Fix architecture: add RMSNorm, change GELU to ReLU (matches Julia training)

Mask phantom token index 28, add valid_vocab to generation

Switch to Python/FastAPI server (RandyGPT pattern)

Rewrite Space for pure-Julia MicroGPT (no Flux)

Aggressive sampling: temp=0.1, top_k=8, rep_penalty=1.3

Fix inference: top-k sampling, repetition penalty, prompt encoding, vocab fixes

Fix space: add README, fix Dockerfile, add SSE streaming support

Replace MicroJulia v1 with JuliaGPTDistill Flux model

Rename all microgpt references to microjulia

Fix package loading: use shared Julia depot and single project path

Fix JSON3 install: use Pkg.add instead of Pkg.instantiate

Add MicroGPT Julia inference server

initial commit