FastLLM (150M) — Modern Causal Language Model

A ~150M parameter decoder-only model (GQA, SwiGLU, RMSNorm, RoPE) from devoppro/FastLLM, still training. Small model, so expect small-model quality — this is a demo, not a chatbot.
16 256
0.1 1.5
0 100
Examples
Prompt Max new tokens Temperature Top-k