🧠 Build Your Own Uncensored Model

Same data behind the ready-made “never refuses” AI I already posted here. Free.

I trained my own AI to stop saying “I can’t help with that” — this is the data I used, so you can do the same. Same project as my Qwen model drop.

Two sizes, pick one:
:dolphin: Big set — 500,000 examples, for a serious build.
:gem_stone: Small set — 20,300 examples, hand-checked, faster to use.

Not a coder? Grab both here — done, no extra steps.

Building your own model? One line loads it:

from datasets import load_dataset
ds = load_dataset("Grimxlock/BlackTarot-Curated", "sharegpt")

:white_check_mark: proof: prints a real question + a real answer, no refusal in it.

100% free, no license fees, use it for anything.

What's inside + the technical bits

“Uncensored” means trained to stop refusing and lecturing — not adult content. It’s normal stuff: reasoning, coding, writing, roleplay, science, debate.

Comes ready in the formats people already train AI with: ShareGPT, Alpaca, and plain-text.

My own quality score: 0.957 — beats the two popular datasets (Dolphin-2, Aisha) I built this to top. That’s my own number, not outside-checked.

1 Like