Abliterated GLM 5.2 & Local Inference Economics

@jcartu ran a 4-hour abliteration on 8x H200s to produce a mostly-uncensored GLM 5.2 with full tool calling intact — jumping on Tor, browsing forums, whatever you ask. The reality check: at 2.6-3T params, serving locally…