diff options
| -rw-r--r-- | CHANGELOG.md | 14 | ||||
| -rw-r--r-- | html/footer.html | 2 | ||||
| -rw-r--r-- | javascript/profilerVisualization.js | 153 | ||||
| -rw-r--r-- | javascript/ui_settings_hints.js | 2 | ||||
| -rw-r--r-- | modules/cmd_args.py | 2 | ||||
| -rw-r--r-- | modules/images.py | 4 | ||||
| -rw-r--r-- | modules/launch_utils.py | 8 | ||||
| -rw-r--r-- | modules/processing.py | 3 | ||||
| -rw-r--r-- | modules/script_callbacks.py | 4 | ||||
| -rw-r--r-- | modules/scripts.py | 3 | ||||
| -rw-r--r-- | modules/sd_hijack.py | 2 | ||||
| -rw-r--r-- | modules/sd_hijack_optimizations.py | 6 | ||||
| -rw-r--r-- | modules/sd_models.py | 2 | ||||
| -rw-r--r-- | modules/shared.py | 10 | ||||
| -rw-r--r-- | modules/timer.py | 46 | ||||
| -rw-r--r-- | modules/ui.py | 6 | ||||
| -rw-r--r-- | style.css | 14 | ||||
| -rw-r--r-- | webui.py | 14 |
18 files changed, 263 insertions, 32 deletions
diff --git a/CHANGELOG.md b/CHANGELOG.md index 6c6ab5e3..c3c57fe0 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,3 +1,17 @@ +## 1.3.1
+
+### Features:
+ * revert default cross attention optimization to Doggettx
+
+### Bug Fixes:
+ * fix bug: LoRA don't apply on dropdown list sd_lora
+ * fix png info always added even if setting is not enabled
+ * fix some fields not applying in xyz plot
+ * fix "hires. fix" prompt sharing same labels with txt2img_prompt
+ * fix lora hashes not being added properly to infotex if there is only one lora
+ * fix --use-cpu failing to work properly at startup
+ * make --disable-opt-split-attention command line option work again
+
## 1.3.0
### Features:
diff --git a/html/footer.html b/html/footer.html index bad87ff6..1ce13295 100644 --- a/html/footer.html +++ b/html/footer.html @@ -5,6 +5,8 @@ •
<a href="https://gradio.app">Gradio</a>
•
+ <a href="#" onclick="showProfile('./internal/profile-startup'); return false;">Startup profile</a>
+ •
<a href="/" onclick="javascript:gradioApp().getElementById('settings_restart_gradio').click(); return false">Reload UI</a>
</div>
<br />
diff --git a/javascript/profilerVisualization.js b/javascript/profilerVisualization.js new file mode 100644 index 00000000..9d8e5f42 --- /dev/null +++ b/javascript/profilerVisualization.js @@ -0,0 +1,153 @@ + +function createRow(table, cellName, items) { + var tr = document.createElement('tr'); + var res = []; + + items.forEach(function(x, i) { + if (x === undefined) { + res.push(null); + return; + } + + var td = document.createElement(cellName); + td.textContent = x; + tr.appendChild(td); + res.push(td); + + var colspan = 1; + for (var n = i + 1; n < items.length; n++) { + if (items[n] !== undefined) { + break; + } + + colspan += 1; + } + + if (colspan > 1) { + td.colSpan = colspan; + } + }); + + table.appendChild(tr); + + return res; +} + +function showProfile(path, cutoff = 0.05) { + requestGet(path, {}, function(data) { + var table = document.createElement('table'); + table.className = 'popup-table'; + + data.records['total'] = data.total; + var keys = Object.keys(data.records).sort(function(a, b) { + return data.records[b] - data.records[a]; + }); + var items = keys.map(function(x) { + return {key: x, parts: x.split('/'), time: data.records[x]}; + }); + var maxLength = items.reduce(function(a, b) { + return Math.max(a, b.parts.length); + }, 0); + + var cols = createRow(table, 'th', ['record', 'seconds']); + cols[0].colSpan = maxLength; + + function arraysEqual(a, b) { + return !(a < b || b < a); + } + + var addLevel = function(level, parent, hide) { + var matching = items.filter(function(x) { + return x.parts[level] && !x.parts[level + 1] && arraysEqual(x.parts.slice(0, level), parent); + }); + var sorted = matching.sort(function(a, b) { + return b.time - a.time; + }); + var othersTime = 0; + var othersList = []; + var othersRows = []; + var childrenRows = []; + sorted.forEach(function(x) { + var visible = x.time >= cutoff && !hide; + + var cells = []; + for (var i = 0; i < maxLength; i++) { + cells.push(x.parts[i]); + } + cells.push(x.time.toFixed(3)); + var cols = createRow(table, 'td', cells); + for (i = 0; i < level; i++) { + cols[i].className = 'muted'; + } + + var tr = cols[0].parentNode; + if (!visible) { + tr.classList.add("hidden"); + } + + if (x.time >= cutoff) { + childrenRows.push(tr); + } else { + othersTime += x.time; + othersList.push(x.parts[level]); + othersRows.push(tr); + } + + var children = addLevel(level + 1, parent.concat([x.parts[level]]), true); + if (children.length > 0) { + var cell = cols[level]; + var onclick = function() { + cell.classList.remove("link"); + cell.removeEventListener("click", onclick); + children.forEach(function(x) { + x.classList.remove("hidden"); + }); + }; + cell.classList.add("link"); + cell.addEventListener("click", onclick); + } + }); + + if (othersTime > 0) { + var cells = []; + for (var i = 0; i < maxLength; i++) { + cells.push(parent[i]); + } + cells.push(othersTime.toFixed(3)); + cells[level] = 'others'; + var cols = createRow(table, 'td', cells); + for (i = 0; i < level; i++) { + cols[i].className = 'muted'; + } + + var cell = cols[level]; + var tr = cell.parentNode; + var onclick = function() { + tr.classList.add("hidden"); + cell.classList.remove("link"); + cell.removeEventListener("click", onclick); + othersRows.forEach(function(x) { + x.classList.remove("hidden"); + }); + }; + + cell.title = othersList.join(", "); + cell.classList.add("link"); + cell.addEventListener("click", onclick); + + if (hide) { + tr.classList.add("hidden"); + } + + childrenRows.push(tr); + } + + return childrenRows; + }; + + addLevel(0, []); + + popup(table); + }); +} + diff --git a/javascript/ui_settings_hints.js b/javascript/ui_settings_hints.js index e216852b..d088f949 100644 --- a/javascript/ui_settings_hints.js +++ b/javascript/ui_settings_hints.js @@ -42,7 +42,7 @@ onOptionsChanged(function() { function settingsHintsShowQuicksettings() { requestGet("./internal/quicksettings-hint", {}, function(data) { var table = document.createElement('table'); - table.className = 'settings-value-table'; + table.className = 'popup-table'; data.forEach(function(obj) { var tr = document.createElement('tr'); diff --git a/modules/cmd_args.py b/modules/cmd_args.py index 0974056d..de905caa 100644 --- a/modules/cmd_args.py +++ b/modules/cmd_args.py @@ -62,7 +62,7 @@ parser.add_argument("--opt-split-attention-invokeai", action='store_true', help= parser.add_argument("--opt-split-attention-v1", action='store_true', help="prefer older version of split attention optimization for automatic choice of optimization")
parser.add_argument("--opt-sdp-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization for automatic choice of optimization; requires PyTorch 2.*")
parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization without memory efficient attention for automatic choice of optimization, makes image generation deterministic; requires PyTorch 2.*")
-parser.add_argument("--disable-opt-split-attention", action='store_true', help="does not do anything")
+parser.add_argument("--disable-opt-split-attention", action='store_true', help="prefer no cross-attention layer optimization for automatic choice of optimization")
parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI")
parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower)
parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests")
diff --git a/modules/images.py b/modules/images.py index 30e9ffc5..a12d252b 100644 --- a/modules/images.py +++ b/modules/images.py @@ -503,10 +503,6 @@ def save_image_with_geninfo(image, geninfo, filename, extension=None, existing_p image_format = Image.registered_extensions()[extension]
- existing_pnginfo = existing_pnginfo or {}
- if opts.enable_pnginfo:
- existing_pnginfo['parameters'] = geninfo
-
if extension.lower() == '.png':
if opts.enable_pnginfo:
pnginfo_data = PngImagePlugin.PngInfo()
diff --git a/modules/launch_utils.py b/modules/launch_utils.py index 6e9bb770..0c8c4db0 100644 --- a/modules/launch_utils.py +++ b/modules/launch_utils.py @@ -68,7 +68,13 @@ def git_tag(): try:
return subprocess.check_output([git, "describe", "--tags"], shell=False, encoding='utf8').strip()
except Exception:
- return "<none>"
+ try:
+ from pathlib import Path
+ changelog_md = Path(__file__).parent.parent / "CHANGELOG.md"
+ with changelog_md.open(encoding="utf-8") as file:
+ return next((line.strip() for line in file if line.strip()), "<none>")
+ except Exception:
+ return "<none>"
def run(command, desc=None, errdesc=None, custom_env=None, live: bool = default_command_live) -> str:
diff --git a/modules/processing.py b/modules/processing.py index baa9b278..362ab4c2 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -595,8 +595,7 @@ def process_images(p: StableDiffusionProcessing) -> Processed: try:
# if no checkpoint override or the override checkpoint can't be found, remove override entry and load opts checkpoint
- override_checkpoint = p.override_settings.get('sd_model_checkpoint')
- if override_checkpoint is not None and sd_models.checkpoint_alisases.get(override_checkpoint) is None:
+ if sd_models.checkpoint_alisases.get(p.override_settings.get('sd_model_checkpoint')) is None:
p.override_settings.pop('sd_model_checkpoint', None)
sd_models.reload_model_weights()
diff --git a/modules/script_callbacks.py b/modules/script_callbacks.py index ec1469d0..f755283c 100644 --- a/modules/script_callbacks.py +++ b/modules/script_callbacks.py @@ -1,11 +1,12 @@ import inspect
+import os
from collections import namedtuple
from typing import Optional, Dict, Any
from fastapi import FastAPI
from gradio import Blocks
-from modules import errors
+from modules import errors, timer
def report_exception(c, job):
@@ -123,6 +124,7 @@ def app_started_callback(demo: Optional[Blocks], app: FastAPI): for c in callback_map['callbacks_app_started']:
try:
c.callback(demo, app)
+ timer.startup_timer.record(os.path.basename(c.script))
except Exception:
report_exception(c, 'app_started_callback')
diff --git a/modules/scripts.py b/modules/scripts.py index b901862d..99bf836a 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -5,7 +5,7 @@ from collections import namedtuple import gradio as gr
-from modules import shared, paths, script_callbacks, extensions, script_loading, scripts_postprocessing, errors
+from modules import shared, paths, script_callbacks, extensions, script_loading, scripts_postprocessing, errors, timer
AlwaysVisible = object()
@@ -280,6 +280,7 @@ def load_scripts(): finally:
sys.path = syspath
current_basedir = paths.script_path
+ timer.startup_timer.record(scriptfile.filename)
global scripts_txt2img, scripts_img2img, scripts_postproc
diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 487dfd60..3b6f95ce 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -68,6 +68,8 @@ def apply_optimizations(option=None): if selection == "None":
matching_optimizer = None
+ elif selection == "Automatic" and shared.cmd_opts.disable_opt_split_attention:
+ matching_optimizer = None
elif matching_optimizer is None:
matching_optimizer = optimizers[0]
diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index 5f0ff513..b41aa419 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -57,7 +57,7 @@ class SdOptimizationSdpNoMem(SdOptimization): name = "sdp-no-mem"
label = "scaled dot product without memory efficient attention"
cmd_opt = "opt_sdp_no_mem_attention"
- priority = 90
+ priority = 80
def is_available(self):
return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention)
@@ -71,7 +71,7 @@ class SdOptimizationSdp(SdOptimizationSdpNoMem): name = "sdp"
label = "scaled dot product"
cmd_opt = "opt_sdp_attention"
- priority = 80
+ priority = 70
def apply(self):
ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward
@@ -114,7 +114,7 @@ class SdOptimizationInvokeAI(SdOptimization): class SdOptimizationDoggettx(SdOptimization):
name = "Doggettx"
cmd_opt = "opt_split_attention"
- priority = 20
+ priority = 90
def apply(self):
ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward
diff --git a/modules/sd_models.py b/modules/sd_models.py index 232eb9c4..918f6fd6 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -314,8 +314,6 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer timer.record("apply half()")
- devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
- devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
devices.dtype_unet = model.model.diffusion_model.dtype
devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16
diff --git a/modules/shared.py b/modules/shared.py index 3c7ae654..7025a754 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -6,6 +6,7 @@ import threading import time
import gradio as gr
+import torch
import tqdm
import modules.interrogate
@@ -63,6 +64,9 @@ cmd_opts.disable_extension_access = (cmd_opts.share or cmd_opts.listen or cmd_op devices.device, devices.device_interrogate, devices.device_gfpgan, devices.device_esrgan, devices.device_codeformer = \
(devices.cpu if any(y in cmd_opts.use_cpu for y in [x, 'all']) else devices.get_optimal_device() for x in ['sd', 'interrogate', 'gfpgan', 'esrgan', 'codeformer'])
+devices.dtype = torch.float32 if cmd_opts.no_half else torch.float16
+devices.dtype_vae = torch.float32 if cmd_opts.no_half or cmd_opts.no_half_vae else torch.float16
+
device = devices.device
weight_load_location = None if cmd_opts.lowram else "cpu"
@@ -474,7 +478,7 @@ options_templates.update(options_section(('ui', "User interface"), { "quicksettings_list": OptionInfo(["sd_model_checkpoint"], "Quicksettings list", ui_components.DropdownMulti, lambda: {"choices": list(opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that appear at the top of page rather than in settings tab").needs_restart(),
"ui_tab_order": OptionInfo([], "UI tab order", ui_components.DropdownMulti, lambda: {"choices": list(tab_names)}).needs_restart(),
"hidden_tabs": OptionInfo([], "Hidden UI tabs", ui_components.DropdownMulti, lambda: {"choices": list(tab_names)}).needs_restart(),
- "ui_reorder": OptionInfo([], "txt2img/img2img UI item order", ui_components.DropdownMulti, lambda: {"choices": list(shared_items.ui_reorder_categories())}).info("selected items appear first").needs_restart(),
+ "ui_reorder_list": OptionInfo([], "txt2img/img2img UI item order", ui_components.DropdownMulti, lambda: {"choices": list(shared_items.ui_reorder_categories())}).info("selected items appear first").needs_restart(),
"hires_fix_show_sampler": OptionInfo(False, "Hires fix: show hires sampler selection").needs_restart(),
"hires_fix_show_prompts": OptionInfo(False, "Hires fix: show hires prompt and negative prompt").needs_restart(),
"disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_restart(),
@@ -626,8 +630,8 @@ class Options: self.data['quicksettings_list'] = [i.strip() for i in self.data.get('quicksettings').split(',')]
# 1.4.0 ui_reorder
- if isinstance(self.data.get('ui_reorder'), str):
- self.data['ui_reorder'] = [i.strip() for i in self.data.get('ui_reorder').split(',')]
+ if isinstance(self.data.get('ui_reorder'), str) and self.data.get('ui_reorder') and "ui_reorder_list" not in self.data:
+ self.data['ui_reorder_list'] = [i.strip() for i in self.data.get('ui_reorder').split(',')]
bad_settings = 0
for k, v in self.data.items():
diff --git a/modules/timer.py b/modules/timer.py index ba92be33..da99e49f 100644 --- a/modules/timer.py +++ b/modules/timer.py @@ -1,11 +1,30 @@ import time
+class TimerSubcategory:
+ def __init__(self, timer, category):
+ self.timer = timer
+ self.category = category
+ self.start = None
+ self.original_base_category = timer.base_category
+
+ def __enter__(self):
+ self.start = time.time()
+ self.timer.base_category = self.original_base_category + self.category + "/"
+
+ def __exit__(self, exc_type, exc_val, exc_tb):
+ elapsed_for_subcategroy = time.time() - self.start
+ self.timer.base_category = self.original_base_category
+ self.timer.add_time_to_record(self.original_base_category + self.category, elapsed_for_subcategroy)
+ self.timer.record(self.category)
+
+
class Timer:
def __init__(self):
self.start = time.time()
self.records = {}
self.total = 0
+ self.base_category = ''
def elapsed(self):
end = time.time()
@@ -13,18 +32,29 @@ class Timer: self.start = end
return res
- def record(self, category, extra_time=0):
- e = self.elapsed()
+ def add_time_to_record(self, category, amount):
if category not in self.records:
self.records[category] = 0
- self.records[category] += e + extra_time
+ self.records[category] += amount
+
+ def record(self, category, extra_time=0):
+ e = self.elapsed()
+
+ self.add_time_to_record(self.base_category + category, e + extra_time)
+
self.total += e + extra_time
+ def subcategory(self, name):
+ self.elapsed()
+
+ subcat = TimerSubcategory(self, name)
+ return subcat
+
def summary(self):
res = f"{self.total:.1f}s"
- additions = [x for x in self.records.items() if x[1] >= 0.1]
+ additions = [(category, time_taken) for category, time_taken in self.records.items() if time_taken >= 0.1 and '/' not in category]
if not additions:
return res
@@ -34,5 +64,13 @@ class Timer: return res
+ def dump(self):
+ return {'total': self.total, 'records': self.records}
+
def reset(self):
self.__init__()
+
+
+startup_timer = Timer()
+
+startup_record = None
diff --git a/modules/ui.py b/modules/ui.py index 4e0cf776..b7459f08 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -11,7 +11,7 @@ import numpy as np from PIL import Image, PngImagePlugin # noqa: F401
from modules.call_queue import wrap_gradio_gpu_call, wrap_queued_call, wrap_gradio_call
-from modules import sd_hijack, sd_models, script_callbacks, ui_extensions, deepbooru, sd_vae, extra_networks, ui_common, ui_postprocessing, progress, ui_loadsave, errors, shared_items, ui_settings
+from modules import sd_hijack, sd_models, script_callbacks, ui_extensions, deepbooru, sd_vae, extra_networks, ui_common, ui_postprocessing, progress, ui_loadsave, errors, shared_items, ui_settings, timer
from modules.ui_components import FormRow, FormGroup, ToolButton, FormHTML
from modules.paths import script_path
from modules.ui_common import create_refresh_button
@@ -388,7 +388,7 @@ def create_sampler_and_steps_selection(choices, tabname): def ordered_ui_categories():
- user_order = {x.strip(): i * 2 + 1 for i, x in enumerate(shared.opts.ui_reorder)}
+ user_order = {x.strip(): i * 2 + 1 for i, x in enumerate(shared.opts.ui_reorder_list)}
for _, category in sorted(enumerate(shared_items.ui_reorder_categories()), key=lambda x: user_order.get(x[1], x[0] * 2 + 0)):
yield category
@@ -1595,3 +1595,5 @@ def setup_ui_api(app): app.add_api_route("/internal/quicksettings-hint", quicksettings_hint, methods=["GET"], response_model=List[QuicksettingsHint])
app.add_api_route("/internal/ping", lambda: {}, methods=["GET"])
+
+ app.add_api_route("/internal/profile-startup", lambda: timer.startup_record, methods=["GET"])
@@ -403,19 +403,29 @@ div#extras_scale_to_tab div.form{ margin: 0 1.2em;
}
-table.settings-value-table{
+table.popup-table{
background: white;
border-collapse: collapse;
margin: 1em;
border: 4px solid white;
}
-table.settings-value-table td{
+table.popup-table td{
padding: 0.4em;
border: 1px solid #ccc;
max-width: 36em;
}
+table.popup-table .muted{
+ color: #aaa;
+}
+
+table.popup-table .link{
+ text-decoration: underline;
+ cursor: pointer;
+ font-weight: bold;
+}
+
.ui-defaults-none{
color: #aaa !important;
}
@@ -22,7 +22,7 @@ logging.getLogger("xformers").addFilter(lambda record: 'A matching Triton is not from modules import paths, timer, import_hook, errors, devices # noqa: F401
-startup_timer = timer.Timer()
+startup_timer = timer.startup_timer
import torch
import pytorch_lightning # noqa: F401 # pytorch_lightning should be imported after torch, but it re-enables warnings on import so import once to disable them
@@ -272,8 +272,8 @@ def initialize_rest(*, reload_script_modules=False): localization.list_localizations(cmd_opts.localizations_dir)
- modules.scripts.load_scripts()
- startup_timer.record("load scripts")
+ with startup_timer.subcategory("load scripts"):
+ modules.scripts.load_scripts()
if reload_script_modules:
for module in [module for name, module in sys.modules.items() if name.startswith("modules.ui")]:
@@ -428,9 +428,12 @@ def webui(): ui_extra_networks.add_pages_to_demo(app)
- modules.script_callbacks.app_started_callback(shared.demo, app)
- startup_timer.record("scripts app_started_callback")
+ startup_timer.record("add APIs")
+
+ with startup_timer.subcategory("app_started_callback"):
+ modules.script_callbacks.app_started_callback(shared.demo, app)
+ timer.startup_record = startup_timer.dump()
print(f"Startup time: {startup_timer.summary()}.")
if cmd_opts.subpath:
@@ -455,6 +458,7 @@ def webui(): # If we catch a keyboard interrupt, we want to stop the server and exit.
shared.demo.close()
break
+
print('Restarting UI...')
shared.demo.close()
time.sleep(0.5)
|