From 85ccc52e9fd0789077de1ea67129704f2ecb0d3a Mon Sep 17 00:00:00 2001 From: Ethanfel Date: Sat, 7 Mar 2026 03:17:36 +0100 Subject: [PATCH] =?UTF-8?q?fix:=20scoring=5Fspeed=20tooltip=20=E2=80=94=20?= =?UTF-8?q?fast=20is=20not=20the=20default,=20turbo=20is?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Co-Authored-By: Claude Opus 4.6 --- lora_optimizer.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lora_optimizer.py b/lora_optimizer.py index 876027f..ace24bf 100644 --- a/lora_optimizer.py +++ b/lora_optimizer.py @@ -4653,7 +4653,7 @@ class LoRAAutoTuner(LoRAOptimizer): "tooltip": "Controls how many prefixes Phase 2 scores per candidate. " "All candidates are scored on the same subset so ranking stays fair.\n\n" "• full — Score every prefix. Best accuracy, slowest. Use when merging very different LoRAs (e.g. style + character + concept) where block behavior varies a lot.\n" - "• fast — Every 2nd prefix (~50%% faster). Good default for most merges.\n" + "• fast — Every 2nd prefix (~50%% faster). Higher accuracy than turbo, still much faster than full.\n" "• turbo — Every 3rd prefix (~67%% faster). Works well when your LoRAs have similar conflict across blocks (e.g. multiple characters from the same trainer).\n" "• turbo+ — Every 4th prefix (~75%% faster). Best for large models (DiT/Flux/WAN) or when iterating quickly. May miss subtle block-level differences on SD/SDXL." }),