribution). Check the opus-candid repo for its terms before commercial use.

Example

user: hello

assistant: Hey! ٩(◕‿◕｡)۶ How's it going? I'm here to chat or help with whatever you need — just let me know!

Use

from datasets import load_dataset
ds = load_dataset("minas2025/warm-chat-12k")