0.5.94: the q8_2f build runs on the card — 357.3 s to 171.5 s 24d7fb2 verified infosave commited on 9 days ago
0.5.93: the latent upscaler, the q8_2f build, reference clips 1669c83 verified infosave commited on 9 days ago
the adapter's cost re-measured on a machine with headroom: 1.06x, which is the noise floor be7e6c4 verified infosave commited on 10 days ago
0.5.92: runtime adapters, the branch router, and what this port does not do yet 72a3ac3 verified infosave commited on 10 days ago
card: field notes from discussions #1/#2/#4 — 24 GB frame budgets, keyframe VAE encode on CPU, kill disarmed during prompt encode (0.5.80), P-core QoS 41b5099 verified infosave commited on 13 days ago
card: the compact fl2va file — start/end frames on 16-24 GB machines 77201ae verified infosave commited on 14 days ago
README: lead with the three-way size comparison, decide by VRAM up front e8a0878 verified infosave commited on 15 days ago
README: one 'Making it smaller' section, a Which-file table keyed on VRAM, three-way sample comparison 0914add verified infosave commited on 15 days ago
ClipProj section: scope the timings to the card they were taken on 79fdefb verified infosave commited on 15 days ago
ClipProj 4B encoder variant: 23.47 GB -> 13.16 GB, 2x faster, still four bits everywhere 7097d8d verified infosave commited on 15 days ago
add the host-only row: 646.1 s against 60.2 on the card, same machine and seed 94c6382 verified infosave commited on 17 days ago
0.5.68: measured on the current build — 60.2 s at four steps, per-stage table, and how to use two cards 60fd8b2 verified infosave commited on 17 days ago
0.5.68: the render is on the card end to end — 91.6 s at 8 steps, and a seed finally reproduces 76c9afc verified infosave commited on 18 days ago
