#389, #401, #407

scripts/A1111/lyco_helpers.py

@@ -7,8 +7,8 @@ def make_weight_cp(t, wa, wb):
 
 
 def rebuild_conventional(up, down, shape, dyn_dim=None):
-    up = up.reshape(up.size(0), -1)
-    down = down.reshape(down.size(0), -1)
+    up = cpufloat(up.reshape(up.size(0), -1))
+    down = cpufloat(down.reshape(down.size(0), -1))
     if dyn_dim is not None:
         up = up[:, :dyn_dim]
         down = down[:dyn_dim, :]
@@ -16,8 +16,9 @@ def rebuild_conventional(up, down, shape, dyn_dim=None):
 
 
 def rebuild_cp_decomposition(up, down, mid):
-    up = up.reshape(up.size(0), -1)
-    down = down.reshape(down.size(0), -1)
+    up = cpufloat(up.reshape(up.size(0), -1))
+    down = cpufloat(down.reshape(down.size(0), -1))
+    mid = cpufloat(mid)
     return torch.einsum('n m k l, i n, m j -> i j k l', mid, up, down)
 
 
@@ -66,3 +67,6 @@ def factorization(dimension: int, factor:int=-1) -> tuple[int, int]:
         n, m = m, n
     return m, n
 
+def cpufloat(module):
+    if not module: return module #None対策
+    return module.to(torch.float) if module.device.type == "cpu" else module

scripts/mergers/mergers.py

@@ -128,13 +128,10 @@ def smergegen(weights_a,weights_b,model_a,model_b,model_c,base_alpha,base_beta,m
 
     save = True if SAVEMODES[0] in save_sets else False
 
-    if not forge:
-        result = savemodel(theta_0,currentmodel,custom_name,save_sets,metadata) if save else "Merged model loaded:"+currentmodel
+    result = savemodel(theta_0,currentmodel,custom_name,save_sets,metadata) if save else "Merged model loaded:"+currentmodel
 
     model_loader(checkpoint_info, theta_0, metadata, currentmodel)
 
-    if forge and save:
-        result = forge_save(custom_name if custom_name else currentmodel.replace(" ","").replace(",","_").replace("(","_").replace(")","_"))
 
     cachedealer(False)
 
@@ -308,7 +305,7 @@ def smerge(weights_a,weights_b,model_a,model_b,model_c,base_alpha,base_beta,mode
             if not(len(weights_b) == 25 or len(weights_b) == 19 or len(weights_a) == 60): return f"ERROR: weights beta value must be 20 or 26 or 61.",*NON4
         
     caster("model load start",hearm)
-    printstart(model_a,model_b,model_c,base_alpha,base_beta,weights_a,weights_b,mode,useblocks,calcmode,deep,lucks['ceed'],fine,inex,ex_blocks,ex_elems)
+    printstart(model_a,model_b,model_c,base_alpha,base_beta,weights_a,weights_b,mode,useblocks,calcmode,deep,lucks['ceed'],fine,inex,ex_blocks,ex_elems,device)
 
     theta_1=load_model_weights_m(model_b,2,cachetarget,device).copy()
 
@@ -417,6 +414,7 @@ def smerge(weights_a,weights_b,model_a,model_b,model_c,base_alpha,base_beta,mode
 
         theta_0[key] = theta_0[key].to(device)
         theta_1[key] = theta_1[key].to(device)
+
         try:
             theta_2[key] = theta_2[key].to(device)
         except Exception as e:
@@ -567,12 +565,14 @@ def smerge(weights_a,weights_b,model_a,model_b,model_c,base_alpha,base_beta,mode
         if flux and not calcmode == "smoothAdd MT":
             theta_1[key] = None
             del theta_1[key]
+
         theta_0[key] = theta_0[key].to("cpu")
         try:
             theta_1[key] = theta_1[key].to("cpu")
         except:
             pass
 
+
     #flux
     if qtype[0]:
         dellist = []
@@ -950,9 +950,9 @@ def elementals(key,weight_index,deep,randomer,num,lucks,deepprint,current_alpha)
 
 def forkforker(filename,device):
     if forge:
-        return load_torch_file(filename)
+        return load_torch_file(filename, device = torch.device(device))
     try:
-        return sd_models.read_state_dict(filename,map_location = device)
+        return sd_models.read_state_dict(filename, map_location = device)
     except:
         return sd_models.read_state_dict(filename)
 
@@ -1550,7 +1550,7 @@ def getcachelist():
 ################################################
 ##### print
 
-def printstart(model_a,model_b,model_c,base_alpha,base_beta,weights_a,weights_b,mode,useblocks,calcmode,deep,lucks,fine,inex,ex_blocks,ex_elems):
+def printstart(model_a,model_b,model_c,base_alpha,base_beta,weights_a,weights_b,mode,useblocks,calcmode,deep,lucks,fine,inex,ex_blocks,ex_elems,device):
     print(f"  model A  \t: {model_a}")
     print(f"  model B  \t: {model_b}")
     print(f"  model C  \t: {model_c}")
@@ -1564,6 +1564,7 @@ def printstart(model_a,model_b,model_c,base_alpha,base_beta,weights_a,weights_b,
     print(f"  Weights Seed\t: {lucks}")
     print(f"  {inex} \t: {ex_blocks,ex_elems}")
     print(f"  Adjust \t: {fine}")
+    print(f"  Device \t: {device}")
 
 def caster(news,hear):
     if hear: print(news)

scripts/mergers/pluslora.py

@@ -296,6 +296,7 @@ def makelora(model_a,model_b,dim,saveto,settings,alpha,beta,save_precision,calc_
     except:
         currentinfo = None
 
+    lowvram.module_in_gpu = None #web-uiのバグ対策
 
     checkpoint_info = sd_models.get_closet_checkpoint_match(model_a)
     load_model(checkpoint_info)