• Choose a model (N-ATLaS-LLM by default).
• Type your message and press Enter (⇧+Enter for newline) or click Send.
• Regenerate (↻) asks the model for a fresh answer to the last turn.
• Step back (⟲) removes the last user+assistant turn.
• Delete (at the top-right of the chat pane) clears the entire chat.

Use this page to chat with N-ATLaS-LLM or LLaMA-3-8B-Instruct in a multi-turn setting across languages.
We want you to ask real questions, switch languages, this tells us where N-ATLaS shines or struggles relative to its LLaMA counterpart.

• Pick the Source and Target languages from the dropdowns..
• Enter the source text and click Translate.
• See outputs from both models side-by-side.
• Vote which is better and Submit.
• Change direction or paste new text anytime.

We welcome Nigerian ↔ Nigerian language translations for well-vast speakers to help explore the models in that direction.
Heads-up: the N-ATLaS model endpoint may be paused intermittently due to cost, please engage while it’s up.

• Enter your email to create a stable ID so you can resume later.
• Choose a Task: MT (English↔Yoruba/Igbo/Hausa) or Math (pick language).
• Click Start Evaluation to load randomized items.
• Navigate with Previous / Next / Random.
• Vote (“A better”, “B better”, “Tie”, “Neither”) — we’ll auto-advance.
• Toggle Skip items I’ve already voted on to continue where you left off.

Here we encounrage you to do a careful, per-item human evaluation of two tasks where human judgment matters most:

• Machine Translation (MT): eng ↔ yor/ibo/hau and yor/ibo/hau ↔ each other.
• Mathematical Reasoning (AfriMGSM): free-form step-by-step solutions.