modules_forge/unet_patcher.py

import copy
import torch

from ldm_patched.modules.model_patcher import ModelPatcher
from ldm_patched.modules.sampler_helpers import convert_cond
from ldm_patched.modules.samplers import encode_model_conds
from ldm_patched.modules.args_parser import args


class UnetPatcher(ModelPatcher):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.controlnet_linked_list = None
        self.extra_preserved_memory_during_sampling = 0
        self.extra_model_patchers_during_sampling = []
        self.extra_concat_condition = None
        self.compiled = False

    def clone(self):
        n = UnetPatcher(self.model, self.load_device, self.offload_device, self.size)
        n.patches = {}
        for k in self.patches:
            n.patches[k] = self.patches[k][:]
        n.object_patches = self.object_patches.copy()
        n.model_options = copy.deepcopy(self.model_options)
        n.controlnet_linked_list = self.controlnet_linked_list
        n.extra_preserved_memory_during_sampling = self.extra_preserved_memory_during_sampling
        n.extra_model_patchers_during_sampling = self.extra_model_patchers_during_sampling.copy()
        n.extra_concat_condition = self.extra_concat_condition
        n.patches_uuid = self.patches_uuid
        n.backup = self.backup
        n.object_patches_backup = self.object_patches_backup
        n.compiled = self.compiled
        n.parent = self
        return n

    def add_extra_preserved_memory_during_sampling(self, memory_in_bytes: int):
        # Use this to ask Forge to preserve a certain amount of memory during sampling.
        # If GPU VRAM is 8 GB, and memory_in_bytes is 2GB, i.e., memory_in_bytes = 2 * 1024 * 1024 * 1024
        # Then the sampling will always use less than 6GB memory by dynamically offload modules to CPU RAM.
        # You can estimate this using model_management.module_size(any_pytorch_model) to get size of any pytorch models.
        self.extra_preserved_memory_during_sampling += memory_in_bytes
        return

    def add_extra_model_patcher_during_sampling(self, model_patcher: ModelPatcher):
        # Use this to ask Forge to move extra model patchers to GPU during sampling.
        # This method will manage GPU memory perfectly.
        self.extra_model_patchers_during_sampling.append(model_patcher)
        return

    def add_extra_torch_module_during_sampling(self, m: torch.nn.Module, cast_to_unet_dtype: bool = True):
        # Use this method to bind an extra torch.nn.Module to this UNet during sampling.
        # This model `m` will be delegated to Forge memory management system.
        # `m` will be loaded to GPU everytime when sampling starts.
        # `m` will be unloaded if necessary.
        # `m` will influence Forge's judgement about use GPU memory or
        # capacity and decide whether to use module offload to make user's batch size larger.
        # Use cast_to_unet_dtype if you want `m` to have same dtype with unet during sampling.

        if cast_to_unet_dtype:
            m.to(self.model.diffusion_model.dtype)

        patcher = ModelPatcher(model=m, load_device=self.load_device, offload_device=self.offload_device)

        self.add_extra_model_patcher_during_sampling(patcher)
        return patcher
    
    # LoRAs don't work (again)
    def compile_model(self, backend="inductor"):
        """Compile the self model using torch.compile"""
        if not hasattr(torch, 'compile'):
            print("torch.compile not available - requires PyTorch 2.0 or newer")
            return
        
        try:
            torch_version = torch.__version__.split('.')
            if int(torch_version[0]) < 2:
                print(f"torch.compile requires PyTorch 2.0 or newer. Current version: {torch.__version__}")
                return

            import torch._dynamo as dynamo
            dynamo.config.suppress_errors = True
            dynamo.config.verbose = True
            dynamo.config.cache_size_limit = 32

            # Get the actual model to compile
            if hasattr(self.model, 'diffusion_model'):
                real_model = self.model.diffusion_model
            else:
                real_model = self.model

            # Check if any individual options are enabled
            has_custom_options = any([
                args.torch_compile_epilogue_fusion,
                args.torch_compile_max_autotune,
                args.torch_compile_fallback_random,
                args.torch_compile_shape_padding,
                args.torch_compile_cudagraphs,
                args.torch_compile_trace,
                args.torch_compile_graph_diagram
            ])

            if backend == "cudagraphs":
                # Simplified settings for cudagraphs
                compile_settings = {
                    "backend": backend,
                    "fullgraph": True,
                    "dynamic": True,
                }
            else:  # inductor and other backends
                compile_settings = {
                    "backend": backend,
                    "fullgraph": False,
                    "dynamic": True,
                }

                if has_custom_options:
                    # If any custom options are specified, use options instead of mode
                    options = {}
                    if args.torch_compile_epilogue_fusion:
                        options["epilogue_fusion"] = True
                    if args.torch_compile_max_autotune:
                        options["max_autotune"] = True
                    if args.torch_compile_fallback_random:
                        options["fallback_random"] = True
                    if args.torch_compile_shape_padding:
                        options["shape_padding"] = True
                    if args.torch_compile_cudagraphs:
                        options["triton.cudagraphs"] = True
                    if args.torch_compile_trace:
                        options["trace.enabled"] = True
                    if args.torch_compile_graph_diagram:
                        options["trace.graph_diagram"] = True

                    compile_settings["options"] = options
                else:
                    # If no custom options, use the selected mode
                    compile_settings["mode"] = args.torch_compile_mode

            print(f"Compiling model using torch.compile with settings: {compile_settings}")

            # Store settings for later recompilation if needed
            real_model.compile_settings = compile_settings
            
            try:
                compiled_model = torch.compile(real_model, **compile_settings)
                if hasattr(self.model, 'diffusion_model'):
                    self.model.diffusion_model = compiled_model
                else:
                    self.model = compiled_model
                print("Model compilation successful with dynamic shapes support")
                return True
            except Exception as e:
                print(f"Warning: torch.compile failed with error: {str(e)}")
                print("Falling back to uncompiled model")
                return False
        except Exception as e:
            print(f"Error during model compilation: {str(e)}")
            return False

    def add_patched_controlnet(self, cnet):
        cnet.set_previous_controlnet(self.controlnet_linked_list)
        self.controlnet_linked_list = cnet
        return

    def list_controlnets(self):
        results = []
        pointer = self.controlnet_linked_list
        while pointer is not None:
            results.append(pointer)
            pointer = pointer.previous_controlnet
        return results

    def append_model_option(self, k, v, ensure_uniqueness=False):
        if k not in self.model_options:
            self.model_options[k] = []

        if ensure_uniqueness and v in self.model_options[k]:
            return

        self.model_options[k].append(v)
        return

    def append_transformer_option(self, k, v, ensure_uniqueness=False):
        if 'transformer_options' not in self.model_options:
            self.model_options['transformer_options'] = {}

        to = self.model_options['transformer_options']

        if k not in to:
            to[k] = []

        if ensure_uniqueness and v in to[k]:
            return

        to[k].append(v)
        return

    def set_transformer_option(self, k, v):
        if 'transformer_options' not in self.model_options:
            self.model_options['transformer_options'] = {}

        self.model_options['transformer_options'][k] = v
        return

    def add_conditioning_modifier(self, modifier, ensure_uniqueness=False):
        self.append_model_option('conditioning_modifiers', modifier, ensure_uniqueness)
        return

    def add_sampler_pre_cfg_function(self, modifier, ensure_uniqueness=False):
        self.append_model_option('sampler_pre_cfg_function', modifier, ensure_uniqueness)
        return

    def set_memory_peak_estimation_modifier(self, modifier):
        self.model_options['memory_peak_estimation_modifier'] = modifier
        return

    def add_alphas_cumprod_modifier(self, modifier, ensure_uniqueness=False):
        """

        For some reasons, this function only works in A1111's Script.process_batch(self, p, *args, **kwargs)

        For example, below is a worked modification:

        class ExampleScript(scripts.Script):

            def process_batch(self, p, *args, **kwargs):
                unet = p.sd_model.forge_objects.unet.clone()

                def modifier(x):
                    return x ** 0.5

                unet.add_alphas_cumprod_modifier(modifier)
                p.sd_model.forge_objects.unet = unet

                return

        This add_alphas_cumprod_modifier is the only patch option that should be used in process_batch()
        All other patch options should be called in process_before_every_sampling()

        """

        self.append_model_option('alphas_cumprod_modifiers', modifier, ensure_uniqueness)
        return

    def add_block_modifier(self, modifier, ensure_uniqueness=False):
        self.append_transformer_option('block_modifiers', modifier, ensure_uniqueness)
        return

    def add_block_inner_modifier(self, modifier, ensure_uniqueness=False):
        self.append_transformer_option('block_inner_modifiers', modifier, ensure_uniqueness)
        return

    def add_controlnet_conditioning_modifier(self, modifier, ensure_uniqueness=False):
        self.append_transformer_option('controlnet_conditioning_modifiers', modifier, ensure_uniqueness)
        return

    def set_controlnet_model_function_wrapper(self, wrapper):
        self.set_transformer_option('controlnet_model_function_wrapper', wrapper)
        return

    def set_model_replace_all(self, patch, target="attn1"):
        for block_name in ['input', 'middle', 'output']:
            for number in range(16):
                for transformer_index in range(16):
                    self.set_model_patch_replace(patch, target, block_name, number, transformer_index)
        return

    def encode_conds_after_clip(self, conds, noise, prompt_type="positive"):
        return encode_model_conds(
            model_function=self.model.extra_conds,
            conds=convert_cond(conds),
            noise=noise,
            device=noise.device,
            prompt_type=prompt_type
        )

    def load_frozen_patcher(self, state_dict, strength):
        patch_dict = {}
        for k, w in state_dict.items():
            model_key, patch_type, weight_index = k.split('::')
            if model_key not in patch_dict:
                patch_dict[model_key] = {}
            if patch_type not in patch_dict[model_key]:
                patch_dict[model_key][patch_type] = [None] * 16
            patch_dict[model_key][patch_type][int(weight_index)] = w

        patch_flat = {}
        for model_key, v in patch_dict.items():
            for patch_type, weight_list in v.items():
                patch_flat[model_key] = (patch_type, weight_list)

        self.add_patches(patches=patch_flat, strength_patch=float(strength), strength_model=1.0)
        return


def copy_and_update_model_options(model_options, patch, name, block_name, number, transformer_index=None):
    model_options = model_options.copy()
    transformer_options = model_options.get("transformer_options", {}).copy()
    patches_replace = transformer_options.get("patches_replace", {}).copy()
    name_patches = patches_replace.get(name, {}).copy()
    block = (block_name, number, transformer_index) if transformer_index is not None else (block_name, number)
    name_patches[block] = patch
    patches_replace[name] = name_patches
    transformer_options["patches_replace"] = patches_replace
    model_options["transformer_options"] = transformer_options
    return model_options