37.6 GB

Ctrl+K

1 contributor

History: 13 commits

chenjiaj109550158

Model card: base_model lists the SANA export the checkpoints were built from

a260293 verified 16 days ago

flux-dev-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
flux-dev
AbsorbQuant v2 release: pixart-sigma, sdxl-turbo, sdxl-base, flux-dev re-derived with the corrected Algorithm-1 Stage B (amax family, tau=0, per-alpha gains); sana-1.6b / flux-schnell remain v1 pending re-derivation about 1 month ago
flux-schnell-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
flux-schnell
AbsorbQuant v2: flux-schnell (lambda 0.003 + S_amax@0.25) - campaign product independently re-derived from scratch on a second RTX 5090 with the same content digest; replaces the v1 file about 1 month ago
pixart-sigma-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
pixart-sigma
AbsorbQuant v2 release: pixart-sigma, sdxl-turbo, sdxl-base, flux-dev re-derived with the corrected Algorithm-1 Stage B (amax family, tau=0, per-alpha gains); sana-1.6b / flux-schnell remain v1 pending re-derivation about 1 month ago
sana-1.6b-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
sana-1.6b
AbsorbQuant v2: sana-1.6b re-derived from scratch on an independent RTX 5090 (lambda 0.003 + S_amax@0.75), bit-identical when rebuilt on the release machine; replaces the v1 file about 1 month ago
sdxl-base-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
sdxl-base
AbsorbQuant v2 release: pixart-sigma, sdxl-turbo, sdxl-base, flux-dev re-derived with the corrected Algorithm-1 Stage B (amax family, tau=0, per-alpha gains); sana-1.6b / flux-schnell remain v1 pending re-derivation about 1 month ago
sdxl-turbo-lam0.1-v4
AbsorbQuant lam0.1-v4: search-free recipe (lambda 0.1, per-layer-type alpha from layer SSE), six models + model card 28 days ago
sdxl-turbo
AbsorbQuant v2 release: pixart-sigma, sdxl-turbo, sdxl-base, flux-dev re-derived with the corrected Algorithm-1 Stage B (amax family, tau=0, per-alpha gains); sana-1.6b / flux-schnell remain v1 pending re-derivation about 1 month ago
.gitattributes

1.52 kB
initial commit about 2 months ago
LICENSE_NOTE

245 Bytes
Model card: per-base-model licenses instead of a blanket apache-2.0 16 days ago
README.md

8.41 kB
Model card: base_model lists the SANA export the checkpoints were built from 16 days ago