From 6021f7a75f7b5208a2be15cda5526028152f922d Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 00:51:36 +0900
Subject: add options to custom hypernetwork layer structure

---
 modules/hypernetworks/hypernetwork.py | 88 ++++++++++++++++++++++++++---------
 1 file changed, 67 insertions(+), 21 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 4905710e..cadb9911 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -1,52 +1,98 @@
+import csv
 import datetime
 import glob
 import html
 import os
 import sys
 import traceback
-import tqdm
-import csv
 
+import modules.textual_inversion.dataset
 import torch
-
-from ldm.util import default
-from modules import devices, shared, processing, sd_models
-import torch
-from torch import einsum
+import tqdm
 from einops import rearrange, repeat
-import modules.textual_inversion.dataset
+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
+
+
+def parse_layer_structure(dim, state_dict):
+    i = 0
+    res = [1]
+    while (key := "linear.{}.weight".format(i)) in state_dict:
+        weight = state_dict[key]
+        res.append(len(weight) // dim)
+        i += 1
+    return res
 
 
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
+    layer_structure = None
+    add_layer_norm = False
 
     def __init__(self, dim, state_dict=None):
         super().__init__()
+        if (state_dict is None or 'linear.0.weight' not in state_dict) and self.layer_structure is None:
+            layer_structure = (1, 2, 1)
+        else:
+            if self.layer_structure is not None:
+                assert self.layer_structure[0] == 1, "Multiplier Sequence should start with size 1!"
+                assert self.layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!"
+                layer_structure = self.layer_structure
+            else:
+                layer_structure = parse_layer_structure(dim, state_dict)
+
+        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 self.add_layer_norm:
+                linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
-        self.linear1 = torch.nn.Linear(dim, dim * 2)
-        self.linear2 = torch.nn.Linear(dim * 2, dim)
+        self.linear = torch.nn.Sequential(*linears)
 
         if state_dict is not None:
-            self.load_state_dict(state_dict, strict=True)
+            try:
+                self.load_state_dict(state_dict)
+            except RuntimeError:
+                self.try_load_previous(state_dict)
         else:
-
-            self.linear1.weight.data.normal_(mean=0.0, std=0.01)
-            self.linear1.bias.data.zero_()
-            self.linear2.weight.data.normal_(mean=0.0, std=0.01)
-            self.linear2.bias.data.zero_()
+            for layer in self.linear:
+                layer.weight.data.normal_(mean = 0.0, std = 0.01)
+                layer.bias.data.zero_()
 
         self.to(devices.device)
 
+    def try_load_previous(self, state_dict):
+        states = self.state_dict()
+        states['linear.0.bias'].copy_(state_dict['linear1.bias'])
+        states['linear.0.weight'].copy_(state_dict['linear1.weight'])
+        states['linear.1.bias'].copy_(state_dict['linear2.bias'])
+        states['linear.1.weight'].copy_(state_dict['linear2.weight'])
+
     def forward(self, x):
-        return x + (self.linear2(self.linear1(x))) * self.multiplier
+        return x + self.linear(x) * self.multiplier
+
+    def trainables(self):
+        res = []
+        for layer in self.linear:
+            res += [layer.weight, layer.bias]
+        return res
 
 
 def apply_strength(value=None):
     HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength
 
 
+def apply_layer_structure(value=None):
+    HypernetworkModule.layer_structure = value if value is not None else shared.opts.sd_hypernetwork_layer_structure
+
+
+def apply_layer_norm(value=None):
+    HypernetworkModule.add_layer_norm = value if value is not None else shared.opts.sd_hypernetwork_add_layer_norm
+
+
 class Hypernetwork:
     filename = None
     name = None
@@ -68,7 +114,7 @@ class Hypernetwork:
         for k, layers in self.layers.items():
             for layer in layers:
                 layer.train()
-                res += [layer.linear1.weight, layer.linear1.bias, layer.linear2.weight, layer.linear2.bias]
+                res += layer.trainables()
 
         return res
 
@@ -226,7 +272,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
     with torch.autocast("cuda"):
         ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=512, height=512, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size)
-
+        assert ds.length > 1, "Dataset should contain more than 1 images"
     if unload:
         shared.sd_model.cond_stage_model.to(devices.cpu)
         shared.sd_model.first_stage_model.to(devices.cpu)
@@ -261,7 +307,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         with torch.autocast("cuda"):
             c = stack_conds([entry.cond for entry in entries]).to(devices.device)
-#            c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
+            c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
             x = torch.stack([entry.latent for entry in entries]).to(devices.device)
             loss = shared.sd_model(x, c)[0]
             del x
@@ -283,7 +329,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
             "loss": f"{mean_loss:.7f}",
-            "learn_rate": scheduler.learn_rate
+            "learn_rate": f"{scheduler.learn_rate:.7f}"
         })
 
         if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0:
-- 
cgit v1.2.1


From a5611ea5026bd8e12d8e84023384c369d0511dda Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 01:00:01 +0900
Subject: update

---
 modules/hypernetworks/hypernetwork.py | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index cadb9911..c5835bce 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -1,20 +1,22 @@
-import csv
 import datetime
 import glob
 import html
 import os
 import sys
 import traceback
+import tqdm
+import csv
 
-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 import devices, shared, processing, sd_models
+import torch
+from torch import einsum
+from einops import rearrange, repeat
+import modules.textual_inversion.dataset
 from modules.textual_inversion import textual_inversion
 from modules.textual_inversion.learn_schedule import LearnRateScheduler
-from torch import einsum
 
 
 def parse_layer_structure(dim, state_dict):
-- 
cgit v1.2.1


From e40ba281f1b419cf99552962ea01d87d699840a5 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 01:03:58 +0900
Subject: update

---
 modules/hypernetworks/hypernetwork.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index c5835bce..082165f4 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -309,7 +309,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         with torch.autocast("cuda"):
             c = stack_conds([entry.cond for entry in entries]).to(devices.device)
-            c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
+            # c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
             x = torch.stack([entry.latent for entry in entries]).to(devices.device)
             loss = shared.sd_model(x, c)[0]
             del x
@@ -331,7 +331,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
             "loss": f"{mean_loss:.7f}",
-            "learn_rate": f"{scheduler.learn_rate:.7f}"
+            "learn_rate": scheduler.learn_rate
         })
 
         if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0:
-- 
cgit v1.2.1


From 42fbda83bb9830af18187fddb50c1bedd01da502 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 14:30:33 +0000
Subject: layer options moves into create hnet ui

---
 modules/hypernetworks/hypernetwork.py | 64 +++++++++++++++++------------------
 modules/hypernetworks/ui.py           |  9 +++--
 2 files changed, 39 insertions(+), 34 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 583ada31..7d519cd9 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -19,37 +19,21 @@ from modules.textual_inversion import textual_inversion
 from modules.textual_inversion.learn_schedule import LearnRateScheduler
 
 
-def parse_layer_structure(dim, state_dict):
-    i = 0
-    res = [1]
-    while (key := "linear.{}.weight".format(i)) in state_dict:
-        weight = state_dict[key]
-        res.append(len(weight) // dim)
-        i += 1
-    return res
-
-
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
-    layer_structure = None
-    add_layer_norm = False
 
-    def __init__(self, dim, state_dict=None):
+    def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False):
         super().__init__()
-        if (state_dict is None or 'linear.0.weight' not in state_dict) and self.layer_structure is None:
-            layer_structure = (1, 2, 1)
+        if layer_structure is not 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!"
         else:
-            if self.layer_structure is not None:
-                assert self.layer_structure[0] == 1, "Multiplier Sequence should start with size 1!"
-                assert self.layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!"
-                layer_structure = self.layer_structure
-            else:
-                layer_structure = parse_layer_structure(dim, state_dict)
+            layer_structure = parse_layer_structure(dim, state_dict)
 
         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 self.add_layer_norm:
+            if add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
         self.linear = torch.nn.Sequential(*linears)
@@ -77,38 +61,47 @@ class HypernetworkModule(torch.nn.Module):
         return x + self.linear(x) * self.multiplier
 
     def trainables(self):
-        res = []
+        layer_structure = []
         for layer in self.linear:
-            res += [layer.weight, layer.bias]
-        return res
+            layer_structure += [layer.weight, layer.bias]
+        return layer_structure
 
 
 def apply_strength(value=None):
     HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength
 
 
-def apply_layer_structure(value=None):
-    HypernetworkModule.layer_structure = value if value is not None else shared.opts.sd_hypernetwork_layer_structure
+def parse_layer_structure(dim, state_dict):
+    i = 0
+    layer_structure = [1]
 
+    while (key := "linear.{}.weight".format(i)) in state_dict:
+        weight = state_dict[key]
+        layer_structure.append(len(weight) // dim)
+        i += 1
 
-def apply_layer_norm(value=None):
-    HypernetworkModule.add_layer_norm = value if value is not None else shared.opts.sd_hypernetwork_add_layer_norm
+    return layer_structure
 
 
 class Hypernetwork:
     filename = None
     name = None
 
-    def __init__(self, name=None, enable_sizes=None):
+    def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False):
         self.filename = None
         self.name = name
         self.layers = {}
         self.step = 0
         self.sd_checkpoint = None
         self.sd_checkpoint_name = None
+        self.layer_structure = layer_structure
+        self.add_layer_norm = add_layer_norm
 
         for size in enable_sizes or []:
-            self.layers[size] = (HypernetworkModule(size), HypernetworkModule(size))
+            self.layers[size] = (
+                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm),
+                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm),
+            )
 
     def weights(self):
         res = []
@@ -128,6 +121,8 @@ 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['sd_checkpoint'] = self.sd_checkpoint
         state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
 
@@ -142,10 +137,15 @@ class Hypernetwork:
 
         for size, sd in state_dict.items():
             if type(size) == int:
-                self.layers[size] = (HypernetworkModule(size, sd[0]), HypernetworkModule(size, sd[1]))
+                self.layers[size] = (
+                    HypernetworkModule(size, sd[0], state_dict["layer_structure"], state_dict["is_layer_norm"]),
+                    HypernetworkModule(size, sd[1], state_dict["layer_structure"], state_dict["is_layer_norm"]),
+                )
 
         self.name = state_dict.get('name', self.name)
         self.step = state_dict.get('step', 0)
+        self.layer_structure = state_dict.get('layer_structure', None)
+        self.add_layer_norm = state_dict.get('is_layer_norm', False)
         self.sd_checkpoint = state_dict.get('sd_checkpoint', None)
         self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None)
 
diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index dfa599af..7e8ea95e 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -9,11 +9,16 @@ from modules import sd_hijack, shared, devices
 from modules.hypernetworks import hypernetwork
 
 
-def create_hypernetwork(name, enable_sizes):
+def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False):
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
     assert not os.path.exists(fn), f"file {fn} already exists"
 
-    hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(name=name, enable_sizes=[int(x) for x in enable_sizes])
+    hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(
+        name=name,
+        enable_sizes=[int(x) for x in enable_sizes],
+        layer_structure=layer_structure,
+        add_layer_norm=add_layer_norm,
+    )
     hypernet.save(fn)
 
     shared.reload_hypernetworks()
-- 
cgit v1.2.1


From 3770b8d2fa62066d472a04739c7b84bce8538832 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 15:28:42 +0000
Subject: enable to write layer structure of hn himself

---
 modules/hypernetworks/ui.py | 4 ++++
 1 file changed, 4 insertions(+)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 7e8ea95e..08f75f15 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -1,5 +1,6 @@
 import html
 import os
+import re
 
 import gradio as gr
 
@@ -13,6 +14,9 @@ def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
     assert not os.path.exists(fn), f"file {fn} already exists"
 
+    if type(layer_structure) == str:
+        layer_structure = tuple(map(int, re.sub(r'\D', '', layer_structure)))
+
     hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(
         name=name,
         enable_sizes=[int(x) for x in enable_sizes],
-- 
cgit v1.2.1


From c6e9fed5003631c87d548e74d6e359678959a453 Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Wed, 19 Oct 2022 19:21:16 +0300
Subject: fix for #3086 failing to load any previous hypernet

---
 modules/hypernetworks/hypernetwork.py | 60 ++++++++++++++++-------------------
 1 file changed, 28 insertions(+), 32 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 7d519cd9..74300122 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -24,11 +24,10 @@ class HypernetworkModule(torch.nn.Module):
 
     def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False):
         super().__init__()
-        if layer_structure is not 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!"
-        else:
-            layer_structure = parse_layer_structure(dim, state_dict)
+
+        assert layer_structure is not None, "layer_structure mut 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):
@@ -39,23 +38,30 @@ class HypernetworkModule(torch.nn.Module):
         self.linear = torch.nn.Sequential(*linears)
 
         if state_dict is not None:
-            try:
-                self.load_state_dict(state_dict)
-            except RuntimeError:
-                self.try_load_previous(state_dict)
+            self.fix_old_state_dict(state_dict)
+            self.load_state_dict(state_dict)
         else:
             for layer in self.linear:
-                layer.weight.data.normal_(mean = 0.0, std = 0.01)
+                layer.weight.data.normal_(mean=0.0, std=0.01)
                 layer.bias.data.zero_()
 
         self.to(devices.device)
 
-    def try_load_previous(self, state_dict):
-        states = self.state_dict()
-        states['linear.0.bias'].copy_(state_dict['linear1.bias'])
-        states['linear.0.weight'].copy_(state_dict['linear1.weight'])
-        states['linear.1.bias'].copy_(state_dict['linear2.bias'])
-        states['linear.1.weight'].copy_(state_dict['linear2.weight'])
+    def fix_old_state_dict(self, state_dict):
+        changes = {
+            'linear1.bias': 'linear.0.bias',
+            'linear1.weight': 'linear.0.weight',
+            'linear2.bias': 'linear.1.bias',
+            'linear2.weight': 'linear.1.weight',
+        }
+
+        for fr, to in changes.items():
+            x = state_dict.get(fr, None)
+            if x is None:
+                continue
+
+            del state_dict[fr]
+            state_dict[to] = x
 
     def forward(self, x):
         return x + self.linear(x) * self.multiplier
@@ -71,18 +77,6 @@ def apply_strength(value=None):
     HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength
 
 
-def parse_layer_structure(dim, state_dict):
-    i = 0
-    layer_structure = [1]
-
-    while (key := "linear.{}.weight".format(i)) in state_dict:
-        weight = state_dict[key]
-        layer_structure.append(len(weight) // dim)
-        i += 1
-
-    return layer_structure
-
-
 class Hypernetwork:
     filename = None
     name = None
@@ -135,17 +129,18 @@ 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)
+
         for size, sd in state_dict.items():
             if type(size) == int:
                 self.layers[size] = (
-                    HypernetworkModule(size, sd[0], state_dict["layer_structure"], state_dict["is_layer_norm"]),
-                    HypernetworkModule(size, sd[1], state_dict["layer_structure"], state_dict["is_layer_norm"]),
+                    HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm),
+                    HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm),
                 )
 
         self.name = state_dict.get('name', self.name)
         self.step = state_dict.get('step', 0)
-        self.layer_structure = state_dict.get('layer_structure', None)
-        self.add_layer_norm = state_dict.get('is_layer_norm', False)
         self.sd_checkpoint = state_dict.get('sd_checkpoint', None)
         self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None)
 
@@ -244,6 +239,7 @@ def stack_conds(conds):
 
     return torch.stack(conds)
 
+
 def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
     assert hypernetwork_name, 'hypernetwork not selected'
 
-- 
cgit v1.2.1


From 2ce52d32e41fb523d1494f45073fd18496e52d35 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Wed, 19 Oct 2022 16:31:12 +0000
Subject: fix for #3086 failing to load any previous hypernet

---
 modules/hypernetworks/hypernetwork.py | 60 ++++++++++++++++-------------------
 1 file changed, 28 insertions(+), 32 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 7d519cd9..74300122 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -24,11 +24,10 @@ class HypernetworkModule(torch.nn.Module):
 
     def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False):
         super().__init__()
-        if layer_structure is not 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!"
-        else:
-            layer_structure = parse_layer_structure(dim, state_dict)
+
+        assert layer_structure is not None, "layer_structure mut 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):
@@ -39,23 +38,30 @@ class HypernetworkModule(torch.nn.Module):
         self.linear = torch.nn.Sequential(*linears)
 
         if state_dict is not None:
-            try:
-                self.load_state_dict(state_dict)
-            except RuntimeError:
-                self.try_load_previous(state_dict)
+            self.fix_old_state_dict(state_dict)
+            self.load_state_dict(state_dict)
         else:
             for layer in self.linear:
-                layer.weight.data.normal_(mean = 0.0, std = 0.01)
+                layer.weight.data.normal_(mean=0.0, std=0.01)
                 layer.bias.data.zero_()
 
         self.to(devices.device)
 
-    def try_load_previous(self, state_dict):
-        states = self.state_dict()
-        states['linear.0.bias'].copy_(state_dict['linear1.bias'])
-        states['linear.0.weight'].copy_(state_dict['linear1.weight'])
-        states['linear.1.bias'].copy_(state_dict['linear2.bias'])
-        states['linear.1.weight'].copy_(state_dict['linear2.weight'])
+    def fix_old_state_dict(self, state_dict):
+        changes = {
+            'linear1.bias': 'linear.0.bias',
+            'linear1.weight': 'linear.0.weight',
+            'linear2.bias': 'linear.1.bias',
+            'linear2.weight': 'linear.1.weight',
+        }
+
+        for fr, to in changes.items():
+            x = state_dict.get(fr, None)
+            if x is None:
+                continue
+
+            del state_dict[fr]
+            state_dict[to] = x
 
     def forward(self, x):
         return x + self.linear(x) * self.multiplier
@@ -71,18 +77,6 @@ def apply_strength(value=None):
     HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength
 
 
-def parse_layer_structure(dim, state_dict):
-    i = 0
-    layer_structure = [1]
-
-    while (key := "linear.{}.weight".format(i)) in state_dict:
-        weight = state_dict[key]
-        layer_structure.append(len(weight) // dim)
-        i += 1
-
-    return layer_structure
-
-
 class Hypernetwork:
     filename = None
     name = None
@@ -135,17 +129,18 @@ 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)
+
         for size, sd in state_dict.items():
             if type(size) == int:
                 self.layers[size] = (
-                    HypernetworkModule(size, sd[0], state_dict["layer_structure"], state_dict["is_layer_norm"]),
-                    HypernetworkModule(size, sd[1], state_dict["layer_structure"], state_dict["is_layer_norm"]),
+                    HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm),
+                    HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm),
                 )
 
         self.name = state_dict.get('name', self.name)
         self.step = state_dict.get('step', 0)
-        self.layer_structure = state_dict.get('layer_structure', None)
-        self.add_layer_norm = state_dict.get('is_layer_norm', False)
         self.sd_checkpoint = state_dict.get('sd_checkpoint', None)
         self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None)
 
@@ -244,6 +239,7 @@ def stack_conds(conds):
 
     return torch.stack(conds)
 
+
 def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
     assert hypernetwork_name, 'hypernetwork not selected'
 
-- 
cgit v1.2.1


From d6ea5841374a28f3f6deb73abc251c8f0bcb240f Mon Sep 17 00:00:00 2001
From: DepFA <35278260+dfaker@users.noreply.github.com>
Date: Thu, 20 Oct 2022 00:07:57 +0100
Subject: change html output

---
 modules/hypernetworks/hypernetwork.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 7d519cd9..73c1cb80 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -380,7 +380,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 Loss: {mean_loss:.7f}<br/>
 Step: {hypernetwork.step}<br/>
 Last prompt: {html.escape(entries[0].cond_text)}<br/>
-Last saved embedding: {html.escape(last_saved_file)}<br/>
+Last saved hypernetwork: {html.escape(last_saved_file)}<br/>
 Last saved image: {html.escape(last_saved_image)}<br/>
 </p>
 """
-- 
cgit v1.2.1


From 166be3919b817cee5e702fd01c34afe9081b952c Mon Sep 17 00:00:00 2001
From: DepFA <35278260+dfaker@users.noreply.github.com>
Date: Thu, 20 Oct 2022 00:09:40 +0100
Subject: allow overwrite old hn

---
 modules/hypernetworks/ui.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 08f75f15..f45345ea 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -10,9 +10,10 @@ from modules import sd_hijack, shared, devices
 from modules.hypernetworks import hypernetwork
 
 
-def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False):
+def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, add_layer_norm=False):
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
-    assert not os.path.exists(fn), f"file {fn} already exists"
+    if not overwrite_old:
+        assert not os.path.exists(fn), f"file {fn} already exists"
 
     if type(layer_structure) == str:
         layer_structure = tuple(map(int, re.sub(r'\D', '', layer_structure)))
-- 
cgit v1.2.1


From 6f98e89486f55b0e4657e96ce640cf1c4675d187 Mon Sep 17 00:00:00 2001
From: discus0434 <discus0434@gmail.com>
Date: Thu, 20 Oct 2022 00:10:45 +0000
Subject: update

---
 modules/hypernetworks/hypernetwork.py | 29 +++++++++++++++++++----------
 modules/hypernetworks/ui.py           |  3 ++-
 2 files changed, 21 insertions(+), 11 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 74300122..7d617680 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -22,16 +22,20 @@ from modules.textual_inversion.learn_schedule import LearnRateScheduler
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
 
-    def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False):
+    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 mut not be None"
+        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 add_layer_norm:
                 linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
 
@@ -42,8 +46,9 @@ class HypernetworkModule(torch.nn.Module):
             self.load_state_dict(state_dict)
         else:
             for layer in self.linear:
-                layer.weight.data.normal_(mean=0.0, std=0.01)
-                layer.bias.data.zero_()
+                if not "ReLU" in layer.__str__():
+                    layer.weight.data.normal_(mean=0.0, std=0.01)
+                    layer.bias.data.zero_()
 
         self.to(devices.device)
 
@@ -69,7 +74,8 @@ class HypernetworkModule(torch.nn.Module):
     def trainables(self):
         layer_structure = []
         for layer in self.linear:
-            layer_structure += [layer.weight, layer.bias]
+            if not "ReLU" in layer.__str__():
+                layer_structure += [layer.weight, layer.bias]
         return layer_structure
 
 
@@ -81,7 +87,7 @@ class Hypernetwork:
     filename = None
     name = None
 
-    def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False):
+    def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False, activation_func=None):
         self.filename = None
         self.name = name
         self.layers = {}
@@ -90,11 +96,12 @@ class Hypernetwork:
         self.sd_checkpoint_name = None
         self.layer_structure = layer_structure
         self.add_layer_norm = add_layer_norm
+        self.activation_func = activation_func
 
         for size in enable_sizes or []:
             self.layers[size] = (
-                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm),
-                HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm),
+                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),
             )
 
     def weights(self):
@@ -117,6 +124,7 @@ class Hypernetwork:
         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['sd_checkpoint'] = self.sd_checkpoint
         state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
 
@@ -131,12 +139,13 @@ class Hypernetwork:
 
         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)
 
         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),
-                    HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm),
+                    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),
                 )
 
         self.name = state_dict.get('name', self.name)
diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 08f75f15..83f9547b 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -10,7 +10,7 @@ from modules import sd_hijack, shared, devices
 from modules.hypernetworks import hypernetwork
 
 
-def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False):
+def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False, activation_func=None):
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
     assert not os.path.exists(fn), f"file {fn} already exists"
 
@@ -22,6 +22,7 @@ def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm
         enable_sizes=[int(x) for x in enable_sizes],
         layer_structure=layer_structure,
         add_layer_norm=add_layer_norm,
+        activation_func=activation_func,
     )
     hypernet.save(fn)
 
-- 
cgit v1.2.1


From 930b4c64f7dbce6918894d53538003e5959fd022 Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Thu, 20 Oct 2022 08:18:02 +0300
Subject: allow float sizes for hypernet's layer_structure

---
 modules/hypernetworks/ui.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 08f75f15..e0741d08 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -15,7 +15,7 @@ def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm
     assert not os.path.exists(fn), f"file {fn} already exists"
 
     if type(layer_structure) == str:
-        layer_structure = tuple(map(int, re.sub(r'\D', '', layer_structure)))
+        layer_structure = [float(x.strip()) for x in layer_structure.split(",")]
 
     hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(
         name=name,
-- 
cgit v1.2.1


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: generalized some functions and option for ignoring first layer

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

(limited to 'modules/hypernetworks')

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:
-- 
cgit v1.2.1


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: only linear

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

(limited to 'modules/hypernetworks')

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:
-- 
cgit v1.2.1


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: fix bugs and optimizations

---
 modules/hypernetworks/hypernetwork.py | 105 +++++++++++++++++++---------------
 1 file changed, 59 insertions(+), 46 deletions(-)

(limited to 'modules/hypernetworks')

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)
+
+                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)
-
-            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
-
-            preview_text = p.prompt
-
-            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 image is not None:
-                shared.state.current_image = image
-                image.save(last_saved_image)
-                last_saved_image += f", prompt: {preview_text}"
+                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
+
+                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 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
 
-- 
cgit v1.2.1


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: Revert "fix bugs and optimizations"

This reverts commit 108be15500aac590b4e00420635d7b61fccfa530.
---
 modules/hypernetworks/hypernetwork.py | 105 +++++++++++++++-------------------
 1 file changed, 46 insertions(+), 59 deletions(-)

(limited to 'modules/hypernetworks')

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,
-                )
 
-                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
-
-                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 image is not None:
-                    shared.state.current_image = image
-                    image.save(last_saved_image)
-                    last_saved_image += f", prompt: {preview_text}"
-
-            hypernetwork.train()
+            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,
+            )
+
+            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
+
+            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 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
 
-- 
cgit v1.2.1


From c23f666dba2b484d521d2dc4be91cf9e09312647 Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Fri, 21 Oct 2022 09:47:43 +0300
Subject: a more strict check for activation type and a more reasonable check
 for type of layer in hypernets

---
 modules/hypernetworks/hypernetwork.py | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 7d617680..84e7e350 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -32,10 +32,16 @@ class HypernetworkModule(torch.nn.Module):
         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":
+            elif activation_func == "leakyrelu":
                 linears.append(torch.nn.LeakyReLU())
+            elif activation_func == 'linear' or activation_func is None:
+                pass
+            else:
+                raise RuntimeError(f'hypernetwork uses an unsupported activation function: {activation_func}')
+
             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 type(layer) == torch.nn.Linear:
                     layer.weight.data.normal_(mean=0.0, std=0.01)
                     layer.bias.data.zero_()
 
@@ -74,7 +80,7 @@ class HypernetworkModule(torch.nn.Module):
     def trainables(self):
         layer_structure = []
         for layer in self.linear:
-            if not "ReLU" in layer.__str__():
+            if type(layer) == torch.nn.Linear:
                 layer_structure += [layer.weight, layer.bias]
         return layer_structure
 
-- 
cgit v1.2.1


From 5245c7a4935f67b677da0f5a1fc2b74c074aa0e2 Mon Sep 17 00:00:00 2001
From: timntorres <timothynarcisotorres@gmail.com>
Date: Wed, 19 Oct 2022 12:21:32 -0700
Subject: Issue #2921-Give PNG info to Hypernet previews.

---
 modules/hypernetworks/hypernetwork.py | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 84e7e350..68c8f26d 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -256,6 +256,9 @@ def stack_conds(conds):
 
 
 def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
+    # images is required here to give training previews their infotext. Importing this at the very top causes a circular dependency.
+    from modules import images
+
     assert hypernetwork_name, 'hypernetwork not selected'
 
     path = shared.hypernetworks.get(hypernetwork_name, None)
@@ -298,6 +301,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
     last_saved_file = "<none>"
     last_saved_image = "<none>"
+    forced_filename = "<none>"
 
     ititial_step = hypernetwork.step or 0
     if ititial_step > steps:
@@ -345,7 +349,8 @@ 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:
-            last_saved_image = os.path.join(images_dir, f'{hypernetwork_name}-{hypernetwork.step}.png')
+            forced_filename = f'{hypernetwork_name}-{hypernetwork.step}'
+            last_saved_image = os.path.join(images_dir, forced_filename)
 
             optimizer.zero_grad()
             shared.sd_model.cond_stage_model.to(devices.device)
@@ -381,7 +386,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
             if image is not None:
                 shared.state.current_image = image
-                image.save(last_saved_image)
+                last_saved_image, last_text_info = images.save_image(image, images_dir, "", p.seed, p.prompt, shared.opts.samples_format, processed.infotexts[0], p=p, forced_filename=forced_filename)
                 last_saved_image += f", prompt: {preview_text}"
 
         shared.state.job_no = hypernetwork.step
-- 
cgit v1.2.1


From 4ff274e1e35bb642687253ce744d2cfa738ab293 Mon Sep 17 00:00:00 2001
From: timntorres <timothynarcisotorres@gmail.com>
Date: Wed, 19 Oct 2022 12:32:22 -0700
Subject: Revise comments.

---
 modules/hypernetworks/hypernetwork.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 68c8f26d..3f96361c 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -256,7 +256,7 @@ def stack_conds(conds):
 
 
 def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
-    # images is required here to give training previews their infotext. Importing this at the very top causes a circular dependency.
+    # images allows training previews to have infotext. Importing it at the top causes a circular import problem.
     from modules import images
 
     assert hypernetwork_name, 'hypernetwork not selected'
-- 
cgit v1.2.1


From 03a1e288c4973dd2dff57a97469b40f146b6fccf Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Fri, 21 Oct 2022 10:13:24 +0300
Subject: turns out LayerNorm also has weight and bias and needs to be
 pre-multiplied and trained for hypernets

---
 modules/hypernetworks/hypernetwork.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 3274a802..b1a5d0c7 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -52,7 +52,7 @@ class HypernetworkModule(torch.nn.Module):
             self.load_state_dict(state_dict)
         else:
             for layer in self.linear:
-                if type(layer) == torch.nn.Linear:
+                if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm:
                     layer.weight.data.normal_(mean=0.0, std=0.01)
                     layer.bias.data.zero_()
 
@@ -80,7 +80,7 @@ class HypernetworkModule(torch.nn.Module):
     def trainables(self):
         layer_structure = []
         for layer in self.linear:
-            if type(layer) == torch.nn.Linear:
+            if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm:
                 layer_structure += [layer.weight, layer.bias]
         return layer_structure
 
-- 
cgit v1.2.1


From 51e3dc9ccad157d7161b697a246e26c868d46a7c Mon Sep 17 00:00:00 2001
From: timntorres <timothynarcisotorres@gmail.com>
Date: Fri, 21 Oct 2022 02:11:12 -0700
Subject: Sanitize hypernet name input.

---
 modules/hypernetworks/ui.py | 3 +++
 1 file changed, 3 insertions(+)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py
index 266f04f6..e6f50a1f 100644
--- a/modules/hypernetworks/ui.py
+++ b/modules/hypernetworks/ui.py
@@ -11,6 +11,9 @@ from modules.hypernetworks import hypernetwork
 
 
 def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, add_layer_norm=False, activation_func=None):
+    # Remove illegal characters from name.
+    name = "".join( x for x in name if (x.isalnum() or x in "._- "))
+
     fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
     if not overwrite_old:
         assert not os.path.exists(fn), f"file {fn} already exists"
-- 
cgit v1.2.1


From 19818f023cfafc472c6c241cab0b72896a168481 Mon Sep 17 00:00:00 2001
From: timntorres <timothynarcisotorres@gmail.com>
Date: Fri, 21 Oct 2022 02:14:02 -0700
Subject: Match hypernet name with filename in all cases.

---
 modules/hypernetworks/hypernetwork.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index b1a5d0c7..6d392be4 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -340,7 +340,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         pbar.set_description(f"loss: {mean_loss:.7f}")
 
         if hypernetwork.step > 0 and hypernetwork_dir is not None and hypernetwork.step % save_hypernetwork_every == 0:
-            last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name}-{hypernetwork.step}.pt')
+            temp = hypernetwork.name
+            # Before saving, change name to match current checkpoint.
+            hypernetwork.name = f'{hypernetwork_name}-{hypernetwork.step}'
+            last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork.name}.pt')
             hypernetwork.save(last_saved_file)
 
         textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
@@ -405,6 +408,9 @@ Last saved image: {html.escape(last_saved_image)}<br/>
 
     hypernetwork.sd_checkpoint = checkpoint.hash
     hypernetwork.sd_checkpoint_name = checkpoint.model_name
+    # Before saving for the last time, change name back to the base name (as opposed to the save_hypernetwork_every step-suffixed naming convention).
+    hypernetwork.name = hypernetwork_name
+    filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork.name}.pt')
     hypernetwork.save(filename)
 
     return hypernetwork, filename
-- 
cgit v1.2.1


From 272fa527bbe93143668ffc16838107b7dca35b40 Mon Sep 17 00:00:00 2001
From: timntorres <timothynarcisotorres@gmail.com>
Date: Fri, 21 Oct 2022 02:41:55 -0700
Subject: Remove unused variable.

---
 modules/hypernetworks/hypernetwork.py | 1 -
 1 file changed, 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 6d392be4..47d91ea5 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -340,7 +340,6 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         pbar.set_description(f"loss: {mean_loss:.7f}")
 
         if hypernetwork.step > 0 and hypernetwork_dir is not None and hypernetwork.step % save_hypernetwork_every == 0:
-            temp = hypernetwork.name
             # Before saving, change name to match current checkpoint.
             hypernetwork.name = f'{hypernetwork_name}-{hypernetwork.step}'
             last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork.name}.pt')
-- 
cgit v1.2.1


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

---
 modules/hypernetworks/hypernetwork.py | 68 +++++++++++++++++++++--------------
 modules/hypernetworks/ui.py           | 10 +++---
 2 files changed, 47 insertions(+), 31 deletions(-)

(limited to 'modules/hypernetworks')

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 modules.textual_inversion.dataset
 import torch
-
-from ldm.util import default
-from modules import devices, shared, processing, sd_models
-import torch
-from torch import einsum
+import tqdm
 from einops import rearrange, repeat
-import modules.textual_inversion.dataset
+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)
 
-- 
cgit v1.2.1


From 7fd90128eb6d1820045bfe2c2c1269661023a712 Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Sat, 22 Oct 2022 14:48:43 +0300
Subject: added a guard for hypernet training that will stop early if weights
 are getting no gradients

---
 modules/hypernetworks/hypernetwork.py | 11 +++++++++++
 1 file changed, 11 insertions(+)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 47d91ea5..46039a49 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -310,6 +310,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
     optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
 
+    steps_without_grad = 0
+
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
@@ -332,8 +334,17 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
             losses[hypernetwork.step % losses.shape[0]] = loss.item()
 
             optimizer.zero_grad()
+            weights[0].grad = None
             loss.backward()
+
+            if weights[0].grad is None:
+                steps_without_grad += 1
+            else:
+                steps_without_grad = 0
+            assert steps_without_grad < 10, 'no gradient found for the trained weight after backward() for 10 steps in a row; this is a bug; training cannot continue'
+
             optimizer.step()
+
         mean_loss = losses.mean()
         if torch.isnan(mean_loss):
             raise RuntimeError("Loss diverged.")
-- 
cgit v1.2.1


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

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

(limited to 'modules/hypernetworks')

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)
-- 
cgit v1.2.1


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

---
 modules/hypernetworks/hypernetwork.py | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

(limited to 'modules/hypernetworks')

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:
-- 
cgit v1.2.1


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

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

(limited to 'modules/hypernetworks')

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)
 
-- 
cgit v1.2.1


From 24694e5983d0944b901892cb101878e6dec89a20 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 01:57:58 +0900
Subject: Update hypernetwork.py

---
 modules/hypernetworks/hypernetwork.py | 55 ++++++++++++++++++++++++++++-------
 1 file changed, 44 insertions(+), 11 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 3bc71ee5..81132be4 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -16,6 +16,7 @@ from modules.textual_inversion import textual_inversion
 from modules.textual_inversion.learn_schedule import LearnRateScheduler
 from torch import einsum
 
+from statistics import stdev, mean
 
 class HypernetworkModule(torch.nn.Module):
     multiplier = 1.0
@@ -268,6 +269,32 @@ def stack_conds(conds):
     return torch.stack(conds)
 
 
+def log_statistics(loss_info:dict, key, value):
+    if key not in loss_info:
+        loss_info[key] = [value]
+    else:
+        loss_info[key].append(value)
+        if len(loss_info) > 1024:
+            loss_info.pop(0)
+
+
+def statistics(data):
+    total_information = f"loss:{mean(data):.3f}"+u"\u00B1"+f"({stdev(data)/ (len(data)**0.5):.3f})"
+    recent_data = data[-32:]
+    recent_information = f"recent 32 loss:{mean(recent_data):.3f}"+u"\u00B1"+f"({stdev(recent_data)/ (len(recent_data)**0.5):.3f})"
+    return total_information, recent_information
+
+
+def report_statistics(loss_info:dict):
+    keys = sorted(loss_info.keys(), key=lambda x: sum(loss_info[x]) / len(loss_info[x]))
+    for key in keys:
+        info, recent = statistics(loss_info[key])
+        print("Loss statistics for file " + key)
+        print(info)
+        print(recent)
+
+
+
 def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
     # images allows training previews to have infotext. Importing it at the top causes a circular import problem.
     from modules import images
@@ -310,7 +337,11 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     for weight in weights:
         weight.requires_grad = True
 
-    losses = torch.zeros((32,))
+    size = len(ds.indexes)
+    loss_dict = {}
+    losses = torch.zeros((size,))
+    previous_mean_loss = 0
+    print("Mean loss of {} elements".format(size))
 
     last_saved_file = "<none>"
     last_saved_image = "<none>"
@@ -329,7 +360,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
-
+        if loss_dict and i % size == 0:
+            previous_mean_loss = sum(i[-1] for i in loss_dict.values()) / len(loss_dict)
+            
         scheduler.apply(optimizer, hypernetwork.step)
         if scheduler.finished:
             break
@@ -346,7 +379,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
             del c
 
             losses[hypernetwork.step % losses.shape[0]] = loss.item()
-
+            for entry in entries:
+                log_statistics(loss_dict, entry.filename, loss.item())
+                
             optimizer.zero_grad()
             weights[0].grad = None
             loss.backward()
@@ -359,10 +394,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
             optimizer.step()
 
-        mean_loss = losses.mean()
-        if torch.isnan(mean_loss):
+        if torch.isnan(losses[hypernetwork.step % losses.shape[0]]):
             raise RuntimeError("Loss diverged.")
-        pbar.set_description(f"loss: {mean_loss:.7f}")
+        pbar.set_description(f"dataset loss: {previous_mean_loss:.7f}")
 
         if hypernetwork.step > 0 and hypernetwork_dir is not None and hypernetwork.step % save_hypernetwork_every == 0:
             # Before saving, change name to match current checkpoint.
@@ -371,7 +405,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
             hypernetwork.save(last_saved_file)
 
         textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
-            "loss": f"{mean_loss:.7f}",
+            "loss": f"{previous_mean_loss:.7f}",
             "learn_rate": scheduler.learn_rate
         })
 
@@ -420,14 +454,15 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         shared.state.textinfo = f"""
 <p>
-Loss: {mean_loss:.7f}<br/>
+Loss: {previous_mean_loss:.7f}<br/>
 Step: {hypernetwork.step}<br/>
 Last prompt: {html.escape(entries[0].cond_text)}<br/>
 Last saved hypernetwork: {html.escape(last_saved_file)}<br/>
 Last saved image: {html.escape(last_saved_image)}<br/>
 </p>
 """
-
+        
+    report_statistics(loss_dict)
     checkpoint = sd_models.select_checkpoint()
 
     hypernetwork.sd_checkpoint = checkpoint.hash
@@ -438,5 +473,3 @@ Last saved image: {html.escape(last_saved_image)}<br/>
     hypernetwork.save(filename)
 
     return hypernetwork, filename
-
-
-- 
cgit v1.2.1


From 48dbf99e84045ee7af55bc5b1b86492a240e631e Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 04:17:16 +0900
Subject: Allow tracking real-time loss

Someone had 6000 images in their dataset, and it was shown as 0, which was confusing.
This will allow tracking real time dataset-average loss for registered objects.
---
 modules/hypernetworks/hypernetwork.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 81132be4..99fd0f8f 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -360,7 +360,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
-        if loss_dict and i % size == 0:
+        if len(loss_dict) > 0:
             previous_mean_loss = sum(i[-1] for i in loss_dict.values()) / len(loss_dict)
             
         scheduler.apply(optimizer, hypernetwork.step)
-- 
cgit v1.2.1


From 1fbfc052eb529d8cf8ce5baf578bcf93d0280c29 Mon Sep 17 00:00:00 2001
From: DepFA <35278260+dfaker@users.noreply.github.com>
Date: Sun, 23 Oct 2022 05:43:34 +0100
Subject: Update hypernetwork.py

---
 modules/hypernetworks/hypernetwork.py | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 99fd0f8f..98a7b62e 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -288,10 +288,13 @@ def statistics(data):
 def report_statistics(loss_info:dict):
     keys = sorted(loss_info.keys(), key=lambda x: sum(loss_info[x]) / len(loss_info[x]))
     for key in keys:
-        info, recent = statistics(loss_info[key])
-        print("Loss statistics for file " + key)
-        print(info)
-        print(recent)
+        try:
+            print("Loss statistics for file " + key)
+            info, recent = statistics(loss_info[key])
+            print(info)
+            print(recent)
+        except Exception as e:
+            print(e)
 
 
 
-- 
cgit v1.2.1


From b297cc3324979ec78d69b2d11dd18030dfad7bcc Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 20:06:42 +0900
Subject: Hypernetworks - fix KeyError in statistics caching

Statistics logging has changed to {filename : list[losses]}, so it has to use loss_info[key].pop()
---
 modules/hypernetworks/hypernetwork.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 98a7b62e..33827210 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -274,8 +274,8 @@ def log_statistics(loss_info:dict, key, value):
         loss_info[key] = [value]
     else:
         loss_info[key].append(value)
-        if len(loss_info) > 1024:
-            loss_info.pop(0)
+        if len(loss_info[key]) > 1024:
+            loss_info[key].pop(0)
 
 
 def statistics(data):
-- 
cgit v1.2.1


From 40b56c9289bf9458ae5ef3c1990ccea851c6c3e2 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 21:07:07 +0900
Subject: cleanup some code

---
 modules/hypernetworks/hypernetwork.py | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 33827210..4072bf54 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -16,6 +16,7 @@ from modules.textual_inversion import textual_inversion
 from modules.textual_inversion.learn_schedule import LearnRateScheduler
 from torch import einsum
 
+from collections import defaultdict, deque
 from statistics import stdev, mean
 
 class HypernetworkModule(torch.nn.Module):
@@ -269,15 +270,6 @@ def stack_conds(conds):
     return torch.stack(conds)
 
 
-def log_statistics(loss_info:dict, key, value):
-    if key not in loss_info:
-        loss_info[key] = [value]
-    else:
-        loss_info[key].append(value)
-        if len(loss_info[key]) > 1024:
-            loss_info[key].pop(0)
-
-
 def statistics(data):
     total_information = f"loss:{mean(data):.3f}"+u"\u00B1"+f"({stdev(data)/ (len(data)**0.5):.3f})"
     recent_data = data[-32:]
@@ -341,7 +333,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
         weight.requires_grad = True
 
     size = len(ds.indexes)
-    loss_dict = {}
+    loss_dict = defaultdict(lambda : deque(maxlen = 1024))
     losses = torch.zeros((size,))
     previous_mean_loss = 0
     print("Mean loss of {} elements".format(size))
@@ -383,7 +375,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
             losses[hypernetwork.step % losses.shape[0]] = loss.item()
             for entry in entries:
-                log_statistics(loss_dict, entry.filename, loss.item())
+                loss_dict[entry.filename].append(loss.item())
                 
             optimizer.zero_grad()
             weights[0].grad = None
-- 
cgit v1.2.1


From 348f89c8d40397c1875cff4a7331018785f9c3b8 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 21:29:53 +0900
Subject: statistics for pbar

---
 modules/hypernetworks/hypernetwork.py | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 4072bf54..48b56029 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -335,6 +335,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     size = len(ds.indexes)
     loss_dict = defaultdict(lambda : deque(maxlen = 1024))
     losses = torch.zeros((size,))
+    previous_mean_losses = [0]
     previous_mean_loss = 0
     print("Mean loss of {} elements".format(size))
 
@@ -356,7 +357,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
         if len(loss_dict) > 0:
-            previous_mean_loss = sum(i[-1] for i in loss_dict.values()) / len(loss_dict)
+            previous_mean_losses = [i[-1] for i in loss_dict.values()]
+            previous_mean_loss = mean(previous_mean_losses)
             
         scheduler.apply(optimizer, hypernetwork.step)
         if scheduler.finished:
@@ -391,7 +393,13 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         if torch.isnan(losses[hypernetwork.step % losses.shape[0]]):
             raise RuntimeError("Loss diverged.")
-        pbar.set_description(f"dataset loss: {previous_mean_loss:.7f}")
+        
+        if len(previous_mean_losses) > 1:
+            std = stdev(previous_mean_losses)
+        else:
+            std = 0
+        dataset_loss_info = f"dataset loss:{mean(previous_mean_losses):.3f}" + u"\u00B1" + f"({std / (len(previous_mean_losses) ** 0.5):.3f})"
+        pbar.set_description(dataset_loss_info)
 
         if hypernetwork.step > 0 and hypernetwork_dir is not None and hypernetwork.step % save_hypernetwork_every == 0:
             # Before saving, change name to match current checkpoint.
-- 
cgit v1.2.1


From 0d2e1dac407a0e2f5b148d314715f0457b2525b7 Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 21:41:39 +0900
Subject: convert deque -> list

I don't feel this being efficient
---
 modules/hypernetworks/hypernetwork.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index 48b56029..fb510fa7 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -282,7 +282,7 @@ def report_statistics(loss_info:dict):
     for key in keys:
         try:
             print("Loss statistics for file " + key)
-            info, recent = statistics(loss_info[key])
+            info, recent = statistics(list(loss_info[key]))
             print(info)
             print(recent)
         except Exception as e:
-- 
cgit v1.2.1


From e9a410b5357612f63528015c5533c2185dcff92e Mon Sep 17 00:00:00 2001
From: AngelBottomless <35677394+aria1th@users.noreply.github.com>
Date: Sun, 23 Oct 2022 21:47:39 +0900
Subject: check length for variance

---
 modules/hypernetworks/hypernetwork.py | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

(limited to 'modules/hypernetworks')

diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py
index fb510fa7..d647ea55 100644
--- a/modules/hypernetworks/hypernetwork.py
+++ b/modules/hypernetworks/hypernetwork.py
@@ -271,9 +271,17 @@ def stack_conds(conds):
 
 
 def statistics(data):
-    total_information = f"loss:{mean(data):.3f}"+u"\u00B1"+f"({stdev(data)/ (len(data)**0.5):.3f})"
+    if len(data) < 2:
+        std = 0
+    else:
+        std = stdev(data)
+    total_information = f"loss:{mean(data):.3f}" + u"\u00B1" + f"({std/ (len(data) ** 0.5):.3f})"
     recent_data = data[-32:]
-    recent_information = f"recent 32 loss:{mean(recent_data):.3f}"+u"\u00B1"+f"({stdev(recent_data)/ (len(recent_data)**0.5):.3f})"
+    if len(recent_data) < 2:
+        std = 0
+    else:
+        std = stdev(recent_data)
+    recent_information = f"recent 32 loss:{mean(recent_data):.3f}" + u"\u00B1" + f"({std / (len(recent_data) ** 0.5):.3f})"
     return total_information, recent_information
 
 
-- 
cgit v1.2.1