GPT-2-XL on Helix — captured run (replay widget)

GPT-2-XL on Helix REPLAY · CAPTURED RUN — NOT LIVE

A real, gated GPT-2-XL run on the from-raw Helix stack (served output == independent oracle, token-for-token 25/25), replayed time-compressed. Real cadence: ≈ 9.8 s/token — trust, not speed. Open the full demo →