8 days ago
-
Attention page: ref2va 768x1344 sink LoRAs (quality, balanced, x6), all released LoRAs re-measured together, SMD 8-step transfer, layer map and speed-vs-error charts
nynxz +1: Claude Opus 5.5 (1M context)
9 days ago
-
Add comfyanonymous' test attention LoRA with sinks (attributed), per-step caveat and blind test
-
Add experimental sink attention LoRAs (need ComfyUI PR #16994) and results
-
Restructure card into index + docs pages; add attention config LoRAs
nynxz +1: Claude Opus 5.5 (1M context) Landing README with contents, headline results and usage; per-experiment pages
(NVFP4 checkpoint, attention LoRAs, AdaLN, methods); balanced / quality / fast
per-layer attention config LoRAs tuned on INT8 ConvRot + dmad 4-step.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
