From d8acd34f66ab35a91f10d66330bcc95a83bfcac6 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Thu, 20 Oct 2022 23:43:03 +0900
Subject: [PATCH 1/8] generalized some functions and option for ignoring first
 layer

---
 modules/hypernetworks/hypernetwork.py | 23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 7d617680..3a44b377 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -21,21 +21,27 @@ from modules.textual_inversion.learn_schedule import LearnRateScheduler
 
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
-
+    activation_dict = {"relu": torch.nn.ReLU, "leakyrelu": torch.nn.LeakyReLU, "elu": torch.nn.ELU,
+                       "swish": torch.nn.Hardswish}
+    
     def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False, activation_func=None):
         super().__init__()
 
         assert layer_structure is not None, "layer_structure must not be None"
         assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!"
         assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!"
-
+        
         linears = []
         for i in range(len(layer_structure) - 1):
             linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1])))
-            if activation_func == "relu":
-                linears.append(torch.nn.ReLU())
-            if activation_func == "leakyrelu":
-                linears.append(torch.nn.LeakyReLU())
+            # if skip_first_layer because first parameters potentially contain negative values
+            if i < 1: continue
+            if activation_func in HypernetworkModule.activation_dict:
+                linears.append(HypernetworkModule.activation_dict[activation_func]())
+            else:
+                print("Invalid key {} encountered as activation function!".format(activation_func))
+            # if use_dropout: 
+            linears.append(torch.nn.Dropout(p=0.3))
             if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
@@ -46,7 +52,7 @@ class HypernetworkModule(torch.nn.Module):
             self.load_state_dict(state_dict)
         else:
             for layer in self.linear:
-                if not "ReLU" in layer.__str__():
+                if isinstance(layer, torch.nn.Linear):
                     layer.weight.data.normal_(mean=0.0, std=0.01)
                     layer.bias.data.zero_()
 
@@ -298,7 +304,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         return hypernetwork, filename
 
     scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-    optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
+    # if optimizer == "Adam": or else Adam / AdamW / etc...
+    optimizer = torch.optim.Adam(weights, lr=scheduler.learn_rate)
 
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:

From a71e0212363979c7cbbb797c9fbd5f8cd03b29d3 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Thu, 20 Oct 2022 23:48:52 +0900
Subject: [PATCH 2/8] only linear

---
 modules/hypernetworks/hypernetwork.py | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 3a44b377..905cbeef 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -35,13 +35,13 @@ class HypernetworkModule(torch.nn.Module):
         for i in range(len(layer_structure) - 1):
             linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1])))
             # if skip_first_layer because first parameters potentially contain negative values
-            if i < 1: continue
+            # if i < 1: continue
             if activation_func in HypernetworkModule.activation_dict:
                 linears.append(HypernetworkModule.activation_dict[activation_func]())
             else:
                 print("Invalid key {} encountered as activation function!".format(activation_func))
             # if use_dropout: 
-            linears.append(torch.nn.Dropout(p=0.3))
+            # linears.append(torch.nn.Dropout(p=0.3))
             if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
@@ -80,7 +80,7 @@ class HypernetworkModule(torch.nn.Module):
     def trainables(self):
         layer_structure = []
         for layer in self.linear:
-            if not "ReLU" in layer.__str__():
+            if isinstance(layer, torch.nn.Linear):
                 layer_structure += [layer.weight, layer.bias]
         return layer_structure
 
@@ -304,8 +304,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         return hypernetwork, filename
 
     scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-    # if optimizer == "Adam": or else Adam / AdamW / etc...
-    optimizer = torch.optim.Adam(weights, lr=scheduler.learn_rate)
+    # if optimizer == "AdamW": or else Adam / AdamW / SGD, etc...
+    optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
 
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:

From 108be15500aac590b4e00420635d7b61fccfa530 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Fri, 21 Oct 2022 01:00:41 +0900
Subject: [PATCH 3/8] fix bugs and optimizations

---
 modules/hypernetworks/hypernetwork.py | 93 +++++++++++++++------------
 1 file changed, 53 insertions(+), 40 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 905cbeef..893ba110 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -36,14 +36,14 @@ class HypernetworkModule(torch.nn.Module):
             linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1])))
             # if skip_first_layer because first parameters potentially contain negative values
             # if i < 1: continue
+            if add_layer_norm:
+                linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
             if activation_func in HypernetworkModule.activation_dict:
                 linears.append(HypernetworkModule.activation_dict[activation_func]())
             else:
                 print("Invalid key {} encountered as activation function!".format(activation_func))
             # if use_dropout: 
             # linears.append(torch.nn.Dropout(p=0.3))
-            if add_layer_norm:
-                linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
         self.linear = torch.nn.Sequential(*linears)
 
@@ -115,11 +115,24 @@ class Hypernetwork:
 
         for k, layers in self.layers.items():
             for layer in layers:
-                layer.train()
                 res += layer.trainables()
 
         return res
 
+    def eval(self):
+        for k, layers in self.layers.items():
+            for layer in layers:
+                layer.eval()
+        for items in self.weights():
+            items.requires_grad = False
+
+    def train(self):
+        for k, layers in self.layers.items():
+            for layer in layers:
+                layer.train()
+        for items in self.weights():
+            items.requires_grad = True
+
     def save(self, filename):
         state_dict = {}
 
@@ -290,10 +303,6 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         shared.sd_model.first_stage_model.to(devices.cpu)
 
     hypernetwork = shared.loaded_hypernetwork
-    weights = hypernetwork.weights()
-    for weight in weights:
-        weight.requires_grad = True
-
     losses = torch.zeros((32,))
 
     last_saved_file = "<none>"
@@ -304,10 +313,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         return hypernetwork, filename
 
     scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-    # if optimizer == "AdamW": or else Adam / AdamW / SGD, etc...
-    optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
+    optimizer = torch.optim.AdamW(hypernetwork.weights(), lr=scheduler.learn_rate)
 
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
+    hypernetwork.train()
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
 
@@ -328,8 +337,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
             losses[hypernetwork.step % losses.shape[0]] = loss.item()
 
-            optimizer.zero_grad()
+            optimizer.zero_grad(set_to_none=True)
             loss.backward()
+            del loss
             optimizer.step()
         mean_loss = losses.mean()
         if torch.isnan(mean_loss):
@@ -346,44 +356,47 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         })
 
         if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0:
+            torch.cuda.empty_cache()
             last_saved_image = os.path.join(images_dir, f'{hypernetwork_name}-{hypernetwork.step}.png')
+            with torch.no_grad():
+                hypernetwork.eval()
+                shared.sd_model.cond_stage_model.to(devices.device)
+                shared.sd_model.first_stage_model.to(devices.device)
 
-            optimizer.zero_grad()
-            shared.sd_model.cond_stage_model.to(devices.device)
-            shared.sd_model.first_stage_model.to(devices.device)
+                p = processing.StableDiffusionProcessingTxt2Img(
+                    sd_model=shared.sd_model,
+                    do_not_save_grid=True,
+                    do_not_save_samples=True,
+                )
 
-            p = processing.StableDiffusionProcessingTxt2Img(
-                sd_model=shared.sd_model,
-                do_not_save_grid=True,
-                do_not_save_samples=True,
-            )
+                if preview_from_txt2img:
+                    p.prompt = preview_prompt
+                    p.negative_prompt = preview_negative_prompt
+                    p.steps = preview_steps
+                    p.sampler_index = preview_sampler_index
+                    p.cfg_scale = preview_cfg_scale
+                    p.seed = preview_seed
+                    p.width = preview_width
+                    p.height = preview_height
+                else:
+                    p.prompt = entries[0].cond_text
+                    p.steps = 20
 
-            if preview_from_txt2img:
-                p.prompt = preview_prompt
-                p.negative_prompt = preview_negative_prompt
-                p.steps = preview_steps
-                p.sampler_index = preview_sampler_index
-                p.cfg_scale = preview_cfg_scale
-                p.seed = preview_seed
-                p.width = preview_width
-                p.height = preview_height
-            else:
-                p.prompt = entries[0].cond_text
-                p.steps = 20
+                preview_text = p.prompt
 
-            preview_text = p.prompt
+                processed = processing.process_images(p)
+                image = processed.images[0] if len(processed.images)>0 else None
 
-            processed = processing.process_images(p)
-            image = processed.images[0] if len(processed.images)>0 else None
+                if unload:
+                    shared.sd_model.cond_stage_model.to(devices.cpu)
+                    shared.sd_model.first_stage_model.to(devices.cpu)
 
-            if unload:
-                shared.sd_model.cond_stage_model.to(devices.cpu)
-                shared.sd_model.first_stage_model.to(devices.cpu)
+                if image is not None:
+                    shared.state.current_image = image
+                    image.save(last_saved_image)
+                    last_saved_image += f", prompt: {preview_text}"
 
-            if image is not None:
-                shared.state.current_image = image
-                image.save(last_saved_image)
-                last_saved_image += f", prompt: {preview_text}"
+            hypernetwork.train()
 
         shared.state.job_no = hypernetwork.step
 

From f89829ec3a0baceb445451ad98d4fb4323e922aa Mon Sep 17 00:00:00 2001
From: aria1th <35677394+aria1th@users.noreply.github.com>
Date: Fri, 21 Oct 2022 01:37:11 +0900
Subject: [PATCH 4/8] Revert "fix bugs and optimizations"

This reverts commit 108be15500aac590b4e00420635d7b61fccfa530.
---
 modules/hypernetworks/hypernetwork.py | 93 ++++++++++++---------------
 1 file changed, 40 insertions(+), 53 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 893ba110..905cbeef 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -36,14 +36,14 @@ class HypernetworkModule(torch.nn.Module):
             linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1])))
             # if skip_first_layer because first parameters potentially contain negative values
             # if i < 1: continue
-            if add_layer_norm:
-                linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
             if activation_func in HypernetworkModule.activation_dict:
                 linears.append(HypernetworkModule.activation_dict[activation_func]())
             else:
                 print("Invalid key {} encountered as activation function!".format(activation_func))
             # if use_dropout: 
             # linears.append(torch.nn.Dropout(p=0.3))
+            if add_layer_norm:
+                linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
         self.linear = torch.nn.Sequential(*linears)
 
@@ -115,24 +115,11 @@ class Hypernetwork:
 
         for k, layers in self.layers.items():
             for layer in layers:
+                layer.train()
                 res += layer.trainables()
 
         return res
 
-    def eval(self):
-        for k, layers in self.layers.items():
-            for layer in layers:
-                layer.eval()
-        for items in self.weights():
-            items.requires_grad = False
-
-    def train(self):
-        for k, layers in self.layers.items():
-            for layer in layers:
-                layer.train()
-        for items in self.weights():
-            items.requires_grad = True
-
     def save(self, filename):
         state_dict = {}
 
@@ -303,6 +290,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         shared.sd_model.first_stage_model.to(devices.cpu)
 
     hypernetwork = shared.loaded_hypernetwork
+    weights = hypernetwork.weights()
+    for weight in weights:
+        weight.requires_grad = True
+
     losses = torch.zeros((32,))
 
     last_saved_file = "<none>"
@@ -313,10 +304,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         return hypernetwork, filename
 
     scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-    optimizer = torch.optim.AdamW(hypernetwork.weights(), lr=scheduler.learn_rate)
+    # if optimizer == "AdamW": or else Adam / AdamW / SGD, etc...
+    optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
 
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
-    hypernetwork.train()
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
 
@@ -337,9 +328,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
             losses[hypernetwork.step % losses.shape[0]] = loss.item()
 
-            optimizer.zero_grad(set_to_none=True)
+            optimizer.zero_grad()
             loss.backward()
-            del loss
             optimizer.step()
         mean_loss = losses.mean()
         if torch.isnan(mean_loss):
@@ -356,47 +346,44 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         })
 
         if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0:
-            torch.cuda.empty_cache()
             last_saved_image = os.path.join(images_dir, f'{hypernetwork_name}-{hypernetwork.step}.png')
-            with torch.no_grad():
-                hypernetwork.eval()
-                shared.sd_model.cond_stage_model.to(devices.device)
-                shared.sd_model.first_stage_model.to(devices.device)
 
-                p = processing.StableDiffusionProcessingTxt2Img(
-                    sd_model=shared.sd_model,
-                    do_not_save_grid=True,
-                    do_not_save_samples=True,
-                )
+            optimizer.zero_grad()
+            shared.sd_model.cond_stage_model.to(devices.device)
+            shared.sd_model.first_stage_model.to(devices.device)
 
-                if preview_from_txt2img:
-                    p.prompt = preview_prompt
-                    p.negative_prompt = preview_negative_prompt
-                    p.steps = preview_steps
-                    p.sampler_index = preview_sampler_index
-                    p.cfg_scale = preview_cfg_scale
-                    p.seed = preview_seed
-                    p.width = preview_width
-                    p.height = preview_height
-                else:
-                    p.prompt = entries[0].cond_text
-                    p.steps = 20
+            p = processing.StableDiffusionProcessingTxt2Img(
+                sd_model=shared.sd_model,
+                do_not_save_grid=True,
+                do_not_save_samples=True,
+            )
 
-                preview_text = p.prompt
+            if preview_from_txt2img:
+                p.prompt = preview_prompt
+                p.negative_prompt = preview_negative_prompt
+                p.steps = preview_steps
+                p.sampler_index = preview_sampler_index
+                p.cfg_scale = preview_cfg_scale
+                p.seed = preview_seed
+                p.width = preview_width
+                p.height = preview_height
+            else:
+                p.prompt = entries[0].cond_text
+                p.steps = 20
 
-                processed = processing.process_images(p)
-                image = processed.images[0] if len(processed.images)>0 else None
+            preview_text = p.prompt
 
-                if unload:
-                    shared.sd_model.cond_stage_model.to(devices.cpu)
-                    shared.sd_model.first_stage_model.to(devices.cpu)
+            processed = processing.process_images(p)
+            image = processed.images[0] if len(processed.images)>0 else None
 
-                if image is not None:
-                    shared.state.current_image = image
-                    image.save(last_saved_image)
-                    last_saved_image += f", prompt: {preview_text}"
+            if unload:
+                shared.sd_model.cond_stage_model.to(devices.cpu)
+                shared.sd_model.first_stage_model.to(devices.cpu)
 
-            hypernetwork.train()
+            if image is not None:
+                shared.state.current_image = image
+                image.save(last_saved_image)
+                last_saved_image += f", prompt: {preview_text}"
 
         shared.state.job_no = hypernetwork.step
 

From 0e8ca8e7af05be22d7d2c07a47c3c7febe0f0ab6 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Sat, 22 Oct 2022 11:07:00 +0000
Subject: [PATCH 5/8] add dropout

---
 modules/hypernetworks/hypernetwork.py | 70 ++++++++++++++++-----------
 modules/hypernetworks/ui.py           | 10 ++--
 modules/ui.py                         | 45 +++++++++--------
 3 files changed, 72 insertions(+), 53 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 905cbeef..e493f366 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -1,47 +1,60 @@
+import csv
 import datetime
 import glob
 import html
 import os
 import sys
 import traceback
-import tqdm
-import csv
 
-import torch
-
-from ldm.util import default
-from modules import devices, shared, processing, sd_models
-import torch
-from torch import einsum
-from einops import rearrange, repeat
 import modules.textual_inversion.dataset
+import torch
+import tqdm
+from einops import rearrange, repeat
+from ldm.util import default
+from modules import devices, processing, sd_models, shared
 from modules.textual_inversion import textual_inversion
 from modules.textual_inversion.learn_schedule import LearnRateScheduler
+from torch import einsum
 
 
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
-    activation_dict = {"relu": torch.nn.ReLU, "leakyrelu": torch.nn.LeakyReLU, "elu": torch.nn.ELU,
-                       "swish": torch.nn.Hardswish}
-    
-    def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False, activation_func=None):
+    activation_dict = {
+        "relu": torch.nn.ReLU,
+        "leakyrelu": torch.nn.LeakyReLU,
+        "elu": torch.nn.ELU,
+        "swish": torch.nn.Hardswish,
+    }
+
+    def __init__(self, dim, state_dict=None, layer_structure=None, activation_func=None, add_layer_norm=False, use_dropout=False):
         super().__init__()
 
         assert layer_structure is not None, "layer_structure must not be None"
         assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!"
         assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!"
-        
+        assert activation_func not in self.activation_dict.keys() + "linear", f"Valid activation funcs: 'linear', 'relu', 'leakyrelu', 'elu', 'swish'"
+
         linears = []
         for i in range(len(layer_structure) - 1):
+
+            # Add a fully-connected layer
             linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1])))
-            # if skip_first_layer because first parameters potentially contain negative values
-            # if i < 1: continue
-            if activation_func in HypernetworkModule.activation_dict:
-                linears.append(HypernetworkModule.activation_dict[activation_func]())
+
+            # Add an activation func
+            if activation_func == "linear":
+                pass
+            elif activation_func in self.activation_dict:
+                linears.append(self.activation_dict[activation_func]())
             else:
-                print("Invalid key {} encountered as activation function!".format(activation_func))
-            # if use_dropout: 
-            # linears.append(torch.nn.Dropout(p=0.3))
+                raise NotImplementedError(
+                    "Valid activation funcs: 'linear', 'relu', 'leakyrelu', 'elu', 'swish'"
+                )
+
+            # Add dropout
+            if use_dropout:
+                linears.append(torch.nn.Dropout(p=0.3))
+
+            # Add layer normalization
             if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
@@ -93,7 +106,7 @@ class Hypernetwork:
     filename = None
     name = None
 
-    def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False, activation_func=None):
+    def __init__(self, name=None, enable_sizes=None, layer_structure=None, activation_func=None, add_layer_norm=False, use_dropout=False):
         self.filename = None
         self.name = name
         self.layers = {}
@@ -101,13 +114,14 @@ class Hypernetwork:
         self.sd_checkpoint = None
         self.sd_checkpoint_name = None
         self.layer_structure = layer_structure
-        self.add_layer_norm = add_layer_norm
         self.activation_func = activation_func
+        self.add_layer_norm = add_layer_norm
+        self.use_dropout = use_dropout
 
         for size in enable_sizes or []:
             self.layers[size] = (
-                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm, self.activation_func),
-                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm, self.activation_func),
+                HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.add_layer_norm, self.use_dropout),
+                HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.add_layer_norm, self.use_dropout),
             )
 
     def weights(self):
@@ -129,8 +143,9 @@ class Hypernetwork:
         state_dict['step'] = self.step
         state_dict['name'] = self.name
         state_dict['layer_structure'] = self.layer_structure
-        state_dict['is_layer_norm'] = self.add_layer_norm
         state_dict['activation_func'] = self.activation_func
+        state_dict['is_layer_norm'] = self.add_layer_norm
+        state_dict['use_dropout'] = self.use_dropout
         state_dict['sd_checkpoint'] = self.sd_checkpoint
         state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
 
@@ -144,8 +159,9 @@ class Hypernetwork:
         state_dict = torch.load(filename, map_location='cpu')
 
         self.layer_structure = state_dict.get('layer_structure', [1, 2, 1])
-        self.add_layer_norm = state_dict.get('is_layer_norm', False)
         self.activation_func = state_dict.get('activation_func', None)
+        self.add_layer_norm = state_dict.get('is_layer_norm', False)
+        self.use_dropout = state_dict.get('use_dropout', False)
 
         for size, sd in state_dict.items():
             if type(size) == int:
diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 1a5a27d8..5f6f17b6 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -3,14 +3,13 @@ import os
 import re
 
 import gradio as gr
-
-import modules.textual_inversion.textual_inversion
 import modules.textual_inversion.preprocess
-from modules import sd_hijack, shared, devices
+import modules.textual_inversion.textual_inversion
+from modules import devices, sd_hijack, shared
 from modules.hypernetworks import hypernetwork
 
 
-def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False, activation_func=None):
+def create_hypernetwork(name, enable_sizes, layer_structure=None, activation_func=None, add_layer_norm=False, use_dropout=False):
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
     assert not os.path.exists(fn), f"file {fn} already exists"
 
@@ -21,8 +20,9 @@ def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm
         name=name,
         enable_sizes=[int(x) for x in enable_sizes],
         layer_structure=layer_structure,
-        add_layer_norm=add_layer_norm,
         activation_func=activation_func,
+        add_layer_norm=add_layer_norm,
+        use_dropout=use_dropout,
     )
     hypernet.save(fn)
 
diff --git a/modules/ui.py b/modules/ui.py
index 716f14b8..d4b32c05 100644
--- a/modules/ui.py
+++ b/modules/ui.py
@@ -5,43 +5,44 @@ import json
 import math
 import mimetypes
 import os
+import platform
 import random
+import subprocess as sp
 import sys
 import tempfile
 import time
 import traceback
-import platform
-import subprocess as sp
 from functools import partial, reduce
 
+import gradio as gr
+import gradio.routes
+import gradio.utils
 import numpy as np
+import piexif
 import torch
 from PIL import Image, PngImagePlugin
-import piexif
 
-import gradio as gr
-import gradio.utils
-import gradio.routes
-
-from modules import sd_hijack, sd_models, localization
+from modules import localization, sd_hijack, sd_models
 from modules.paths import script_path
-from modules.shared import opts, cmd_opts, restricted_opts
+from modules.shared import cmd_opts, opts, restricted_opts
+
 if cmd_opts.deepdanbooru:
     from modules.deepbooru import get_deepbooru_tags
-import modules.shared as shared
-from modules.sd_samplers import samplers, samplers_for_img2img
-from modules.sd_hijack import model_hijack
-import modules.ldsr_model
-import modules.scripts
-import modules.gfpgan_model
+
 import modules.codeformer_model
-import modules.styles
 import modules.generation_parameters_copypaste
-from modules import prompt_parser
-from modules.images import save_image
-import modules.textual_inversion.ui
+import modules.gfpgan_model
 import modules.hypernetworks.ui
 import modules.images_history as img_his
+import modules.ldsr_model
+import modules.scripts
+import modules.shared as shared
+import modules.styles
+import modules.textual_inversion.ui
+from modules import prompt_parser
+from modules.images import save_image
+from modules.sd_hijack import model_hijack
+from modules.sd_samplers import samplers, samplers_for_img2img
 
 # this is a fix for Windows users. Without it, javascript files will be served with text/html content-type and the browser will not show any UI
 mimetypes.init()
@@ -1223,8 +1224,9 @@ def create_ui(wrap_gradio_gpu_call):
                     new_hypernetwork_name = gr.Textbox(label="Name")
                     new_hypernetwork_sizes = gr.CheckboxGroup(label="Modules", value=["768", "320", "640", "1280"], choices=["768", "320", "640", "1280"])
                     new_hypernetwork_layer_structure = gr.Textbox("1, 2, 1", label="Enter hypernetwork layer structure", placeholder="1st and last digit must be 1. ex:'1, 2, 1'")
+                    new_hypernetwork_activation_func = gr.Dropdown(value="relu", label="Select activation function of hypernetwork", choices=["linear", "relu", "leakyrelu", "elu", "swish"])
                     new_hypernetwork_add_layer_norm = gr.Checkbox(label="Add layer normalization")
-                    new_hypernetwork_activation_func = gr.Dropdown(value="relu", label="Select activation function of hypernetwork", choices=["linear", "relu", "leakyrelu"])
+                    new_hypernetwork_use_dropout = gr.Checkbox(label="Use dropout")
 
                     with gr.Row():
                         with gr.Column(scale=3):
@@ -1308,8 +1310,9 @@ def create_ui(wrap_gradio_gpu_call):
                 new_hypernetwork_name,
                 new_hypernetwork_sizes,
                 new_hypernetwork_layer_structure,
-                new_hypernetwork_add_layer_norm,
                 new_hypernetwork_activation_func,
+                new_hypernetwork_add_layer_norm,
+                new_hypernetwork_use_dropout
             ],
             outputs=[
                 train_hypernetwork_name,

From fccba4729db341a299db3343e3264fecd9459a07 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Sat, 22 Oct 2022 12:02:41 +0000
Subject: [PATCH 6/8] add an option to avoid dying relu

---
 modules/hypernetworks/hypernetwork.py | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index b7a04038..3132a56c 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -32,7 +32,6 @@ class HypernetworkModule(torch.nn.Module):
         assert layer_structure is not None, "layer_structure must not be None"
         assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!"
         assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!"
-        assert activation_func not in self.activation_dict.keys() + "linear", f"Valid activation funcs: 'linear', 'relu', 'leakyrelu', 'elu', 'swish'"
 
         linears = []
         for i in range(len(layer_structure) - 1):
@@ -43,12 +42,13 @@ class HypernetworkModule(torch.nn.Module):
             # Add an activation func
             if activation_func == "linear" or activation_func is None:
                 pass
+            # If ReLU, Skip adding it to the first layer to avoid dying ReLU
+            elif activation_func == "relu" and i < 1:
+                pass
             elif activation_func in self.activation_dict:
                 linears.append(self.activation_dict[activation_func]())
             else:
-                raise RuntimeError(
-                    "Valid activation funcs: 'linear', 'relu', 'leakyrelu', 'elu', 'swish'"
-                )
+                raise RuntimeError(f'hypernetwork uses an unsupported activation function: {activation_func}')
 
             # Add dropout
             if use_dropout:
@@ -166,8 +166,8 @@ class Hypernetwork:
         for size, sd in state_dict.items():
             if type(size) == int:
                 self.layers[size] = (
-                    HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm, self.activation_func),
-                    HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm, self.activation_func),
+                    HypernetworkModule(size, sd[0], self.layer_structure, self.activation_func, self.add_layer_norm, self.use_dropout),
+                    HypernetworkModule(size, sd[1], self.layer_structure, self.activation_func, self.add_layer_norm, self.use_dropout),
                 )
 
         self.name = state_dict.get('name', self.name)

From 7912acef725832debef58c4c7bf8ec22fb446c0b Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Sat, 22 Oct 2022 13:00:44 +0000
Subject: [PATCH 7/8] small fix

---
 modules/hypernetworks/hypernetwork.py | 12 +++++-------
 modules/ui.py                         |  1 -
 2 files changed, 5 insertions(+), 8 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 3132a56c..7d12e0ff 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -42,22 +42,20 @@ class HypernetworkModule(torch.nn.Module):
             # Add an activation func
             if activation_func == "linear" or activation_func is None:
                 pass
-            # If ReLU, Skip adding it to the first layer to avoid dying ReLU
-            elif activation_func == "relu" and i < 1:
-                pass
             elif activation_func in self.activation_dict:
                 linears.append(self.activation_dict[activation_func]())
             else:
                 raise RuntimeError(f'hypernetwork uses an unsupported activation function: {activation_func}')
 
-            # Add dropout
-            if use_dropout:
-                linears.append(torch.nn.Dropout(p=0.3))
-
             # Add layer normalization
             if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
+            # Add dropout
+            if use_dropout:
+                p = 0.5 if 0 <= i <= len(layer_structure) - 3 else 0.2
+                linears.append(torch.nn.Dropout(p=p))
+
         self.linear = torch.nn.Sequential(*linears)
 
         if state_dict is not None:
diff --git a/modules/ui.py b/modules/ui.py
index cd118552..eca887ca 100644
--- a/modules/ui.py
+++ b/modules/ui.py
@@ -1244,7 +1244,6 @@ def create_ui(wrap_gradio_gpu_call):
                     new_hypernetwork_add_layer_norm = gr.Checkbox(label="Add layer normalization")
                     new_hypernetwork_use_dropout = gr.Checkbox(label="Use dropout")
                     overwrite_old_hypernetwork = gr.Checkbox(value=False, label="Overwrite Old Hypernetwork")
-                    new_hypernetwork_activation_func = gr.Dropdown(value="relu", label="Select activation function of hypernetwork", choices=["linear", "relu", "leakyrelu"])
 
                     with gr.Row():
                         with gr.Column(scale=3):

From 6a4fa73a38935a18779ce1809892730fd1572bee Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Sat, 22 Oct 2022 13:44:39 +0000
Subject: [PATCH 8/8] small fix

---
 modules/hypernetworks/hypernetwork.py | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 3372aae2..3bc71ee5 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -51,10 +51,9 @@ class HypernetworkModule(torch.nn.Module):
             if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
-            # Add dropout
-            if use_dropout:
-                p = 0.5 if 0 <= i <= len(layer_structure) - 3 else 0.2
-                linears.append(torch.nn.Dropout(p=p))
+            # Add dropout expect last layer
+            if use_dropout and i < len(layer_structure) - 3:
+                linears.append(torch.nn.Dropout(p=0.3))
 
         self.linear = torch.nn.Sequential(*linears)