mu0n commited on
Commit
31f962d
·
1 Parent(s): 78e3334

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -0
README.md CHANGED
@@ -116,4 +116,14 @@ Therefore, LCM (LORA) was applied only to the MID layer. The speedup was -25%, b
116
  As a result, I think this is the best solution.<br>
117
  ※This is the case when LCM 0.5 is applied. If the intensity is increased, the speed should be further improved.<br>
118
  (<<NON LCM STEP:16<<rev 020 LCM ALL Layer STEP:6>>LCM MID Layer STEP:12>>)<br>
 
 
 
 
 
 
 
 
 
 
119
  <br>
 
116
  As a result, I think this is the best solution.<br>
117
  ※This is the case when LCM 0.5 is applied. If the intensity is increased, the speed should be further improved.<br>
118
  (<<NON LCM STEP:16<<rev 020 LCM ALL Layer STEP:6>>LCM MID Layer STEP:12>>)<br>
119
+ <br>
120
+ [おすすめ] 高効率な LCM 利用法<br>
121
+ LCM(LORA) をモデルにマージする場合 BM1280ch (FreeU相当) にすると STEP 設定に余裕が生まれます<br>
122
+ 今回のテスト版では STEP:10 以上の場合に破綻のない画像を高頻度で得られました<br>
123
+ ※BM1280ch=1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0<br>
124
+ [Recommendation] Efficient LCM usage<br>
125
+ When merging LCM(LORA) into a model, set BM1280ch (equivalent to FreeU) to give more leeway for STEP setting.<br>
126
+ In this test version, we obtained images with no collapses at a high frequency when STEP:10 or higher was used.<br>
127
+ ※BM1280ch = 1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0<br>
128
+ <br>
129
  <br>