I pretrained a 0.5B model from scratch - Dexter
Modern Llama-style architecture: RoPE, RMSNorm, SwiGLU. Trains nicely on consumer GPUs and already produces coherent, sensible outputs.
Right now it barely competes with low-parameter Qwen models .
https://t.co/IDQUe6YdLM
so after all that , they shipped a uh.. what is it a landing page or dashboard or what ? whatever I am a "Social Media Manager" not a techie