From 762265eab58cdb8f2d6398769bab43d8b8db0075 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Wed, 10 May 2023 07:52:45 +0300 Subject: autofixes from ruff --- modules/sd_hijack_optimizations.py | 1 - 1 file changed, 1 deletion(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index f10865cd..b623d53d 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -296,7 +296,6 @@ def sub_quad_attention(q, k, v, q_chunk_size=1024, kv_chunk_size=None, kv_chunk_ if chunk_threshold_bytes is not None and qk_matmul_size_bytes <= chunk_threshold_bytes: # the big matmul fits into our memory limit; do everything in 1 chunk, # i.e. send it down the unchunked fast-path - query_chunk_size = q_tokens kv_chunk_size = k_tokens with devices.without_autocast(disable=q.dtype == v.dtype): -- cgit v1.2.3 From 028d3f6425d85f122027c127fba8bcbf4f66ee75 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Wed, 10 May 2023 11:05:02 +0300 Subject: ruff auto fixes --- extensions-builtin/LDSR/sd_hijack_autoencoder.py | 4 ++-- extensions-builtin/LDSR/sd_hijack_ddpm_v1.py | 12 ++++++------ extensions-builtin/Lora/lora.py | 12 ++++++------ extensions-builtin/Lora/scripts/lora_script.py | 2 +- modules/config_states.py | 2 +- modules/deepbooru.py | 2 +- modules/devices.py | 2 +- modules/hypernetworks/hypernetwork.py | 2 +- modules/hypernetworks/ui.py | 4 ++-- modules/interrogate.py | 2 +- modules/modelloader.py | 2 +- modules/models/diffusion/ddpm_edit.py | 4 ++-- modules/scripts_auto_postprocessing.py | 2 +- modules/sd_hijack.py | 2 +- modules/sd_hijack_optimizations.py | 14 +++++++------- modules/sd_samplers_compvis.py | 2 +- modules/sd_samplers_kdiffusion.py | 2 +- modules/shared.py | 6 +++--- modules/textual_inversion/textual_inversion.py | 2 +- modules/ui.py | 8 ++++---- modules/ui_extra_networks.py | 4 ++-- modules/ui_tempdir.py | 2 +- 22 files changed, 47 insertions(+), 47 deletions(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/extensions-builtin/LDSR/sd_hijack_autoencoder.py b/extensions-builtin/LDSR/sd_hijack_autoencoder.py index 6303fed5..f457ca93 100644 --- a/extensions-builtin/LDSR/sd_hijack_autoencoder.py +++ b/extensions-builtin/LDSR/sd_hijack_autoencoder.py @@ -288,5 +288,5 @@ class VQModelInterface(VQModel): dec = self.decoder(quant) return dec -setattr(ldm.models.autoencoder, "VQModel", VQModel) -setattr(ldm.models.autoencoder, "VQModelInterface", VQModelInterface) +ldm.models.autoencoder.VQModel = VQModel +ldm.models.autoencoder.VQModelInterface = VQModelInterface diff --git a/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py index 4d3f6c56..d8fc30e3 100644 --- a/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py +++ b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py @@ -1116,7 +1116,7 @@ class LatentDiffusionV1(DDPMV1): if cond is not None: if isinstance(cond, dict): cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else - list(map(lambda x: x[:batch_size], cond[key])) for key in cond} + [x[:batch_size] for x in cond[key]] for key in cond} else: cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] @@ -1215,7 +1215,7 @@ class LatentDiffusionV1(DDPMV1): if cond is not None: if isinstance(cond, dict): cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else - list(map(lambda x: x[:batch_size], cond[key])) for key in cond} + [x[:batch_size] for x in cond[key]] for key in cond} else: cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] return self.p_sample_loop(cond, @@ -1437,7 +1437,7 @@ class Layout2ImgDiffusionV1(LatentDiffusionV1): logs['bbox_image'] = cond_img return logs -setattr(ldm.models.diffusion.ddpm, "DDPMV1", DDPMV1) -setattr(ldm.models.diffusion.ddpm, "LatentDiffusionV1", LatentDiffusionV1) -setattr(ldm.models.diffusion.ddpm, "DiffusionWrapperV1", DiffusionWrapperV1) -setattr(ldm.models.diffusion.ddpm, "Layout2ImgDiffusionV1", Layout2ImgDiffusionV1) +ldm.models.diffusion.ddpm.DDPMV1 = DDPMV1 +ldm.models.diffusion.ddpm.LatentDiffusionV1 = LatentDiffusionV1 +ldm.models.diffusion.ddpm.DiffusionWrapperV1 = DiffusionWrapperV1 +ldm.models.diffusion.ddpm.Layout2ImgDiffusionV1 = Layout2ImgDiffusionV1 diff --git a/extensions-builtin/Lora/lora.py b/extensions-builtin/Lora/lora.py index 0ab43229..9795540f 100644 --- a/extensions-builtin/Lora/lora.py +++ b/extensions-builtin/Lora/lora.py @@ -172,7 +172,7 @@ def load_lora(name, filename): else: print(f'Lora layer {key_diffusers} matched a layer with unsupported type: {type(sd_module).__name__}') continue - assert False, f'Lora layer {key_diffusers} matched a layer with unsupported type: {type(sd_module).__name__}' + raise AssertionError(f"Lora layer {key_diffusers} matched a layer with unsupported type: {type(sd_module).__name__}") with torch.no_grad(): module.weight.copy_(weight) @@ -184,7 +184,7 @@ def load_lora(name, filename): elif lora_key == "lora_down.weight": lora_module.down = module else: - assert False, f'Bad Lora layer name: {key_diffusers} - must end in lora_up.weight, lora_down.weight or alpha' + raise AssertionError(f"Bad Lora layer name: {key_diffusers} - must end in lora_up.weight, lora_down.weight or alpha") if len(keys_failed_to_match) > 0: print(f"Failed to match keys when loading Lora {filename}: {keys_failed_to_match}") @@ -202,7 +202,7 @@ def load_loras(names, multipliers=None): loaded_loras.clear() loras_on_disk = [available_lora_aliases.get(name, None) for name in names] - if any([x is None for x in loras_on_disk]): + if any(x is None for x in loras_on_disk): list_available_loras() loras_on_disk = [available_lora_aliases.get(name, None) for name in names] @@ -309,7 +309,7 @@ def lora_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.Mu print(f'failed to calculate lora weights for layer {lora_layer_name}') - setattr(self, "lora_current_names", wanted_names) + self.lora_current_names = wanted_names def lora_forward(module, input, original_forward): @@ -343,8 +343,8 @@ def lora_forward(module, input, original_forward): def lora_reset_cached_weight(self: Union[torch.nn.Conv2d, torch.nn.Linear]): - setattr(self, "lora_current_names", ()) - setattr(self, "lora_weights_backup", None) + self.lora_current_names = () + self.lora_weights_backup = None def lora_Linear_forward(self, input): diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index 7db971fd..b70e2de7 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -53,7 +53,7 @@ script_callbacks.on_infotext_pasted(lora.infotext_pasted) shared.options_templates.update(shared.options_section(('extra_networks', "Extra Networks"), { - "sd_lora": shared.OptionInfo("None", "Add Lora to prompt", gr.Dropdown, lambda: {"choices": ["None"] + [x for x in lora.available_loras]}, refresh=lora.list_available_loras), + "sd_lora": shared.OptionInfo("None", "Add Lora to prompt", gr.Dropdown, lambda: {"choices": ["None"] + list(lora.available_loras)}, refresh=lora.list_available_loras), })) diff --git a/modules/config_states.py b/modules/config_states.py index 8f1ff428..75da862a 100644 --- a/modules/config_states.py +++ b/modules/config_states.py @@ -35,7 +35,7 @@ def list_config_states(): j["filepath"] = path config_states.append(j) - config_states = list(sorted(config_states, key=lambda cs: cs["created_at"], reverse=True)) + config_states = sorted(config_states, key=lambda cs: cs["created_at"], reverse=True) for cs in config_states: timestamp = time.asctime(time.gmtime(cs["created_at"])) diff --git a/modules/deepbooru.py b/modules/deepbooru.py index 1c4554a2..547e1b4c 100644 --- a/modules/deepbooru.py +++ b/modules/deepbooru.py @@ -78,7 +78,7 @@ class DeepDanbooru: res = [] - filtertags = set([x.strip().replace(' ', '_') for x in shared.opts.deepbooru_filter_tags.split(",")]) + filtertags = {x.strip().replace(' ', '_') for x in shared.opts.deepbooru_filter_tags.split(",")} for tag in [x for x in tags if x not in filtertags]: probability = probability_dict[tag] diff --git a/modules/devices.py b/modules/devices.py index c705a3cb..d8a34a0f 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -65,7 +65,7 @@ def enable_tf32(): # enabling benchmark option seems to enable a range of cards to do fp16 when they otherwise can't # see https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/4407 - if any([torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())]): + if any(torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())): torch.backends.cudnn.benchmark = True torch.backends.cuda.matmul.allow_tf32 = True diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 9fe749b7..6ef0bfdf 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -403,7 +403,7 @@ def attention_CrossAttention_forward(self, x, context=None, mask=None): k = self.to_k(context_k) v = self.to_v(context_v) - q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v)) + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q, k, v)) sim = einsum('b i d, b j d -> b i j', q, k) * self.scale diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py index be168736..e3f9eb13 100644 --- a/modules/hypernetworks/ui.py +++ b/modules/hypernetworks/ui.py @@ -5,13 +5,13 @@ import modules.hypernetworks.hypernetwork from modules import devices, sd_hijack, shared not_available = ["hardswish", "multiheadattention"] -keys = list(x for x in modules.hypernetworks.hypernetwork.HypernetworkModule.activation_dict.keys() if x not in not_available) +keys = [x for x in modules.hypernetworks.hypernetwork.HypernetworkModule.activation_dict.keys() if x not in not_available] def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, dropout_structure=None): filename = modules.hypernetworks.hypernetwork.create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure, activation_func, weight_init, add_layer_norm, use_dropout, dropout_structure) - return gr.Dropdown.update(choices=sorted([x for x in shared.hypernetworks.keys()])), f"Created: {filename}", "" + return gr.Dropdown.update(choices=sorted(shared.hypernetworks.keys())), f"Created: {filename}", "" def train_hypernetwork(*args): diff --git a/modules/interrogate.py b/modules/interrogate.py index 22df9216..a1c8e537 100644 --- a/modules/interrogate.py +++ b/modules/interrogate.py @@ -159,7 +159,7 @@ class InterrogateModels: text_array = text_array[0:int(shared.opts.interrogate_clip_dict_limit)] top_count = min(top_count, len(text_array)) - text_tokens = clip.tokenize([text for text in text_array], truncate=True).to(devices.device_interrogate) + text_tokens = clip.tokenize(list(text_array), truncate=True).to(devices.device_interrogate) text_features = self.clip_model.encode_text(text_tokens).type(self.dtype) text_features /= text_features.norm(dim=-1, keepdim=True) diff --git a/modules/modelloader.py b/modules/modelloader.py index 92ada694..25612bf8 100644 --- a/modules/modelloader.py +++ b/modules/modelloader.py @@ -39,7 +39,7 @@ def load_models(model_path: str, model_url: str = None, command_path: str = None if os.path.islink(full_path) and not os.path.exists(full_path): print(f"Skipping broken symlink: {full_path}") continue - if ext_blacklist is not None and any([full_path.endswith(x) for x in ext_blacklist]): + if ext_blacklist is not None and any(full_path.endswith(x) for x in ext_blacklist): continue if full_path not in output: output.append(full_path) diff --git a/modules/models/diffusion/ddpm_edit.py b/modules/models/diffusion/ddpm_edit.py index 611c2b69..09432117 100644 --- a/modules/models/diffusion/ddpm_edit.py +++ b/modules/models/diffusion/ddpm_edit.py @@ -1130,7 +1130,7 @@ class LatentDiffusion(DDPM): if cond is not None: if isinstance(cond, dict): cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else - list(map(lambda x: x[:batch_size], cond[key])) for key in cond} + [x[:batch_size] for x in cond[key]] for key in cond} else: cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] @@ -1229,7 +1229,7 @@ class LatentDiffusion(DDPM): if cond is not None: if isinstance(cond, dict): cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else - list(map(lambda x: x[:batch_size], cond[key])) for key in cond} + [x[:batch_size] for x in cond[key]] for key in cond} else: cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] return self.p_sample_loop(cond, diff --git a/modules/scripts_auto_postprocessing.py b/modules/scripts_auto_postprocessing.py index 30d6d658..d63078de 100644 --- a/modules/scripts_auto_postprocessing.py +++ b/modules/scripts_auto_postprocessing.py @@ -17,7 +17,7 @@ class ScriptPostprocessingForMainUI(scripts.Script): return self.postprocessing_controls.values() def postprocess_image(self, p, script_pp, *args): - args_dict = {k: v for k, v in zip(self.postprocessing_controls, args)} + args_dict = dict(zip(self.postprocessing_controls, args)) pp = scripts_postprocessing.PostprocessedImage(script_pp.image) pp.info = {} diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 81573b78..e374aeb8 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -37,7 +37,7 @@ def apply_optimizations(): optimization_method = None - can_use_sdp = hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(getattr(torch.nn.functional, "scaled_dot_product_attention")) # not everyone has torch 2.x to use sdp + can_use_sdp = hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) # not everyone has torch 2.x to use sdp if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)): print("Applying xformers cross attention optimization.") diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index b623d53d..a174bbe1 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -49,7 +49,7 @@ def split_cross_attention_forward_v1(self, x, context=None, mask=None): v_in = self.to_v(context_v) del context, context_k, context_v, x - q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q_in, k_in, v_in)) + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q_in, k_in, v_in)) del q_in, k_in, v_in dtype = q.dtype @@ -98,7 +98,7 @@ def split_cross_attention_forward(self, x, context=None, mask=None): del context, x - q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q_in, k_in, v_in)) + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q_in, k_in, v_in)) del q_in, k_in, v_in r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype) @@ -229,7 +229,7 @@ def split_cross_attention_forward_invokeAI(self, x, context=None, mask=None): with devices.without_autocast(disable=not shared.opts.upcast_attn): k = k * self.scale - q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v)) + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q, k, v)) r = einsum_op(q, k, v) r = r.to(dtype) return self.to_out(rearrange(r, '(b h) n d -> b n (h d)', h=h)) @@ -334,7 +334,7 @@ def xformers_attention_forward(self, x, context=None, mask=None): k_in = self.to_k(context_k) v_in = self.to_v(context_v) - q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b n h d', h=h), (q_in, k_in, v_in)) + q, k, v = (rearrange(t, 'b n (h d) -> b n h d', h=h) for t in (q_in, k_in, v_in)) del q_in, k_in, v_in dtype = q.dtype @@ -460,7 +460,7 @@ def xformers_attnblock_forward(self, x): k = self.k(h_) v = self.v(h_) b, c, h, w = q.shape - q, k, v = map(lambda t: rearrange(t, 'b c h w -> b (h w) c'), (q, k, v)) + q, k, v = (rearrange(t, 'b c h w -> b (h w) c') for t in (q, k, v)) dtype = q.dtype if shared.opts.upcast_attn: q, k = q.float(), k.float() @@ -482,7 +482,7 @@ def sdp_attnblock_forward(self, x): k = self.k(h_) v = self.v(h_) b, c, h, w = q.shape - q, k, v = map(lambda t: rearrange(t, 'b c h w -> b (h w) c'), (q, k, v)) + q, k, v = (rearrange(t, 'b c h w -> b (h w) c') for t in (q, k, v)) dtype = q.dtype if shared.opts.upcast_attn: q, k = q.float(), k.float() @@ -506,7 +506,7 @@ def sub_quad_attnblock_forward(self, x): k = self.k(h_) v = self.v(h_) b, c, h, w = q.shape - q, k, v = map(lambda t: rearrange(t, 'b c h w -> b (h w) c'), (q, k, v)) + q, k, v = (rearrange(t, 'b c h w -> b (h w) c') for t in (q, k, v)) q = q.contiguous() k = k.contiguous() v = v.contiguous() diff --git a/modules/sd_samplers_compvis.py b/modules/sd_samplers_compvis.py index bfcc5574..7427648f 100644 --- a/modules/sd_samplers_compvis.py +++ b/modules/sd_samplers_compvis.py @@ -83,7 +83,7 @@ class VanillaStableDiffusionSampler: conds_list, tensor = prompt_parser.reconstruct_multicond_batch(cond, self.step) unconditional_conditioning = prompt_parser.reconstruct_cond_batch(unconditional_conditioning, self.step) - assert all([len(conds) == 1 for conds in conds_list]), 'composition via AND is not supported for DDIM/PLMS samplers' + assert all(len(conds) == 1 for conds in conds_list), 'composition via AND is not supported for DDIM/PLMS samplers' cond = tensor # for DDIM, shapes must match, we can't just process cond and uncond independently; diff --git a/modules/sd_samplers_kdiffusion.py b/modules/sd_samplers_kdiffusion.py index 3b8e9622..2f733cf5 100644 --- a/modules/sd_samplers_kdiffusion.py +++ b/modules/sd_samplers_kdiffusion.py @@ -86,7 +86,7 @@ class CFGDenoiser(torch.nn.Module): conds_list, tensor = prompt_parser.reconstruct_multicond_batch(cond, self.step) uncond = prompt_parser.reconstruct_cond_batch(uncond, self.step) - assert not is_edit_model or all([len(conds) == 1 for conds in conds_list]), "AND is not supported for InstructPix2Pix checkpoint (unless using Image CFG scale = 1.0)" + assert not is_edit_model or all(len(conds) == 1 for conds in conds_list), "AND is not supported for InstructPix2Pix checkpoint (unless using Image CFG scale = 1.0)" batch_size = len(conds_list) repeats = [len(conds_list[i]) for i in range(batch_size)] diff --git a/modules/shared.py b/modules/shared.py index 7d70f041..e2691585 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -381,7 +381,7 @@ options_templates.update(options_section(('extra_networks', "Extra Networks"), { "extra_networks_card_width": OptionInfo(0, "Card width for Extra Networks (px)"), "extra_networks_card_height": OptionInfo(0, "Card height for Extra Networks (px)"), "extra_networks_add_text_separator": OptionInfo(" ", "Extra text to add before <...> when adding extra network to prompt"), - "sd_hypernetwork": OptionInfo("None", "Add hypernetwork to prompt", gr.Dropdown, lambda: {"choices": ["None"] + [x for x in hypernetworks.keys()]}, refresh=reload_hypernetworks), + "sd_hypernetwork": OptionInfo("None", "Add hypernetwork to prompt", gr.Dropdown, lambda: {"choices": ["None"] + list(hypernetworks.keys())}, refresh=reload_hypernetworks), })) options_templates.update(options_section(('ui', "User interface"), { @@ -403,7 +403,7 @@ options_templates.update(options_section(('ui', "User interface"), { "keyedit_precision_extra": OptionInfo(0.05, "Ctrl+up/down precision when editing ", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), "keyedit_delimiters": OptionInfo(".,\\/!?%^*;:{}=`~()", "Ctrl+up/down word delimiters"), "quicksettings_list": OptionInfo(["sd_model_checkpoint"], "Quicksettings list", ui_components.DropdownMulti, lambda: {"choices": list(opts.data_labels.keys())}), - "hidden_tabs": OptionInfo([], "Hidden UI tabs (requires restart)", ui_components.DropdownMulti, lambda: {"choices": [x for x in tab_names]}), + "hidden_tabs": OptionInfo([], "Hidden UI tabs (requires restart)", ui_components.DropdownMulti, lambda: {"choices": list(tab_names)}), "ui_reorder": OptionInfo(", ".join(ui_reorder_categories), "txt2img/img2img UI item order"), "ui_extra_networks_tab_reorder": OptionInfo("", "Extra networks tab order"), "localization": OptionInfo("None", "Localization (requires restart)", gr.Dropdown, lambda: {"choices": ["None"] + list(localization.localizations.keys())}, refresh=lambda: localization.list_localizations(cmd_opts.localizations_dir)), @@ -583,7 +583,7 @@ class Options: if item.section not in section_ids: section_ids[item.section] = len(section_ids) - self.data_labels = {k: v for k, v in sorted(settings_items, key=lambda x: section_ids[x[1].section])} + self.data_labels = dict(sorted(settings_items, key=lambda x: section_ids[x[1].section])) def cast_value(self, key, value): """casts an arbitrary to the same type as this setting's value with key diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py index 9ed9ba45..c37bb2ad 100644 --- a/modules/textual_inversion/textual_inversion.py +++ b/modules/textual_inversion/textual_inversion.py @@ -167,7 +167,7 @@ class EmbeddingDatabase: if 'string_to_param' in data: param_dict = data['string_to_param'] if hasattr(param_dict, '_parameters'): - param_dict = getattr(param_dict, '_parameters') # fix for torch 1.12.1 loading saved file from torch 1.11 + param_dict = param_dict._parameters # fix for torch 1.12.1 loading saved file from torch 1.11 assert len(param_dict) == 1, 'embedding file has multiple terms in it' emb = next(iter(param_dict.items()))[1] # diffuser concepts diff --git a/modules/ui.py b/modules/ui.py index 782b569d..84d661b2 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -1222,7 +1222,7 @@ def create_ui(): ) def get_textual_inversion_template_names(): - return sorted([x for x in textual_inversion.textual_inversion_templates]) + return sorted(textual_inversion.textual_inversion_templates) with gr.Tab(label="Train", id="train"): gr.HTML(value="

Train an embedding or Hypernetwork; you must specify a directory with a set of 1:1 ratio images [wiki]

") @@ -1230,8 +1230,8 @@ def create_ui(): train_embedding_name = gr.Dropdown(label='Embedding', elem_id="train_embedding", choices=sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys())) create_refresh_button(train_embedding_name, sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings, lambda: {"choices": sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys())}, "refresh_train_embedding_name") - train_hypernetwork_name = gr.Dropdown(label='Hypernetwork', elem_id="train_hypernetwork", choices=[x for x in shared.hypernetworks.keys()]) - create_refresh_button(train_hypernetwork_name, shared.reload_hypernetworks, lambda: {"choices": sorted([x for x in shared.hypernetworks.keys()])}, "refresh_train_hypernetwork_name") + train_hypernetwork_name = gr.Dropdown(label='Hypernetwork', elem_id="train_hypernetwork", choices=list(shared.hypernetworks.keys())) + create_refresh_button(train_hypernetwork_name, shared.reload_hypernetworks, lambda: {"choices": sorted(shared.hypernetworks.keys())}, "refresh_train_hypernetwork_name") with FormRow(): embedding_learn_rate = gr.Textbox(label='Embedding Learning rate', placeholder="Embedding Learning rate", value="0.005", elem_id="train_embedding_learn_rate") @@ -1808,7 +1808,7 @@ def create_ui(): if type(x) == gr.Dropdown: def check_dropdown(val): if getattr(x, 'multiselect', False): - return all([value in x.choices for value in val]) + return all(value in x.choices for value in val) else: return val in x.choices diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 800e467a..ab585917 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -26,7 +26,7 @@ def register_page(page): def fetch_file(filename: str = ""): from starlette.responses import FileResponse - if not any([Path(x).absolute() in Path(filename).absolute().parents for x in allowed_dirs]): + if not any(Path(x).absolute() in Path(filename).absolute().parents for x in allowed_dirs): raise ValueError(f"File cannot be fetched: {filename}. Must be in one of directories registered by extra pages.") ext = os.path.splitext(filename)[1].lower() @@ -326,7 +326,7 @@ def setup_ui(ui, gallery): is_allowed = False for extra_page in ui.stored_extra_pages: - if any([path_is_parent(x, filename) for x in extra_page.allowed_directories_for_previews()]): + if any(path_is_parent(x, filename) for x in extra_page.allowed_directories_for_previews()): is_allowed = True break diff --git a/modules/ui_tempdir.py b/modules/ui_tempdir.py index 46fa9cb0..cac73c51 100644 --- a/modules/ui_tempdir.py +++ b/modules/ui_tempdir.py @@ -23,7 +23,7 @@ def register_tmp_file(gradio, filename): def check_tmp_file(gradio, filename): if hasattr(gradio, 'temp_file_sets'): - return any([filename in fileset for fileset in gradio.temp_file_sets]) + return any(filename in fileset for fileset in gradio.temp_file_sets) if hasattr(gradio, 'temp_dirs'): return any(Path(temp_dir).resolve() in Path(filename).resolve().parents for temp_dir in gradio.temp_dirs) -- cgit v1.2.3 From 49a55b410b66b7dd9be9335d8a2e3a71e4f8b15c Mon Sep 17 00:00:00 2001 From: Aarni Koskela Date: Thu, 11 May 2023 18:28:15 +0300 Subject: Autofix Ruff W (not W605) (mostly whitespace) --- extensions-builtin/LDSR/ldsr_model_arch.py | 4 +- extensions-builtin/LDSR/sd_hijack_ddpm_v1.py | 6 +-- extensions-builtin/ScuNET/scunet_model_arch.py | 2 +- extensions-builtin/SwinIR/scripts/swinir_model.py | 2 +- extensions-builtin/SwinIR/swinir_model_arch.py | 2 +- extensions-builtin/SwinIR/swinir_model_arch_v2.py | 52 +++++++++++------------ launch.py | 2 +- modules/api/api.py | 4 +- modules/api/models.py | 2 +- modules/cmd_args.py | 2 +- modules/codeformer/codeformer_arch.py | 14 +++--- modules/codeformer/vqgan_arch.py | 38 ++++++++--------- modules/esrgan_model_arch.py | 4 +- modules/extras.py | 2 +- modules/hypernetworks/hypernetwork.py | 12 +++--- modules/images.py | 2 +- modules/mac_specific.py | 4 +- modules/masking.py | 2 +- modules/ngrok.py | 4 +- modules/processing.py | 2 +- modules/script_callbacks.py | 14 +++--- modules/sd_hijack.py | 12 +++--- modules/sd_hijack_optimizations.py | 32 +++++++------- modules/sd_models.py | 4 +- modules/sd_samplers_kdiffusion.py | 18 ++++---- modules/sub_quadratic_attention.py | 2 +- modules/textual_inversion/dataset.py | 4 +- modules/textual_inversion/preprocess.py | 2 +- modules/textual_inversion/textual_inversion.py | 16 +++---- modules/ui.py | 18 ++++---- modules/ui_extensions.py | 6 +-- modules/xlmr.py | 6 +-- pyproject.toml | 5 ++- scripts/img2imgalt.py | 14 +++--- scripts/loopback.py | 8 ++-- scripts/poor_mans_outpainting.py | 2 +- scripts/prompt_matrix.py | 2 +- scripts/prompts_from_file.py | 4 +- scripts/sd_upscale.py | 2 +- 39 files changed, 167 insertions(+), 166 deletions(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/extensions-builtin/LDSR/ldsr_model_arch.py b/extensions-builtin/LDSR/ldsr_model_arch.py index 2173de79..7f450086 100644 --- a/extensions-builtin/LDSR/ldsr_model_arch.py +++ b/extensions-builtin/LDSR/ldsr_model_arch.py @@ -130,11 +130,11 @@ class LDSR: im_og = im_og.resize((width_downsampled_pre, height_downsampled_pre), Image.LANCZOS) else: print(f"Down sample rate is 1 from {target_scale} / 4 (Not downsampling)") - + # pad width and height to multiples of 64, pads with the edge values of image to avoid artifacts pad_w, pad_h = np.max(((2, 2), np.ceil(np.array(im_og.size) / 64).astype(int)), axis=0) * 64 - im_og.size im_padded = Image.fromarray(np.pad(np.array(im_og), ((0, pad_h), (0, pad_w), (0, 0)), mode='edge')) - + logs = self.run(model["model"], im_padded, diffusion_steps, eta) sample = logs["sample"] diff --git a/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py index 57c02d12..631a08ef 100644 --- a/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py +++ b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py @@ -460,7 +460,7 @@ class LatentDiffusionV1(DDPMV1): self.instantiate_cond_stage(cond_stage_config) self.cond_stage_forward = cond_stage_forward self.clip_denoised = False - self.bbox_tokenizer = None + self.bbox_tokenizer = None self.restarted_from_ckpt = False if ckpt_path is not None: @@ -792,7 +792,7 @@ class LatentDiffusionV1(DDPMV1): z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1])) # (bn, nc, ks[0], ks[1], L ) # 2. apply model loop over last dim - if isinstance(self.first_stage_model, VQModelInterface): + if isinstance(self.first_stage_model, VQModelInterface): output_list = [self.first_stage_model.decode(z[:, :, :, :, i], force_not_quantize=predict_cids or force_not_quantize) for i in range(z.shape[-1])] @@ -890,7 +890,7 @@ class LatentDiffusionV1(DDPMV1): if hasattr(self, "split_input_params"): assert len(cond) == 1 # todo can only deal with one conditioning atm - assert not return_ids + assert not return_ids ks = self.split_input_params["ks"] # eg. (128, 128) stride = self.split_input_params["stride"] # eg. (64, 64) diff --git a/extensions-builtin/ScuNET/scunet_model_arch.py b/extensions-builtin/ScuNET/scunet_model_arch.py index 8028918a..b51a8806 100644 --- a/extensions-builtin/ScuNET/scunet_model_arch.py +++ b/extensions-builtin/ScuNET/scunet_model_arch.py @@ -265,4 +265,4 @@ class SCUNet(nn.Module): nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) - nn.init.constant_(m.weight, 1.0) \ No newline at end of file + nn.init.constant_(m.weight, 1.0) diff --git a/extensions-builtin/SwinIR/scripts/swinir_model.py b/extensions-builtin/SwinIR/scripts/swinir_model.py index 55dd94ab..0ba50487 100644 --- a/extensions-builtin/SwinIR/scripts/swinir_model.py +++ b/extensions-builtin/SwinIR/scripts/swinir_model.py @@ -150,7 +150,7 @@ def inference(img, model, tile, tile_overlap, window_size, scale): for w_idx in w_idx_list: if state.interrupted or state.skipped: break - + in_patch = img[..., h_idx: h_idx + tile, w_idx: w_idx + tile] out_patch = model(in_patch) out_patch_mask = torch.ones_like(out_patch) diff --git a/extensions-builtin/SwinIR/swinir_model_arch.py b/extensions-builtin/SwinIR/swinir_model_arch.py index 73e37cfa..93b93274 100644 --- a/extensions-builtin/SwinIR/swinir_model_arch.py +++ b/extensions-builtin/SwinIR/swinir_model_arch.py @@ -805,7 +805,7 @@ class SwinIR(nn.Module): def forward(self, x): H, W = x.shape[2:] x = self.check_image_size(x) - + self.mean = self.mean.type_as(x) x = (x - self.mean) * self.img_range diff --git a/extensions-builtin/SwinIR/swinir_model_arch_v2.py b/extensions-builtin/SwinIR/swinir_model_arch_v2.py index 3ca9be78..dad22cca 100644 --- a/extensions-builtin/SwinIR/swinir_model_arch_v2.py +++ b/extensions-builtin/SwinIR/swinir_model_arch_v2.py @@ -241,7 +241,7 @@ class SwinTransformerBlock(nn.Module): attn_mask = None self.register_buffer("attn_mask", attn_mask) - + def calculate_mask(self, x_size): # calculate attention mask for SW-MSA H, W = x_size @@ -263,7 +263,7 @@ class SwinTransformerBlock(nn.Module): attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) - return attn_mask + return attn_mask def forward(self, x, x_size): H, W = x_size @@ -288,7 +288,7 @@ class SwinTransformerBlock(nn.Module): attn_windows = self.attn(x_windows, mask=self.attn_mask) # nW*B, window_size*window_size, C else: attn_windows = self.attn(x_windows, mask=self.calculate_mask(x_size).to(x.device)) - + # merge windows attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) shifted_x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C @@ -369,7 +369,7 @@ class PatchMerging(nn.Module): H, W = self.input_resolution flops = (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim flops += H * W * self.dim // 2 - return flops + return flops class BasicLayer(nn.Module): """ A basic Swin Transformer layer for one stage. @@ -447,7 +447,7 @@ class BasicLayer(nn.Module): nn.init.constant_(blk.norm1.weight, 0) nn.init.constant_(blk.norm2.bias, 0) nn.init.constant_(blk.norm2.weight, 0) - + class PatchEmbed(nn.Module): r""" Image to Patch Embedding Args: @@ -492,7 +492,7 @@ class PatchEmbed(nn.Module): flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1]) if self.norm is not None: flops += Ho * Wo * self.embed_dim - return flops + return flops class RSTB(nn.Module): """Residual Swin Transformer Block (RSTB). @@ -531,7 +531,7 @@ class RSTB(nn.Module): num_heads=num_heads, window_size=window_size, mlp_ratio=mlp_ratio, - qkv_bias=qkv_bias, + qkv_bias=qkv_bias, drop=drop, attn_drop=attn_drop, drop_path=drop_path, norm_layer=norm_layer, @@ -622,7 +622,7 @@ class Upsample(nn.Sequential): else: raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.') super(Upsample, self).__init__(*m) - + class Upsample_hf(nn.Sequential): """Upsample module. @@ -642,7 +642,7 @@ class Upsample_hf(nn.Sequential): m.append(nn.PixelShuffle(3)) else: raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.') - super(Upsample_hf, self).__init__(*m) + super(Upsample_hf, self).__init__(*m) class UpsampleOneStep(nn.Sequential): @@ -667,8 +667,8 @@ class UpsampleOneStep(nn.Sequential): H, W = self.input_resolution flops = H * W * self.num_feat * 3 * 9 return flops - - + + class Swin2SR(nn.Module): r""" Swin2SR @@ -699,7 +699,7 @@ class Swin2SR(nn.Module): def __init__(self, img_size=64, patch_size=1, in_chans=3, embed_dim=96, depths=(6, 6, 6, 6), num_heads=(6, 6, 6, 6), - window_size=7, mlp_ratio=4., qkv_bias=True, + window_size=7, mlp_ratio=4., qkv_bias=True, drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1, norm_layer=nn.LayerNorm, ape=False, patch_norm=True, use_checkpoint=False, upscale=2, img_range=1., upsampler='', resi_connection='1conv', @@ -764,7 +764,7 @@ class Swin2SR(nn.Module): num_heads=num_heads[i_layer], window_size=window_size, mlp_ratio=self.mlp_ratio, - qkv_bias=qkv_bias, + qkv_bias=qkv_bias, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], # no impact on SR results norm_layer=norm_layer, @@ -776,7 +776,7 @@ class Swin2SR(nn.Module): ) self.layers.append(layer) - + if self.upsampler == 'pixelshuffle_hf': self.layers_hf = nn.ModuleList() for i_layer in range(self.num_layers): @@ -787,7 +787,7 @@ class Swin2SR(nn.Module): num_heads=num_heads[i_layer], window_size=window_size, mlp_ratio=self.mlp_ratio, - qkv_bias=qkv_bias, + qkv_bias=qkv_bias, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], # no impact on SR results norm_layer=norm_layer, @@ -799,7 +799,7 @@ class Swin2SR(nn.Module): ) self.layers_hf.append(layer) - + self.norm = norm_layer(self.num_features) # build the last conv layer in deep feature extraction @@ -829,10 +829,10 @@ class Swin2SR(nn.Module): self.conv_aux = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) self.conv_after_aux = nn.Sequential( nn.Conv2d(3, num_feat, 3, 1, 1), - nn.LeakyReLU(inplace=True)) + nn.LeakyReLU(inplace=True)) self.upsample = Upsample(upscale, num_feat) self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) - + elif self.upsampler == 'pixelshuffle_hf': self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True)) @@ -846,7 +846,7 @@ class Swin2SR(nn.Module): nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True)) self.conv_last_hf = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) - + elif self.upsampler == 'pixelshuffledirect': # for lightweight SR (to save parameters) self.upsample = UpsampleOneStep(upscale, embed_dim, num_out_ch, @@ -905,7 +905,7 @@ class Swin2SR(nn.Module): x = self.patch_unembed(x, x_size) return x - + def forward_features_hf(self, x): x_size = (x.shape[2], x.shape[3]) x = self.patch_embed(x) @@ -919,7 +919,7 @@ class Swin2SR(nn.Module): x = self.norm(x) # B L C x = self.patch_unembed(x, x_size) - return x + return x def forward(self, x): H, W = x.shape[2:] @@ -951,7 +951,7 @@ class Swin2SR(nn.Module): x = self.conv_after_body(self.forward_features(x)) + x x_before = self.conv_before_upsample(x) x_out = self.conv_last(self.upsample(x_before)) - + x_hf = self.conv_first_hf(x_before) x_hf = self.conv_after_body_hf(self.forward_features_hf(x_hf)) + x_hf x_hf = self.conv_before_upsample_hf(x_hf) @@ -977,15 +977,15 @@ class Swin2SR(nn.Module): x_first = self.conv_first(x) res = self.conv_after_body(self.forward_features(x_first)) + x_first x = x + self.conv_last(res) - + x = x / self.img_range + self.mean if self.upsampler == "pixelshuffle_aux": return x[:, :, :H*self.upscale, :W*self.upscale], aux - + elif self.upsampler == "pixelshuffle_hf": x_out = x_out / self.img_range + self.mean return x_out[:, :, :H*self.upscale, :W*self.upscale], x[:, :, :H*self.upscale, :W*self.upscale], x_hf[:, :, :H*self.upscale, :W*self.upscale] - + else: return x[:, :, :H*self.upscale, :W*self.upscale] @@ -1014,4 +1014,4 @@ if __name__ == '__main__': x = torch.randn((1, 3, height, width)) x = model(x) - print(x.shape) \ No newline at end of file + print(x.shape) diff --git a/launch.py b/launch.py index 670af87c..62b33f14 100644 --- a/launch.py +++ b/launch.py @@ -327,7 +327,7 @@ def prepare_environment(): if args.update_all_extensions: git_pull_recursive(extensions_dir) - + if "--exit" in sys.argv: print("Exiting because of --exit argument") exit(0) diff --git a/modules/api/api.py b/modules/api/api.py index 594fa655..165985c3 100644 --- a/modules/api/api.py +++ b/modules/api/api.py @@ -227,7 +227,7 @@ class Api: script_idx = script_name_to_index(script_name, script_runner.selectable_scripts) script = script_runner.selectable_scripts[script_idx] return script, script_idx - + def get_scripts_list(self): t2ilist = [str(title.lower()) for title in scripts.scripts_txt2img.titles] i2ilist = [str(title.lower()) for title in scripts.scripts_img2img.titles] @@ -237,7 +237,7 @@ class Api: def get_script(self, script_name, script_runner): if script_name is None or script_name == "": return None, None - + script_idx = script_name_to_index(script_name, script_runner.scripts) return script_runner.scripts[script_idx] diff --git a/modules/api/models.py b/modules/api/models.py index 4d291076..006ccdb7 100644 --- a/modules/api/models.py +++ b/modules/api/models.py @@ -289,4 +289,4 @@ class MemoryResponse(BaseModel): class ScriptsList(BaseModel): txt2img: list = Field(default=None,title="Txt2img", description="Titles of scripts (txt2img)") - img2img: list = Field(default=None,title="Img2img", description="Titles of scripts (img2img)") \ No newline at end of file + img2img: list = Field(default=None,title="Img2img", description="Titles of scripts (img2img)") diff --git a/modules/cmd_args.py b/modules/cmd_args.py index e01ca655..f4a4ab36 100644 --- a/modules/cmd_args.py +++ b/modules/cmd_args.py @@ -102,4 +102,4 @@ parser.add_argument("--no-gradio-queue", action='store_true', help="Disables gra parser.add_argument("--skip-version-check", action='store_true', help="Do not check versions of torch and xformers") parser.add_argument("--no-hashing", action='store_true', help="disable sha256 hashing of checkpoints to help loading performance", default=False) parser.add_argument("--no-download-sd-model", action='store_true', help="don't download SD1.5 model even if no model is found in --ckpt-dir", default=False) -parser.add_argument('--subpath', type=str, help='customize the subpath for gradio, use with reverse proxy') \ No newline at end of file +parser.add_argument('--subpath', type=str, help='customize the subpath for gradio, use with reverse proxy') diff --git a/modules/codeformer/codeformer_arch.py b/modules/codeformer/codeformer_arch.py index 45c70f84..12db6814 100644 --- a/modules/codeformer/codeformer_arch.py +++ b/modules/codeformer/codeformer_arch.py @@ -119,7 +119,7 @@ class TransformerSALayer(nn.Module): tgt_mask: Optional[Tensor] = None, tgt_key_padding_mask: Optional[Tensor] = None, query_pos: Optional[Tensor] = None): - + # self attention tgt2 = self.norm1(tgt) q = k = self.with_pos_embed(tgt2, query_pos) @@ -159,7 +159,7 @@ class Fuse_sft_block(nn.Module): @ARCH_REGISTRY.register() class CodeFormer(VQAutoEncoder): - def __init__(self, dim_embd=512, n_head=8, n_layers=9, + def __init__(self, dim_embd=512, n_head=8, n_layers=9, codebook_size=1024, latent_size=256, connect_list=('32', '64', '128', '256'), fix_modules=('quantize', 'generator')): @@ -179,14 +179,14 @@ class CodeFormer(VQAutoEncoder): self.feat_emb = nn.Linear(256, self.dim_embd) # transformer - self.ft_layers = nn.Sequential(*[TransformerSALayer(embed_dim=dim_embd, nhead=n_head, dim_mlp=self.dim_mlp, dropout=0.0) + self.ft_layers = nn.Sequential(*[TransformerSALayer(embed_dim=dim_embd, nhead=n_head, dim_mlp=self.dim_mlp, dropout=0.0) for _ in range(self.n_layers)]) # logits_predict head self.idx_pred_layer = nn.Sequential( nn.LayerNorm(dim_embd), nn.Linear(dim_embd, codebook_size, bias=False)) - + self.channels = { '16': 512, '32': 256, @@ -221,7 +221,7 @@ class CodeFormer(VQAutoEncoder): enc_feat_dict = {} out_list = [self.fuse_encoder_block[f_size] for f_size in self.connect_list] for i, block in enumerate(self.encoder.blocks): - x = block(x) + x = block(x) if i in out_list: enc_feat_dict[str(x.shape[-1])] = x.clone() @@ -266,11 +266,11 @@ class CodeFormer(VQAutoEncoder): fuse_list = [self.fuse_generator_block[f_size] for f_size in self.connect_list] for i, block in enumerate(self.generator.blocks): - x = block(x) + x = block(x) if i in fuse_list: # fuse after i-th block f_size = str(x.shape[-1]) if w>0: x = self.fuse_convs_dict[f_size](enc_feat_dict[f_size].detach(), x, w) out = x # logits doesn't need softmax before cross_entropy loss - return out, logits, lq_feat \ No newline at end of file + return out, logits, lq_feat diff --git a/modules/codeformer/vqgan_arch.py b/modules/codeformer/vqgan_arch.py index b24a0394..09ee6660 100644 --- a/modules/codeformer/vqgan_arch.py +++ b/modules/codeformer/vqgan_arch.py @@ -13,7 +13,7 @@ from basicsr.utils.registry import ARCH_REGISTRY def normalize(in_channels): return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) - + @torch.jit.script def swish(x): @@ -210,15 +210,15 @@ class AttnBlock(nn.Module): # compute attention b, c, h, w = q.shape q = q.reshape(b, c, h*w) - q = q.permute(0, 2, 1) + q = q.permute(0, 2, 1) k = k.reshape(b, c, h*w) - w_ = torch.bmm(q, k) + w_ = torch.bmm(q, k) w_ = w_ * (int(c)**(-0.5)) w_ = F.softmax(w_, dim=2) # attend to values v = v.reshape(b, c, h*w) - w_ = w_.permute(0, 2, 1) + w_ = w_.permute(0, 2, 1) h_ = torch.bmm(v, w_) h_ = h_.reshape(b, c, h, w) @@ -270,18 +270,18 @@ class Encoder(nn.Module): def forward(self, x): for block in self.blocks: x = block(x) - + return x class Generator(nn.Module): def __init__(self, nf, emb_dim, ch_mult, res_blocks, img_size, attn_resolutions): super().__init__() - self.nf = nf - self.ch_mult = ch_mult + self.nf = nf + self.ch_mult = ch_mult self.num_resolutions = len(self.ch_mult) self.num_res_blocks = res_blocks - self.resolution = img_size + self.resolution = img_size self.attn_resolutions = attn_resolutions self.in_channels = emb_dim self.out_channels = 3 @@ -315,24 +315,24 @@ class Generator(nn.Module): blocks.append(nn.Conv2d(block_in_ch, self.out_channels, kernel_size=3, stride=1, padding=1)) self.blocks = nn.ModuleList(blocks) - + def forward(self, x): for block in self.blocks: x = block(x) - + return x - + @ARCH_REGISTRY.register() class VQAutoEncoder(nn.Module): def __init__(self, img_size, nf, ch_mult, quantizer="nearest", res_blocks=2, attn_resolutions=None, codebook_size=1024, emb_dim=256, beta=0.25, gumbel_straight_through=False, gumbel_kl_weight=1e-8, model_path=None): super().__init__() logger = get_root_logger() - self.in_channels = 3 - self.nf = nf - self.n_blocks = res_blocks + self.in_channels = 3 + self.nf = nf + self.n_blocks = res_blocks self.codebook_size = codebook_size self.embed_dim = emb_dim self.ch_mult = ch_mult @@ -363,11 +363,11 @@ class VQAutoEncoder(nn.Module): self.kl_weight ) self.generator = Generator( - self.nf, + self.nf, self.embed_dim, - self.ch_mult, - self.n_blocks, - self.resolution, + self.ch_mult, + self.n_blocks, + self.resolution, self.attn_resolutions ) @@ -432,4 +432,4 @@ class VQGANDiscriminator(nn.Module): raise ValueError('Wrong params!') def forward(self, x): - return self.main(x) \ No newline at end of file + return self.main(x) diff --git a/modules/esrgan_model_arch.py b/modules/esrgan_model_arch.py index 4de9dd8d..2b9888ba 100644 --- a/modules/esrgan_model_arch.py +++ b/modules/esrgan_model_arch.py @@ -105,7 +105,7 @@ class ResidualDenseBlock_5C(nn.Module): Modified options that can be used: - "Partial Convolution based Padding" arXiv:1811.11718 - "Spectral normalization" arXiv:1802.05957 - - "ICASSP 2020 - ESRGAN+ : Further Improving ESRGAN" N. C. + - "ICASSP 2020 - ESRGAN+ : Further Improving ESRGAN" N. C. {Rakotonirina} and A. {Rasoanaivo} """ @@ -170,7 +170,7 @@ class GaussianNoise(nn.Module): scale = self.sigma * x.detach() if self.is_relative_detach else self.sigma * x sampled_noise = self.noise.repeat(*x.size()).normal_() * scale x = x + sampled_noise - return x + return x def conv1x1(in_planes, out_planes, stride=1): return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False) diff --git a/modules/extras.py b/modules/extras.py index eb4f0b42..bdf9b3b7 100644 --- a/modules/extras.py +++ b/modules/extras.py @@ -199,7 +199,7 @@ def run_modelmerger(id_task, primary_model_name, secondary_model_name, tertiary_ result_is_inpainting_model = True else: theta_0[key] = theta_func2(a, b, multiplier) - + theta_0[key] = to_half(theta_0[key], save_as_half) shared.state.sampling_step += 1 diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 38ef074f..570b5603 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -540,7 +540,7 @@ def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradi return hypernetwork, filename scheduler = LearnRateScheduler(learn_rate, steps, initial_step) - + clip_grad = torch.nn.utils.clip_grad_value_ if clip_grad_mode == "value" else torch.nn.utils.clip_grad_norm_ if clip_grad_mode == "norm" else None if clip_grad: clip_grad_sched = LearnRateScheduler(clip_grad_value, steps, initial_step, verbose=False) @@ -593,7 +593,7 @@ def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradi print(e) scaler = torch.cuda.amp.GradScaler() - + batch_size = ds.batch_size gradient_step = ds.gradient_step # n steps = batch_size * gradient_step * n image processed @@ -636,7 +636,7 @@ def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradi if clip_grad: clip_grad_sched.step(hypernetwork.step) - + with devices.autocast(): x = batch.latent_sample.to(devices.device, non_blocking=pin_memory) if use_weight: @@ -657,14 +657,14 @@ def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradi _loss_step += loss.item() scaler.scale(loss).backward() - + # go back until we reach gradient accumulation steps if (j + 1) % gradient_step != 0: continue loss_logging.append(_loss_step) if clip_grad: clip_grad(weights, clip_grad_sched.learn_rate) - + scaler.step(optimizer) scaler.update() hypernetwork.step += 1 @@ -674,7 +674,7 @@ def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradi _loss_step = 0 steps_done = hypernetwork.step + 1 - + epoch_num = hypernetwork.step // steps_per_epoch epoch_step = hypernetwork.step % steps_per_epoch diff --git a/modules/images.py b/modules/images.py index 3b8b62d9..b2de3662 100644 --- a/modules/images.py +++ b/modules/images.py @@ -367,7 +367,7 @@ class FilenameGenerator: self.seed = seed self.prompt = prompt self.image = image - + def hasprompt(self, *args): lower = self.prompt.lower() if self.p is None or self.prompt is None: diff --git a/modules/mac_specific.py b/modules/mac_specific.py index 5c2f92a1..d74c6b95 100644 --- a/modules/mac_specific.py +++ b/modules/mac_specific.py @@ -42,7 +42,7 @@ if has_mps: # MPS workaround for https://github.com/pytorch/pytorch/issues/79383 CondFunc('torch.Tensor.to', lambda orig_func, self, *args, **kwargs: orig_func(self.contiguous(), *args, **kwargs), lambda _, self, *args, **kwargs: self.device.type != 'mps' and (args and isinstance(args[0], torch.device) and args[0].type == 'mps' or isinstance(kwargs.get('device'), torch.device) and kwargs['device'].type == 'mps')) - # MPS workaround for https://github.com/pytorch/pytorch/issues/80800 + # MPS workaround for https://github.com/pytorch/pytorch/issues/80800 CondFunc('torch.nn.functional.layer_norm', lambda orig_func, *args, **kwargs: orig_func(*([args[0].contiguous()] + list(args[1:])), **kwargs), lambda _, *args, **kwargs: args and isinstance(args[0], torch.Tensor) and args[0].device.type == 'mps') # MPS workaround for https://github.com/pytorch/pytorch/issues/90532 @@ -60,4 +60,4 @@ if has_mps: # MPS workaround for https://github.com/pytorch/pytorch/issues/92311 if platform.processor() == 'i386': for funcName in ['torch.argmax', 'torch.Tensor.argmax']: - CondFunc(funcName, lambda _, input, *args, **kwargs: torch.max(input.float() if input.dtype == torch.int64 else input, *args, **kwargs)[1], lambda _, input, *args, **kwargs: input.device.type == 'mps') \ No newline at end of file + CondFunc(funcName, lambda _, input, *args, **kwargs: torch.max(input.float() if input.dtype == torch.int64 else input, *args, **kwargs)[1], lambda _, input, *args, **kwargs: input.device.type == 'mps') diff --git a/modules/masking.py b/modules/masking.py index a5c4d2da..be9f84c7 100644 --- a/modules/masking.py +++ b/modules/masking.py @@ -4,7 +4,7 @@ from PIL import Image, ImageFilter, ImageOps def get_crop_region(mask, pad=0): """finds a rectangular region that contains all masked ares in an image. Returns (x1, y1, x2, y2) coordinates of the rectangle. For example, if a user has painted the top-right part of a 512x512 image", the result may be (256, 0, 512, 256)""" - + h, w = mask.shape crop_left = 0 diff --git a/modules/ngrok.py b/modules/ngrok.py index 7a7b4b26..67a74e85 100644 --- a/modules/ngrok.py +++ b/modules/ngrok.py @@ -13,7 +13,7 @@ def connect(token, port, region): config = conf.PyngrokConfig( auth_token=token, region=region ) - + # Guard for existing tunnels existing = ngrok.get_tunnels(pyngrok_config=config) if existing: @@ -24,7 +24,7 @@ def connect(token, port, region): print(f'ngrok has already been connected to localhost:{port}! URL: {public_url}\n' 'You can use this link after the launch is complete.') return - + try: if account is None: public_url = ngrok.connect(port, pyngrok_config=config, bind_tls=True).public_url diff --git a/modules/processing.py b/modules/processing.py index c3932d6b..f902b9df 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -164,7 +164,7 @@ class StableDiffusionProcessing: self.all_subseeds = None self.iteration = 0 self.is_hr_pass = False - + @property def sd_model(self): diff --git a/modules/script_callbacks.py b/modules/script_callbacks.py index 17109732..7d9dd736 100644 --- a/modules/script_callbacks.py +++ b/modules/script_callbacks.py @@ -32,22 +32,22 @@ class CFGDenoiserParams: def __init__(self, x, image_cond, sigma, sampling_step, total_sampling_steps, text_cond, text_uncond): self.x = x """Latent image representation in the process of being denoised""" - + self.image_cond = image_cond """Conditioning image""" - + self.sigma = sigma """Current sigma noise step value""" - + self.sampling_step = sampling_step """Current Sampling step number""" - + self.total_sampling_steps = total_sampling_steps """Total number of sampling steps planned""" - + self.text_cond = text_cond """ Encoder hidden states of text conditioning from prompt""" - + self.text_uncond = text_uncond """ Encoder hidden states of text conditioning from negative prompt""" @@ -240,7 +240,7 @@ def add_callback(callbacks, fun): callbacks.append(ScriptCallback(filename, fun)) - + def remove_current_script_callbacks(): stack = [x for x in inspect.stack() if x.filename != __file__] filename = stack[0].filename if len(stack) > 0 else 'unknown file' diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index e374aeb8..7e50f1ab 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -34,7 +34,7 @@ def apply_optimizations(): ldm.modules.diffusionmodules.model.nonlinearity = silu ldm.modules.diffusionmodules.openaimodel.th = sd_hijack_unet.th - + optimization_method = None can_use_sdp = hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) # not everyone has torch 2.x to use sdp @@ -92,12 +92,12 @@ def fix_checkpoint(): def weighted_loss(sd_model, pred, target, mean=True): #Calculate the weight normally, but ignore the mean loss = sd_model._old_get_loss(pred, target, mean=False) - + #Check if we have weights available weight = getattr(sd_model, '_custom_loss_weight', None) if weight is not None: loss *= weight - + #Return the loss, as mean if specified return loss.mean() if mean else loss @@ -105,7 +105,7 @@ def weighted_forward(sd_model, x, c, w, *args, **kwargs): try: #Temporarily append weights to a place accessible during loss calc sd_model._custom_loss_weight = w - + #Replace 'get_loss' with a weight-aware one. Otherwise we need to reimplement 'forward' completely #Keep 'get_loss', but don't overwrite the previous old_get_loss if it's already set if not hasattr(sd_model, '_old_get_loss'): @@ -120,7 +120,7 @@ def weighted_forward(sd_model, x, c, w, *args, **kwargs): del sd_model._custom_loss_weight except AttributeError: pass - + #If we have an old loss function, reset the loss function to the original one if hasattr(sd_model, '_old_get_loss'): sd_model.get_loss = sd_model._old_get_loss @@ -184,7 +184,7 @@ class StableDiffusionModelHijack: def undo_hijack(self, m): if type(m.cond_stage_model) == xlmr.BertSeriesModelWithTransformation: - m.cond_stage_model = m.cond_stage_model.wrapped + m.cond_stage_model = m.cond_stage_model.wrapped elif type(m.cond_stage_model) == sd_hijack_clip.FrozenCLIPEmbedderWithCustomWords: m.cond_stage_model = m.cond_stage_model.wrapped diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index a174bbe1..f00fe55c 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -62,10 +62,10 @@ def split_cross_attention_forward_v1(self, x, context=None, mask=None): end = i + 2 s1 = einsum('b i d, b j d -> b i j', q[i:end], k[i:end]) s1 *= self.scale - + s2 = s1.softmax(dim=-1) del s1 - + r1[i:end] = einsum('b i j, b j d -> b i d', s2, v[i:end]) del s2 del q, k, v @@ -95,43 +95,43 @@ def split_cross_attention_forward(self, x, context=None, mask=None): with devices.without_autocast(disable=not shared.opts.upcast_attn): k_in = k_in * self.scale - + del context, x - + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q_in, k_in, v_in)) del q_in, k_in, v_in - + r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype) - + mem_free_total = get_available_vram() - + gb = 1024 ** 3 tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * q.element_size() modifier = 3 if q.element_size() == 2 else 2.5 mem_required = tensor_size * modifier steps = 1 - + if mem_required > mem_free_total: steps = 2 ** (math.ceil(math.log(mem_required / mem_free_total, 2))) # print(f"Expected tensor size:{tensor_size/gb:0.1f}GB, cuda free:{mem_free_cuda/gb:0.1f}GB " # f"torch free:{mem_free_torch/gb:0.1f} total:{mem_free_total/gb:0.1f} steps:{steps}") - + if steps > 64: max_res = math.floor(math.sqrt(math.sqrt(mem_free_total / 2.5)) / 8) * 64 raise RuntimeError(f'Not enough memory, use lower resolution (max approx. {max_res}x{max_res}). ' f'Need: {mem_required / 64 / gb:0.1f}GB free, Have:{mem_free_total / gb:0.1f}GB free') - + slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1] for i in range(0, q.shape[1], slice_size): end = i + slice_size s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k) - + s2 = s1.softmax(dim=-1, dtype=q.dtype) del s1 - + r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v) del s2 - + del q, k, v r1 = r1.to(dtype) @@ -228,7 +228,7 @@ def split_cross_attention_forward_invokeAI(self, x, context=None, mask=None): with devices.without_autocast(disable=not shared.opts.upcast_attn): k = k * self.scale - + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q, k, v)) r = einsum_op(q, k, v) r = r.to(dtype) @@ -369,7 +369,7 @@ def scaled_dot_product_attention_forward(self, x, context=None, mask=None): q = q_in.view(batch_size, -1, h, head_dim).transpose(1, 2) k = k_in.view(batch_size, -1, h, head_dim).transpose(1, 2) v = v_in.view(batch_size, -1, h, head_dim).transpose(1, 2) - + del q_in, k_in, v_in dtype = q.dtype @@ -451,7 +451,7 @@ def cross_attention_attnblock_forward(self, x): h3 += x return h3 - + def xformers_attnblock_forward(self, x): try: h_ = x diff --git a/modules/sd_models.py b/modules/sd_models.py index d1e946a5..3316d021 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -165,7 +165,7 @@ def model_hash(filename): def select_checkpoint(): model_checkpoint = shared.opts.sd_model_checkpoint - + checkpoint_info = checkpoint_alisases.get(model_checkpoint, None) if checkpoint_info is not None: return checkpoint_info @@ -372,7 +372,7 @@ def enable_midas_autodownload(): if not os.path.exists(path): if not os.path.exists(midas_path): mkdir(midas_path) - + print(f"Downloading midas model weights for {model_type} to {path}") request.urlretrieve(midas_urls[model_type], path) print(f"{model_type} downloaded") diff --git a/modules/sd_samplers_kdiffusion.py b/modules/sd_samplers_kdiffusion.py index 2f733cf5..e9e41818 100644 --- a/modules/sd_samplers_kdiffusion.py +++ b/modules/sd_samplers_kdiffusion.py @@ -93,10 +93,10 @@ class CFGDenoiser(torch.nn.Module): if shared.sd_model.model.conditioning_key == "crossattn-adm": image_uncond = torch.zeros_like(image_cond) - make_condition_dict = lambda c_crossattn, c_adm: {"c_crossattn": c_crossattn, "c_adm": c_adm} + make_condition_dict = lambda c_crossattn, c_adm: {"c_crossattn": c_crossattn, "c_adm": c_adm} else: image_uncond = image_cond - make_condition_dict = lambda c_crossattn, c_concat: {"c_crossattn": c_crossattn, "c_concat": [c_concat]} + make_condition_dict = lambda c_crossattn, c_concat: {"c_crossattn": c_crossattn, "c_concat": [c_concat]} if not is_edit_model: x_in = torch.cat([torch.stack([x[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [x]) @@ -316,7 +316,7 @@ class KDiffusionSampler: sigma_sched = sigmas[steps - t_enc - 1:] xi = x + noise * sigma_sched[0] - + extra_params_kwargs = self.initialize(p) parameters = inspect.signature(self.func).parameters @@ -339,9 +339,9 @@ class KDiffusionSampler: self.model_wrap_cfg.init_latent = x self.last_latent = x extra_args={ - 'cond': conditioning, - 'image_cond': image_conditioning, - 'uncond': unconditional_conditioning, + 'cond': conditioning, + 'image_cond': image_conditioning, + 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale, 's_min_uncond': self.s_min_uncond } @@ -374,9 +374,9 @@ class KDiffusionSampler: self.last_latent = x samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args={ - 'cond': conditioning, - 'image_cond': image_conditioning, - 'uncond': unconditional_conditioning, + 'cond': conditioning, + 'image_cond': image_conditioning, + 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale, 's_min_uncond': self.s_min_uncond }, disable=False, callback=self.callback_state, **extra_params_kwargs)) diff --git a/modules/sub_quadratic_attention.py b/modules/sub_quadratic_attention.py index cc38debd..497568eb 100644 --- a/modules/sub_quadratic_attention.py +++ b/modules/sub_quadratic_attention.py @@ -179,7 +179,7 @@ def efficient_dot_product_attention( chunk_idx, min(query_chunk_size, q_tokens) ) - + summarize_chunk: SummarizeChunk = partial(_summarize_chunk, scale=scale) summarize_chunk: SummarizeChunk = partial(checkpoint, summarize_chunk) if use_checkpoint else summarize_chunk compute_query_chunk_attn: ComputeQueryChunkAttn = partial( diff --git a/modules/textual_inversion/dataset.py b/modules/textual_inversion/dataset.py index 41610e03..b9621fc9 100644 --- a/modules/textual_inversion/dataset.py +++ b/modules/textual_inversion/dataset.py @@ -118,7 +118,7 @@ class PersonalizedBase(Dataset): weight = torch.ones(latent_sample.shape) else: weight = None - + if latent_sampling_method == "random": entry = DatasetEntry(filename=path, filename_text=filename_text, latent_dist=latent_dist, weight=weight) else: @@ -243,4 +243,4 @@ class BatchLoaderRandom(BatchLoader): return self def collate_wrapper_random(batch): - return BatchLoaderRandom(batch) \ No newline at end of file + return BatchLoaderRandom(batch) diff --git a/modules/textual_inversion/preprocess.py b/modules/textual_inversion/preprocess.py index d0cad09e..a009d8e8 100644 --- a/modules/textual_inversion/preprocess.py +++ b/modules/textual_inversion/preprocess.py @@ -125,7 +125,7 @@ def multicrop_pic(image: Image, mindim, maxdim, minarea, maxarea, objective, thr default=None ) return wh and center_crop(image, *wh) - + def preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_keep_original_size, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2, process_focal_crop=False, process_focal_crop_face_weight=0.9, process_focal_crop_entropy_weight=0.3, process_focal_crop_edges_weight=0.5, process_focal_crop_debug=False, process_multicrop=None, process_multicrop_mindim=None, process_multicrop_maxdim=None, process_multicrop_minarea=None, process_multicrop_maxarea=None, process_multicrop_objective=None, process_multicrop_threshold=None): width = process_width diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py index 9e1b2b9a..d489ed1e 100644 --- a/modules/textual_inversion/textual_inversion.py +++ b/modules/textual_inversion/textual_inversion.py @@ -323,16 +323,16 @@ def tensorboard_add(tensorboard_writer, loss, global_step, step, learn_rate, epo tensorboard_add_scaler(tensorboard_writer, f"Learn rate/train/epoch-{epoch_num}", learn_rate, step) def tensorboard_add_scaler(tensorboard_writer, tag, value, step): - tensorboard_writer.add_scalar(tag=tag, + tensorboard_writer.add_scalar(tag=tag, scalar_value=value, global_step=step) def tensorboard_add_image(tensorboard_writer, tag, pil_image, step): # Convert a pil image to a torch tensor img_tensor = torch.as_tensor(np.array(pil_image, copy=True)) - img_tensor = img_tensor.view(pil_image.size[1], pil_image.size[0], + img_tensor = img_tensor.view(pil_image.size[1], pil_image.size[0], len(pil_image.getbands())) img_tensor = img_tensor.permute((2, 0, 1)) - + tensorboard_writer.add_image(tag, img_tensor, global_step=step) def validate_train_inputs(model_name, learn_rate, batch_size, gradient_step, data_root, template_file, template_filename, steps, save_model_every, create_image_every, log_directory, name="embedding"): @@ -402,7 +402,7 @@ def train_embedding(id_task, embedding_name, learn_rate, batch_size, gradient_st if initial_step >= steps: shared.state.textinfo = "Model has already been trained beyond specified max steps" return embedding, filename - + scheduler = LearnRateScheduler(learn_rate, steps, initial_step) clip_grad = torch.nn.utils.clip_grad_value_ if clip_grad_mode == "value" else \ torch.nn.utils.clip_grad_norm_ if clip_grad_mode == "norm" else \ @@ -412,7 +412,7 @@ def train_embedding(id_task, embedding_name, learn_rate, batch_size, gradient_st # dataset loading may take a while, so input validations and early returns should be done before this shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..." old_parallel_processing_allowed = shared.parallel_processing_allowed - + if shared.opts.training_enable_tensorboard: tensorboard_writer = tensorboard_setup(log_directory) @@ -439,7 +439,7 @@ def train_embedding(id_task, embedding_name, learn_rate, batch_size, gradient_st optimizer_saved_dict = torch.load(f"{filename}.optim", map_location='cpu') if embedding.checksum() == optimizer_saved_dict.get('hash', None): optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None) - + if optimizer_state_dict is not None: optimizer.load_state_dict(optimizer_state_dict) print("Loaded existing optimizer from checkpoint") @@ -485,7 +485,7 @@ def train_embedding(id_task, embedding_name, learn_rate, batch_size, gradient_st if clip_grad: clip_grad_sched.step(embedding.step) - + with devices.autocast(): x = batch.latent_sample.to(devices.device, non_blocking=pin_memory) if use_weight: @@ -513,7 +513,7 @@ def train_embedding(id_task, embedding_name, learn_rate, batch_size, gradient_st # go back until we reach gradient accumulation steps if (j + 1) % gradient_step != 0: continue - + if clip_grad: clip_grad(embedding.vec, clip_grad_sched.learn_rate) diff --git a/modules/ui.py b/modules/ui.py index 1efb656a..ff82fff6 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -1171,7 +1171,7 @@ def create_ui(): process_focal_crop_entropy_weight = gr.Slider(label='Focal point entropy weight', value=0.15, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_focal_crop_entropy_weight") process_focal_crop_edges_weight = gr.Slider(label='Focal point edges weight', value=0.5, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_focal_crop_edges_weight") process_focal_crop_debug = gr.Checkbox(label='Create debug image', elem_id="train_process_focal_crop_debug") - + with gr.Column(visible=False) as process_multicrop_col: gr.Markdown('Each image is center-cropped with an automatically chosen width and height.') with gr.Row(): @@ -1183,7 +1183,7 @@ def create_ui(): with gr.Row(): process_multicrop_objective = gr.Radio(["Maximize area", "Minimize error"], value="Maximize area", label="Resizing objective", elem_id="train_process_multicrop_objective") process_multicrop_threshold = gr.Slider(minimum=0, maximum=1, step=0.01, label="Error threshold", value=0.1, elem_id="train_process_multicrop_threshold") - + with gr.Row(): with gr.Column(scale=3): gr.HTML(value="") @@ -1226,7 +1226,7 @@ def create_ui(): with FormRow(): embedding_learn_rate = gr.Textbox(label='Embedding Learning rate', placeholder="Embedding Learning rate", value="0.005", elem_id="train_embedding_learn_rate") hypernetwork_learn_rate = gr.Textbox(label='Hypernetwork Learning rate', placeholder="Hypernetwork Learning rate", value="0.00001", elem_id="train_hypernetwork_learn_rate") - + with FormRow(): clip_grad_mode = gr.Dropdown(value="disabled", label="Gradient Clipping", choices=["disabled", "value", "norm"]) clip_grad_value = gr.Textbox(placeholder="Gradient clip value", value="0.1", show_label=False) @@ -1565,7 +1565,7 @@ def create_ui(): gr.HTML(shared.html("licenses.html"), elem_id="licenses") gr.Button(value="Show all pages", elem_id="settings_show_all_pages") - + def unload_sd_weights(): modules.sd_models.unload_model_weights() @@ -1841,15 +1841,15 @@ def versions_html(): return f""" version: {tag} - •  + • python: {python_version} - •  + • torch: {getattr(torch, '__long_version__',torch.__version__)} - •  + • xformers: {xformers_version} - •  + • gradio: {gr.__version__} - •  + • checkpoint: N/A """ diff --git a/modules/ui_extensions.py b/modules/ui_extensions.py index ed70abe5..af497733 100644 --- a/modules/ui_extensions.py +++ b/modules/ui_extensions.py @@ -467,7 +467,7 @@ def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text=" {html.escape(description)}

Added: {html.escape(added)}

{install_code} - + """ for tag in [x for x in extension_tags if x not in tags]: @@ -535,9 +535,9 @@ def create_ui(): hide_tags = gr.CheckboxGroup(value=["ads", "localization", "installed"], label="Hide extensions with tags", choices=["script", "ads", "localization", "installed"]) sort_column = gr.Radio(value="newest first", label="Order", choices=["newest first", "oldest first", "a-z", "z-a", "internal order", ], type="index") - with gr.Row(): + with gr.Row(): search_extensions_text = gr.Text(label="Search").style(container=False) - + install_result = gr.HTML() available_extensions_table = gr.HTML() diff --git a/modules/xlmr.py b/modules/xlmr.py index e056c3f6..a407a3ca 100644 --- a/modules/xlmr.py +++ b/modules/xlmr.py @@ -28,7 +28,7 @@ class BertSeriesModelWithTransformation(BertPreTrainedModel): config_class = BertSeriesConfig def __init__(self, config=None, **kargs): - # modify initialization for autoloading + # modify initialization for autoloading if config is None: config = XLMRobertaConfig() config.attention_probs_dropout_prob= 0.1 @@ -74,7 +74,7 @@ class BertSeriesModelWithTransformation(BertPreTrainedModel): text["attention_mask"] = torch.tensor( text['attention_mask']).to(device) features = self(**text) - return features['projection_state'] + return features['projection_state'] def forward( self, @@ -134,4 +134,4 @@ class BertSeriesModelWithTransformation(BertPreTrainedModel): class RobertaSeriesModelWithTransformation(BertSeriesModelWithTransformation): base_model_prefix = 'roberta' - config_class= RobertaSeriesConfig \ No newline at end of file + config_class= RobertaSeriesConfig diff --git a/pyproject.toml b/pyproject.toml index c88907be..d4a1bbf4 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -6,6 +6,7 @@ extend-select = [ "B", "C", "I", + "W", ] exclude = [ @@ -20,7 +21,7 @@ ignore = [ "I001", # Import block is un-sorted or un-formatted "C901", # Function is too complex "C408", # Rewrite as a literal - + "W605", # invalid escape sequence, messes with some docstrings ] [tool.ruff.per-file-ignores] @@ -28,4 +29,4 @@ ignore = [ [tool.ruff.flake8-bugbear] # Allow default arguments like, e.g., `data: List[str] = fastapi.Query(None)`. -extend-immutable-calls = ["fastapi.Depends", "fastapi.security.HTTPBasic"] \ No newline at end of file +extend-immutable-calls = ["fastapi.Depends", "fastapi.security.HTTPBasic"] diff --git a/scripts/img2imgalt.py b/scripts/img2imgalt.py index bb00fb3f..1e833fa8 100644 --- a/scripts/img2imgalt.py +++ b/scripts/img2imgalt.py @@ -149,9 +149,9 @@ class Script(scripts.Script): sigma_adjustment = gr.Checkbox(label="Sigma adjustment for finding noise for image", value=False, elem_id=self.elem_id("sigma_adjustment")) return [ - info, + info, override_sampler, - override_prompt, original_prompt, original_negative_prompt, + override_prompt, original_prompt, original_negative_prompt, override_steps, st, override_strength, cfg, randomness, sigma_adjustment, @@ -191,17 +191,17 @@ class Script(scripts.Script): self.cache = Cached(rec_noise, cfg, st, lat, original_prompt, original_negative_prompt, sigma_adjustment) rand_noise = processing.create_random_tensors(p.init_latent.shape[1:], seeds=seeds, subseeds=subseeds, subseed_strength=p.subseed_strength, seed_resize_from_h=p.seed_resize_from_h, seed_resize_from_w=p.seed_resize_from_w, p=p) - + combined_noise = ((1 - randomness) * rec_noise + randomness * rand_noise) / ((randomness**2 + (1-randomness)**2) ** 0.5) - + sampler = sd_samplers.create_sampler(p.sampler_name, p.sd_model) sigmas = sampler.model_wrap.get_sigmas(p.steps) - + noise_dt = combined_noise - (p.init_latent / sigmas[0]) - + p.seed = p.seed + 1 - + return sampler.sample_img2img(p, p.init_latent, noise_dt, conditioning, unconditional_conditioning, image_conditioning=p.image_conditioning) p.sample = sample_extra diff --git a/scripts/loopback.py b/scripts/loopback.py index ad6609be..2d5feaf9 100644 --- a/scripts/loopback.py +++ b/scripts/loopback.py @@ -14,7 +14,7 @@ class Script(scripts.Script): def show(self, is_img2img): return is_img2img - def ui(self, is_img2img): + def ui(self, is_img2img): loops = gr.Slider(minimum=1, maximum=32, step=1, label='Loops', value=4, elem_id=self.elem_id("loops")) final_denoising_strength = gr.Slider(minimum=0, maximum=1, step=0.01, label='Final denoising strength', value=0.5, elem_id=self.elem_id("final_denoising_strength")) denoising_curve = gr.Dropdown(label="Denoising strength curve", choices=["Aggressive", "Linear", "Lazy"], value="Linear") @@ -104,7 +104,7 @@ class Script(scripts.Script): p.seed = processed.seed + 1 p.denoising_strength = calculate_denoising_strength(i + 1) - + if state.skipped: break @@ -121,7 +121,7 @@ class Script(scripts.Script): all_images.append(last_image) p.inpainting_fill = original_inpainting_fill - + if state.interrupted: break @@ -132,7 +132,7 @@ class Script(scripts.Script): if opts.return_grid: grids.append(grid) - + all_images = grids + all_images processed = Processed(p, all_images, initial_seed, initial_info) diff --git a/scripts/poor_mans_outpainting.py b/scripts/poor_mans_outpainting.py index c0bbecc1..ea0632b6 100644 --- a/scripts/poor_mans_outpainting.py +++ b/scripts/poor_mans_outpainting.py @@ -19,7 +19,7 @@ class Script(scripts.Script): def ui(self, is_img2img): if not is_img2img: return None - + pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128, elem_id=self.elem_id("pixels")) mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id=self.elem_id("mask_blur")) inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='fill', type="index", elem_id=self.elem_id("inpainting_fill")) diff --git a/scripts/prompt_matrix.py b/scripts/prompt_matrix.py index fb06beab..88324fe6 100644 --- a/scripts/prompt_matrix.py +++ b/scripts/prompt_matrix.py @@ -96,7 +96,7 @@ class Script(scripts.Script): p.prompt_for_display = positive_prompt processed = process_images(p) - grid = images.image_grid(processed.images, p.batch_size, rows=1 << ((len(prompt_matrix_parts) - 1) // 2)) + grid = images.image_grid(processed.images, p.batch_size, rows=1 << ((len(prompt_matrix_parts) - 1) // 2)) grid = images.draw_prompt_matrix(grid, processed.images[0].width, processed.images[0].height, prompt_matrix_parts, margin_size) processed.images.insert(0, grid) processed.index_of_first_image = 1 diff --git a/scripts/prompts_from_file.py b/scripts/prompts_from_file.py index 9607077a..2378816f 100644 --- a/scripts/prompts_from_file.py +++ b/scripts/prompts_from_file.py @@ -109,7 +109,7 @@ class Script(scripts.Script): def title(self): return "Prompts from file or textbox" - def ui(self, is_img2img): + def ui(self, is_img2img): checkbox_iterate = gr.Checkbox(label="Iterate seed every line", value=False, elem_id=self.elem_id("checkbox_iterate")) checkbox_iterate_batch = gr.Checkbox(label="Use same random seed for all lines", value=False, elem_id=self.elem_id("checkbox_iterate_batch")) @@ -166,7 +166,7 @@ class Script(scripts.Script): proc = process_images(copy_p) images += proc.images - + if checkbox_iterate: p.seed = p.seed + (p.batch_size * p.n_iter) all_prompts += proc.all_prompts diff --git a/scripts/sd_upscale.py b/scripts/sd_upscale.py index 0b1d3096..e614c23b 100644 --- a/scripts/sd_upscale.py +++ b/scripts/sd_upscale.py @@ -16,7 +16,7 @@ class Script(scripts.Script): def show(self, is_img2img): return is_img2img - def ui(self, is_img2img): + def ui(self, is_img2img): info = gr.HTML("

Will upscale the image by the selected scale factor; use width and height sliders to set tile size

") overlap = gr.Slider(minimum=0, maximum=256, step=16, label='Tile overlap', value=64, elem_id=self.elem_id("overlap")) scale_factor = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label='Scale Factor', value=2.0, elem_id=self.elem_id("scale_factor")) -- cgit v1.2.3 From 2582a0fd3b3e91c5fba9e5e561cbdf5fee835063 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Thu, 18 May 2023 22:48:28 +0300 Subject: make it possible for scripts to add cross attention optimizations add UI selection for cross attention optimization --- modules/cmd_args.py | 14 ++-- modules/script_callbacks.py | 21 ++++++ modules/sd_hijack.py | 90 ++++++++++++++----------- modules/sd_hijack_optimizations.py | 135 ++++++++++++++++++++++++++++++++++++- modules/shared.py | 1 + modules/shared_items.py | 8 +++ webui.py | 10 +++ 7 files changed, 228 insertions(+), 51 deletions(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/modules/cmd_args.py b/modules/cmd_args.py index 7bde161e..85db93f3 100644 --- a/modules/cmd_args.py +++ b/modules/cmd_args.py @@ -53,16 +53,16 @@ parser.add_argument("--xformers", action='store_true', help="enable xformers for parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work") parser.add_argument("--xformers-flash-attention", action='store_true', help="enable xformers with Flash Attention to improve reproducibility (supported for SD2.x or variant only)") parser.add_argument("--deepdanbooru", action='store_true', help="does not do anything") -parser.add_argument("--opt-split-attention", action='store_true', help="force-enables Doggettx's cross-attention layer optimization. By default, it's on for torch cuda.") -parser.add_argument("--opt-sub-quad-attention", action='store_true', help="enable memory efficient sub-quadratic cross-attention layer optimization") +parser.add_argument("--opt-split-attention", action='store_true', help="prefer Doggettx's cross-attention layer optimization for automatic choice of optimization") +parser.add_argument("--opt-sub-quad-attention", action='store_true', help="prefer memory efficient sub-quadratic cross-attention layer optimization for automatic choice of optimization") parser.add_argument("--sub-quad-q-chunk-size", type=int, help="query chunk size for the sub-quadratic cross-attention layer optimization to use", default=1024) parser.add_argument("--sub-quad-kv-chunk-size", type=int, help="kv chunk size for the sub-quadratic cross-attention layer optimization to use", default=None) parser.add_argument("--sub-quad-chunk-threshold", type=int, help="the percentage of VRAM threshold for the sub-quadratic cross-attention layer optimization to use chunking", default=None) -parser.add_argument("--opt-split-attention-invokeai", action='store_true', help="force-enables InvokeAI's cross-attention layer optimization. By default, it's on when cuda is unavailable.") -parser.add_argument("--opt-split-attention-v1", action='store_true', help="enable older version of split attention optimization that does not consume all the VRAM it can find") -parser.add_argument("--opt-sdp-attention", action='store_true', help="enable scaled dot product cross-attention layer optimization; requires PyTorch 2.*") -parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="enable scaled dot product cross-attention layer optimization without memory efficient attention, makes image generation deterministic; requires PyTorch 2.*") -parser.add_argument("--disable-opt-split-attention", action='store_true', help="force-disables cross-attention layer optimization") +parser.add_argument("--opt-split-attention-invokeai", action='store_true', help="prefer InvokeAI's cross-attention layer optimization for automatic choice of optimization") +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-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/script_callbacks.py b/modules/script_callbacks.py index 3c21a362..40f388a5 100644 --- a/modules/script_callbacks.py +++ b/modules/script_callbacks.py @@ -110,6 +110,7 @@ callback_map = dict( callbacks_script_unloaded=[], callbacks_before_ui=[], callbacks_on_reload=[], + callbacks_list_optimizers=[], ) @@ -258,6 +259,18 @@ def before_ui_callback(): report_exception(c, 'before_ui') +def list_optimizers_callback(): + res = [] + + for c in callback_map['callbacks_list_optimizers']: + try: + c.callback(res) + except Exception: + report_exception(c, 'list_optimizers') + + return res + + def add_callback(callbacks, fun): stack = [x for x in inspect.stack() if x.filename != __file__] filename = stack[0].filename if len(stack) > 0 else 'unknown file' @@ -409,3 +422,11 @@ def on_before_ui(callback): """register a function to be called before the UI is created.""" add_callback(callback_map['callbacks_before_ui'], callback) + + +def on_list_optimizers(callback): + """register a function to be called when UI is making a list of cross attention optimization options. + The function will be called with one argument, a list, and shall add objects of type modules.sd_hijack_optimizations.SdOptimization + to it.""" + + add_callback(callback_map['callbacks_list_optimizers'], callback) diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 14e7f799..39193be8 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -3,8 +3,9 @@ from torch.nn.functional import silu from types import MethodType import modules.textual_inversion.textual_inversion -from modules import devices, sd_hijack_optimizations, shared +from modules import devices, sd_hijack_optimizations, shared, script_callbacks, errors from modules.hypernetworks import hypernetwork +from modules.sd_hijack_optimizations import diffusionmodules_model_AttnBlock_forward from modules.shared import cmd_opts from modules import sd_hijack_clip, sd_hijack_open_clip, sd_hijack_unet, sd_hijack_xlmr, xlmr @@ -28,57 +29,56 @@ ldm.modules.attention.BasicTransformerBlock.ATTENTION_MODES["softmax-xformers"] ldm.modules.attention.print = lambda *args: None ldm.modules.diffusionmodules.model.print = lambda *args: None +optimizers = [] +current_optimizer: sd_hijack_optimizations.SdOptimization = None + + +def list_optimizers(): + new_optimizers = script_callbacks.list_optimizers_callback() + + new_optimizers = [x for x in new_optimizers if x.is_available()] + + new_optimizers = sorted(new_optimizers, key=lambda x: x.priority(), reverse=True) + + optimizers.clear() + optimizers.extend(new_optimizers) + def apply_optimizations(): + global current_optimizer + undo_optimizations() ldm.modules.diffusionmodules.model.nonlinearity = silu ldm.modules.diffusionmodules.openaimodel.th = sd_hijack_unet.th - optimization_method = None + if current_optimizer is not None: + current_optimizer.undo() + current_optimizer = None + + selection = shared.opts.cross_attention_optimization + if selection == "Automatic" and len(optimizers) > 0: + matching_optimizer = next(iter([x for x in optimizers if x.cmd_opt and getattr(shared.cmd_opts, x.cmd_opt, False)]), optimizers[0]) + else: + matching_optimizer = next(iter([x for x in optimizers if x.title() == selection]), None) - can_use_sdp = hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) # not everyone has torch 2.x to use sdp - - if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)): - print("Applying xformers cross attention optimization.") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.xformers_attention_forward - ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.xformers_attnblock_forward - optimization_method = 'xformers' - elif cmd_opts.opt_sdp_no_mem_attention and can_use_sdp: - print("Applying scaled dot product cross attention optimization (without memory efficient attention).") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.scaled_dot_product_no_mem_attention_forward - ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.sdp_no_mem_attnblock_forward - optimization_method = 'sdp-no-mem' - elif cmd_opts.opt_sdp_attention and can_use_sdp: - print("Applying scaled dot product cross attention optimization.") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.scaled_dot_product_attention_forward - ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.sdp_attnblock_forward - optimization_method = 'sdp' - elif cmd_opts.opt_sub_quad_attention: - print("Applying sub-quadratic cross attention optimization.") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.sub_quad_attention_forward - ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.sub_quad_attnblock_forward - optimization_method = 'sub-quadratic' - elif cmd_opts.opt_split_attention_v1: - print("Applying v1 cross attention optimization.") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_v1 - optimization_method = 'V1' - elif not cmd_opts.disable_opt_split_attention and (cmd_opts.opt_split_attention_invokeai or not cmd_opts.opt_split_attention and not torch.cuda.is_available()): - print("Applying cross attention optimization (InvokeAI).") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_invokeAI - optimization_method = 'InvokeAI' - elif not cmd_opts.disable_opt_split_attention and (cmd_opts.opt_split_attention or torch.cuda.is_available()): - print("Applying cross attention optimization (Doggettx).") - ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward - ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.cross_attention_attnblock_forward - optimization_method = 'Doggettx' - - return optimization_method + if selection == "None": + matching_optimizer = None + elif matching_optimizer is None: + matching_optimizer = optimizers[0] + + if matching_optimizer is not None: + print(f"Applying optimization: {matching_optimizer.name}") + matching_optimizer.apply() + current_optimizer = matching_optimizer + return current_optimizer.name + else: + return '' def undo_optimizations(): - ldm.modules.attention.CrossAttention.forward = hypernetwork.attention_CrossAttention_forward ldm.modules.diffusionmodules.model.nonlinearity = diffusionmodules_model_nonlinearity + ldm.modules.attention.CrossAttention.forward = hypernetwork.attention_CrossAttention_forward ldm.modules.diffusionmodules.model.AttnBlock.forward = diffusionmodules_model_AttnBlock_forward @@ -169,7 +169,11 @@ class StableDiffusionModelHijack: if m.cond_stage_key == "edit": sd_hijack_unet.hijack_ddpm_edit() - self.optimization_method = apply_optimizations() + try: + self.optimization_method = apply_optimizations() + except Exception as e: + errors.display(e, "applying cross attention optimization") + undo_optimizations() self.clip = m.cond_stage_model @@ -223,6 +227,10 @@ class StableDiffusionModelHijack: return token_count, self.clip.get_target_prompt_token_count(token_count) + def redo_hijack(self, m): + self.undo_hijack(m) + self.hijack(m) + class EmbeddingsWithFixes(torch.nn.Module): def __init__(self, wrapped, embeddings): diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index f00fe55c..1c5b709b 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -9,10 +9,139 @@ from torch import einsum from ldm.util import default from einops import rearrange -from modules import shared, errors, devices +from modules import shared, errors, devices, sub_quadratic_attention, script_callbacks from modules.hypernetworks import hypernetwork -from .sub_quadratic_attention import efficient_dot_product_attention +import ldm.modules.attention +import ldm.modules.diffusionmodules.model + +diffusionmodules_model_AttnBlock_forward = ldm.modules.diffusionmodules.model.AttnBlock.forward + + +class SdOptimization: + def __init__(self, name, label=None, cmd_opt=None): + self.name = name + self.label = label + self.cmd_opt = cmd_opt + + def title(self): + if self.label is None: + return self.name + + return f"{self.name} - {self.label}" + + def is_available(self): + return True + + def priority(self): + return 0 + + def apply(self): + pass + + def undo(self): + ldm.modules.attention.CrossAttention.forward = hypernetwork.attention_CrossAttention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = diffusionmodules_model_AttnBlock_forward + + +class SdOptimizationXformers(SdOptimization): + def __init__(self): + super().__init__("xformers", cmd_opt="xformers") + + def is_available(self): + return shared.cmd_opts.force_enable_xformers or (shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)) + + def priority(self): + return 100 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = xformers_attention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = xformers_attnblock_forward + + +class SdOptimizationSdpNoMem(SdOptimization): + def __init__(self, name="sdp-no-mem", label="scaled dot product without memory efficient attention", cmd_opt="opt_sdp_no_mem_attention"): + super().__init__(name, label, cmd_opt) + + def is_available(self): + return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) + + def priority(self): + return 90 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = scaled_dot_product_no_mem_attention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = sdp_no_mem_attnblock_forward + + +class SdOptimizationSdp(SdOptimizationSdpNoMem): + def __init__(self): + super().__init__("sdp", "scaled dot product", cmd_opt="opt_sdp_attention") + + def priority(self): + return 80 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = sdp_attnblock_forward + + +class SdOptimizationSubQuad(SdOptimization): + def __init__(self): + super().__init__("sub-quadratic", cmd_opt="opt_sub_quad_attention") + + def priority(self): + return 10 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = sub_quad_attention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = sub_quad_attnblock_forward + + +class SdOptimizationV1(SdOptimization): + def __init__(self): + super().__init__("V1", "original v1", cmd_opt="opt_split_attention_v1") + + def priority(self): + return 10 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward_v1 + + +class SdOptimizationInvokeAI(SdOptimization): + def __init__(self): + super().__init__("InvokeAI", cmd_opt="opt_split_attention_invokeai") + + def priority(self): + return 1000 if not torch.cuda.is_available() else 10 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward_invokeAI + + +class SdOptimizationDoggettx(SdOptimization): + def __init__(self): + super().__init__("Doggettx", cmd_opt="opt_split_attention") + + def priority(self): + return 20 + + def apply(self): + ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward + ldm.modules.diffusionmodules.model.AttnBlock.forward = cross_attention_attnblock_forward + + +def list_optimizers(res): + res.extend([ + SdOptimizationXformers(), + SdOptimizationSdpNoMem(), + SdOptimizationSdp(), + SdOptimizationSubQuad(), + SdOptimizationV1(), + SdOptimizationInvokeAI(), + SdOptimizationDoggettx(), + ]) if shared.cmd_opts.xformers or shared.cmd_opts.force_enable_xformers: @@ -299,7 +428,7 @@ def sub_quad_attention(q, k, v, q_chunk_size=1024, kv_chunk_size=None, kv_chunk_ kv_chunk_size = k_tokens with devices.without_autocast(disable=q.dtype == v.dtype): - return efficient_dot_product_attention( + return sub_quadratic_attention.efficient_dot_product_attention( q, k, v, diff --git a/modules/shared.py b/modules/shared.py index fdbab5c4..7cfbaa0c 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -417,6 +417,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), { })) options_templates.update(options_section(('optimizations', "Optimizations"), { + "cross_attention_optimization": OptionInfo("Automatic", "Cross attention optimization", gr.Dropdown, lambda: {"choices": shared_items.cross_attention_optimizations()}), "s_min_uncond": OptionInfo(0, "Negative Guidance minimum sigma", gr.Slider, {"minimum": 0.0, "maximum": 4.0, "step": 0.01}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9177").info("skip negative prompt for some steps when the image is almost ready; 0=disable, higher=faster"), "token_merging_ratio": OptionInfo(0.0, "Token merging ratio", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9256").info("0=disable, higher=faster"), "token_merging_ratio_img2img": OptionInfo(0.0, "Token merging ratio for img2img", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}).info("only applies if non-zero and overrides above"), diff --git a/modules/shared_items.py b/modules/shared_items.py index e792a134..2a8713c8 100644 --- a/modules/shared_items.py +++ b/modules/shared_items.py @@ -21,3 +21,11 @@ def refresh_vae_list(): import modules.sd_vae modules.sd_vae.refresh_vae_list() + + +def cross_attention_optimizations(): + import modules.sd_hijack + + return ["Automatic"] + [x.title() for x in modules.sd_hijack.optimizers] + ["None"] + + diff --git a/webui.py b/webui.py index b4a21e73..afe3c5fa 100644 --- a/webui.py +++ b/webui.py @@ -52,6 +52,7 @@ import modules.img2img import modules.lowvram import modules.scripts import modules.sd_hijack +import modules.sd_hijack_optimizations import modules.sd_models import modules.sd_vae import modules.txt2img @@ -200,6 +201,10 @@ def initialize(): modules.textual_inversion.textual_inversion.list_textual_inversion_templates() startup_timer.record("refresh textual inversion templates") + modules.script_callbacks.on_list_optimizers(modules.sd_hijack_optimizations.list_optimizers) + modules.sd_hijack.list_optimizers() + startup_timer.record("scripts list_optimizers") + # load model in parallel to other startup stuff Thread(target=lambda: shared.sd_model).start() @@ -208,6 +213,7 @@ def initialize(): shared.opts.onchange("sd_vae_as_default", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False) shared.opts.onchange("temp_dir", ui_tempdir.on_tmpdir_changed) shared.opts.onchange("gradio_theme", shared.reload_gradio_theme) + shared.opts.onchange("cross_attention_optimization", wrap_queued_call(lambda: modules.sd_hijack.model_hijack.redo_hijack(shared.sd_model)), call=False) startup_timer.record("opts onchange") shared.reload_hypernetworks() @@ -428,6 +434,10 @@ def webui(): extra_networks.register_extra_network(extra_networks_hypernet.ExtraNetworkHypernet()) startup_timer.record("initialize extra networks") + modules.script_callbacks.on_list_optimizers(modules.sd_hijack_optimizations.list_optimizers) + modules.sd_hijack.list_optimizers() + startup_timer.record("scripts list_optimizers") + if __name__ == "__main__": if cmd_opts.nowebui: -- cgit v1.2.3 From 8a3d232839930376898634f65bd6c16f3a41e5b4 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Fri, 19 May 2023 00:03:27 +0300 Subject: fix linter issues --- modules/sd_hijack.py | 1 - modules/sd_hijack_optimizations.py | 2 +- 2 files changed, 1 insertion(+), 2 deletions(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 39193be8..75f1c540 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -5,7 +5,6 @@ from types import MethodType import modules.textual_inversion.textual_inversion from modules import devices, sd_hijack_optimizations, shared, script_callbacks, errors from modules.hypernetworks import hypernetwork -from modules.sd_hijack_optimizations import diffusionmodules_model_AttnBlock_forward from modules.shared import cmd_opts from modules import sd_hijack_clip, sd_hijack_open_clip, sd_hijack_unet, sd_hijack_xlmr, xlmr diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index 1c5b709b..db1e4367 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -9,7 +9,7 @@ from torch import einsum from ldm.util import default from einops import rearrange -from modules import shared, errors, devices, sub_quadratic_attention, script_callbacks +from modules import shared, errors, devices, sub_quadratic_attention from modules.hypernetworks import hypernetwork import ldm.modules.attention -- cgit v1.2.3 From 1e5afd4fa9774314d649bddb3d18a9a75871902b Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Fri, 19 May 2023 09:17:36 +0300 Subject: Apply suggestions from code review Co-authored-by: Aarni Koskela --- modules/sd_hijack_optimizations.py | 66 ++++++++++++++++---------------------- 1 file changed, 28 insertions(+), 38 deletions(-) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index db1e4367..0eb4c525 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -19,10 +19,10 @@ diffusionmodules_model_AttnBlock_forward = ldm.modules.diffusionmodules.model.At class SdOptimization: - def __init__(self, name, label=None, cmd_opt=None): - self.name = name - self.label = label - self.cmd_opt = cmd_opt + name: str = None + label: str | None = None + cmd_opt: str | None = None + priority: int = 0 def title(self): if self.label is None: @@ -33,9 +33,6 @@ class SdOptimization: def is_available(self): return True - def priority(self): - return 0 - def apply(self): pass @@ -45,41 +42,37 @@ class SdOptimization: class SdOptimizationXformers(SdOptimization): - def __init__(self): - super().__init__("xformers", cmd_opt="xformers") + name = "xformers" + cmd_opt = "xformers" + priority = 100 def is_available(self): return shared.cmd_opts.force_enable_xformers or (shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)) - def priority(self): - return 100 - def apply(self): ldm.modules.attention.CrossAttention.forward = xformers_attention_forward ldm.modules.diffusionmodules.model.AttnBlock.forward = xformers_attnblock_forward class SdOptimizationSdpNoMem(SdOptimization): - def __init__(self, name="sdp-no-mem", label="scaled dot product without memory efficient attention", cmd_opt="opt_sdp_no_mem_attention"): - super().__init__(name, label, cmd_opt) + name = "sdp-no-mem" + label = "scaled dot product without memory efficient attention" + cmd_opt = "opt_sdp_no_mem_attention" + priority = 90 def is_available(self): return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) - def priority(self): - return 90 - def apply(self): ldm.modules.attention.CrossAttention.forward = scaled_dot_product_no_mem_attention_forward ldm.modules.diffusionmodules.model.AttnBlock.forward = sdp_no_mem_attnblock_forward class SdOptimizationSdp(SdOptimizationSdpNoMem): - def __init__(self): - super().__init__("sdp", "scaled dot product", cmd_opt="opt_sdp_attention") - - def priority(self): - return 80 + name = "sdp" + label = "scaled dot product" + cmd_opt = "opt_sdp_attention" + priority = 80 def apply(self): ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward @@ -87,11 +80,9 @@ class SdOptimizationSdp(SdOptimizationSdpNoMem): class SdOptimizationSubQuad(SdOptimization): - def __init__(self): - super().__init__("sub-quadratic", cmd_opt="opt_sub_quad_attention") - - def priority(self): - return 10 + name = "sub-quadratic" + cmd_opt = "opt_sub_quad_attention" + priority = 10 def apply(self): ldm.modules.attention.CrossAttention.forward = sub_quad_attention_forward @@ -99,20 +90,21 @@ class SdOptimizationSubQuad(SdOptimization): class SdOptimizationV1(SdOptimization): - def __init__(self): - super().__init__("V1", "original v1", cmd_opt="opt_split_attention_v1") + name = "V1" + label = "original v1" + cmd_opt = "opt_split_attention_v1" + priority = 10 - def priority(self): - return 10 def apply(self): ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward_v1 class SdOptimizationInvokeAI(SdOptimization): - def __init__(self): - super().__init__("InvokeAI", cmd_opt="opt_split_attention_invokeai") + name = "InvokeAI" + cmd_opt = "opt_split_attention_invokeai" + @property def priority(self): return 1000 if not torch.cuda.is_available() else 10 @@ -121,11 +113,9 @@ class SdOptimizationInvokeAI(SdOptimization): class SdOptimizationDoggettx(SdOptimization): - def __init__(self): - super().__init__("Doggettx", cmd_opt="opt_split_attention") - - def priority(self): - return 20 + name = "Doggettx" + cmd_opt = "opt_split_attention" + priority = 20 def apply(self): ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward -- cgit v1.2.3 From df004be2fc4b2c68adfb75565d97551a1a5e7ed6 Mon Sep 17 00:00:00 2001 From: Aarni Koskela Date: Sun, 21 May 2023 00:26:16 +0300 Subject: Add a couple `from __future__ import annotations`es for Py3.9 compat --- modules/sd_hijack_optimizations.py | 1 + webui.py | 2 ++ 2 files changed, 3 insertions(+) (limited to 'modules/sd_hijack_optimizations.py') diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py index 0eb4c525..2ec0b049 100644 --- a/modules/sd_hijack_optimizations.py +++ b/modules/sd_hijack_optimizations.py @@ -1,3 +1,4 @@ +from __future__ import annotations import math import sys import traceback diff --git a/webui.py b/webui.py index a76e377c..d4402f55 100644 --- a/webui.py +++ b/webui.py @@ -1,3 +1,5 @@ +from __future__ import annotations + import os import sys import time -- cgit v1.2.3