fix para_to_weights() (issue #103)

sd_modelmixer/hyper.py

@@ -22,14 +22,19 @@
 
 classifiers = get_classifiers()
 
-def para_to_weights(para, weights=None, isxl=False):
+def para_to_weights(para, weights=None, alpha=None, isxl=False):
     BLOCKS = all_blocks(isxl)
     BLOCKLEN = (12 if not isxl else 9)*2 + 2
 
     weights = {} if weights is None else dict(zip(range(len(weights)), weights))
+    alpha = {}
     for k in para:
         name = k.split(".")
         modelidx = ord(name[0].split("_")[1]) - 98
+        if name[1] == "alpha":
+            alpha[modelidx] = para[k]
+            continue
+
         weight = weights.get(modelidx, [0.0]*BLOCKLEN)
         j = BLOCKS.index(name[1])
         weight[j] = para[k]
@@ -40,7 +45,12 @@ def para_to_weights(para, weights=None, isxl=False):
     for i in weights.keys():
         nweights[i] = ",".join([("0" if float(f) == 0.0 else str(f)) for f in weights[i]])
 
-    return nweights
+    maxid = max(alpha.keys())
+    nalpha = [""] * (maxid + 1)
+    for i in alpha.keys():
+        nalpha[i] = alpha[i]
+
+    return nweights, nalpha
 
 
 def normalize_mbw(mbw, isxl):
@@ -546,12 +556,13 @@ def hyper_score(localargs):
             shared.state.end()
 
         if best_para is not None:
-            best_weights = para_to_weights(best_para, weights, isxl)
+            best_weights, best_alpha = para_to_weights(best_para, weights, alpha, isxl)
             print(" - Best weights para = ", best_weights, override_uses)
+            print(" - Best alpha para = ", best_alpha)
 
             # setup override weights. will be replaced with mm_weights
 
-            shared.modelmixer_overrides = {"weights": best_weights, "uses": override_uses}
+            shared.modelmixer_overrides = {"weights": best_weights, "alpha": best_alpha, "uses": override_uses}
 
             # generate image with the optimized parameter
             ret = None