Miséricorde's frankenstein DS V4.1 quant (calibrated 2.75 bpw EXL3 experts plus hot experts in mxfp4, fp4 KV) on 4 cards; GLM-5.3-Flash recipe for 3 cards at 3.5 bpw and for 1 to 2 cards

Swimlane: Default swimlane
Column: Idea
Position: 15

Description

Reported by / on it: @wakeupmbrc1, @cibernox, @finishtilt14, @shirogeek

Source: https://discord.com/channels/1518634379604136178/1528826330454429706/1561796404446367916

Wiki page: Quantization on the 170HX; GLM 5.3 Flash on 3x 8 GB

Cards: both

Card D-099, seeded from the cmpunlocker Discord with credit (reference discord:1561796404446367916).

Sub-Tasks

Internal links

Comments