Model (instruction-tuned chat models only; the base models are not offered here)

🪨 Quartz Micro Preview Chat

Model: Quartz Micro Preview V2 (instruction-tuned, 100M parameters)

A tiny dense model trained from scratch on one consumer GPU, then tuned to chat. It writes short, fluent, on-topic text but often states wrong facts with confidence, is weak at maths and code, and can loop on longer answers. It sees the earlier messages of the chat (up to its 1,024-token window). Don't rely on it for anything that matters. Ask it "what are your limitations?" for a longer straight answer (that reply is fixed text, not generated).
0 1.5
0.1 1
10 400
1 1.5
Examples (the context in the first one is used by V1 only)
Your message Context (V1 only, optional): paste a short passage and ask a question about it. This is V1's best-measured use.

Models: V2 (100M, stock transformers / MLX / GGUF) · V1 (~1B MoE, no GGUF) · Apache-2.0 · Running on ZeroGPU (shared GPU, so there can be a queue). This app's code does not store your messages.