1 files changed, 144 insertions, 5 deletions
diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py
index 9fc89dc6..8863107a 100644
--- a/modules/sd_disable_initialization.py
+++ b/modules/sd_disable_initialization.py
@@ -3,8 +3,31 @@ import open_clip
 import torch
 import transformers.utils.hub
 
+from modules import shared
 
-class DisableInitialization:
+
+class ReplaceHelper:
+    def __init__(self):
+        self.replaced = []
+
+    def replace(self, obj, field, func):
+        original = getattr(obj, field, None)
+        if original is None:
+            return None
+
+        self.replaced.append((obj, field, original))
+        setattr(obj, field, func)
+
+        return original
+
+    def restore(self):
+        for obj, field, original in self.replaced:
+            setattr(obj, field, original)
+
+        self.replaced.clear()
+
+
+class DisableInitialization(ReplaceHelper):
     """
     When an object of this class enters a `with` block, it starts:
     - preventing torch's layer initialization functions from working
@@ -21,7 +44,7 @@ class DisableInitialization:
     """
 
     def __init__(self, disable_clip=True):
-        self.replaced = []
+        super().__init__()
         self.disable_clip = disable_clip
 
     def replace(self, obj, field, func):
@@ -86,8 +109,124 @@ class DisableInitialization:
             self.transformers_utils_hub_get_from_cache = self.replace(transformers.utils.hub, 'get_from_cache', transformers_utils_hub_get_from_cache)
 
     def __exit__(self, exc_type, exc_val, exc_tb):
-        for obj, field, original in self.replaced:
-            setattr(obj, field, original)
+        self.restore()
 
-        self.replaced.clear()
 
+class InitializeOnMeta(ReplaceHelper):
+    """
+    Context manager that causes all parameters for linear/conv2d/mha layers to be allocated on meta device,
+    which results in those parameters having no values and taking no memory. model.to() will be broken and
+    will need to be repaired by using LoadStateDictOnMeta below when loading params from state dict.
+
+    Usage:
+    ```
+    with sd_disable_initialization.InitializeOnMeta():
+        sd_model = instantiate_from_config(sd_config.model)
+    ```
+    """
+
+    def __enter__(self):
+        if shared.cmd_opts.disable_model_loading_ram_optimization:
+            return
+
+        def set_device(x):
+            x["device"] = "meta"
+            return x
+
+        linear_init = self.replace(torch.nn.Linear, '__init__', lambda *args, **kwargs: linear_init(*args, **set_device(kwargs)))
+        conv2d_init = self.replace(torch.nn.Conv2d, '__init__', lambda *args, **kwargs: conv2d_init(*args, **set_device(kwargs)))
+        mha_init = self.replace(torch.nn.MultiheadAttention, '__init__', lambda *args, **kwargs: mha_init(*args, **set_device(kwargs)))
+        self.replace(torch.nn.Module, 'to', lambda *args, **kwargs: None)
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.restore()
+
+
+class LoadStateDictOnMeta(ReplaceHelper):
+    """
+    Context manager that allows to read parameters from state_dict into a model that has some of its parameters in the meta device.
+    As those parameters are read from state_dict, they will be deleted from it, so by the end state_dict will be mostly empty, to save memory.
+    Meant to be used together with InitializeOnMeta above.
+
+    Usage:
+    ```
+    with sd_disable_initialization.LoadStateDictOnMeta(state_dict):
+        model.load_state_dict(state_dict, strict=False)
+    ```
+    """
+
+    def __init__(self, state_dict, device, weight_dtype_conversion=None):
+        super().__init__()
+        self.state_dict = state_dict
+        self.device = device
+        self.weight_dtype_conversion = weight_dtype_conversion or {}
+        self.default_dtype = self.weight_dtype_conversion.get('')
+
+    def get_weight_dtype(self, key):
+        key_first_term, _ = key.split('.', 1)
+        return self.weight_dtype_conversion.get(key_first_term, self.default_dtype)
+
+    def __enter__(self):
+        if shared.cmd_opts.disable_model_loading_ram_optimization:
+            return
+
+        sd = self.state_dict
+        device = self.device
+
+        def load_from_state_dict(original, module, state_dict, prefix, *args, **kwargs):
+            used_param_keys = []
+
+            for name, param in module._parameters.items():
+                if param is None:
+                    continue
+
+                key = prefix + name
+                sd_param = sd.pop(key, None)
+                if sd_param is not None:
+                    state_dict[key] = sd_param.to(dtype=self.get_weight_dtype(key))
+                    used_param_keys.append(key)
+
+                if param.is_meta:
+                    dtype = sd_param.dtype if sd_param is not None else param.dtype
+                    module._parameters[name] = torch.nn.parameter.Parameter(torch.zeros_like(param, device=device, dtype=dtype), requires_grad=param.requires_grad)
+
+            for name in module._buffers:
+                key = prefix + name
+
+                sd_param = sd.pop(key, None)
+                if sd_param is not None:
+                    state_dict[key] = sd_param
+                    used_param_keys.append(key)
+
+            original(module, state_dict, prefix, *args, **kwargs)
+
+            for key in used_param_keys:
+                state_dict.pop(key, None)
+
+        def load_state_dict(original, module, state_dict, strict=True):
+            """torch makes a lot of copies of the dictionary with weights, so just deleting entries from state_dict does not help
+            because the same values are stored in multiple copies of the dict. The trick used here is to give torch a dict with
+            all weights on meta device, i.e. deleted, and then it doesn't matter how many copies torch makes.
+
+            In _load_from_state_dict, the correct weight will be obtained from a single dict with the right weights (sd).
+
+            The dangerous thing about this is if _load_from_state_dict is not called, (if some exotic module overloads
+            the function and does not call the original) the state dict will just fail to load because weights
+            would be on the meta device.
+            """
+
+            if state_dict == sd:
+                state_dict = {k: v.to(device="meta", dtype=v.dtype) for k, v in state_dict.items()}
+
+            original(module, state_dict, strict=strict)
+
+        module_load_state_dict = self.replace(torch.nn.Module, 'load_state_dict', lambda *args, **kwargs: load_state_dict(module_load_state_dict, *args, **kwargs))
+        module_load_from_state_dict = self.replace(torch.nn.Module, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(module_load_from_state_dict, *args, **kwargs))
+        linear_load_from_state_dict = self.replace(torch.nn.Linear, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(linear_load_from_state_dict, *args, **kwargs))
+        conv2d_load_from_state_dict = self.replace(torch.nn.Conv2d, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(conv2d_load_from_state_dict, *args, **kwargs))
+        mha_load_from_state_dict = self.replace(torch.nn.MultiheadAttention, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(mha_load_from_state_dict, *args, **kwargs))
+        layer_norm_load_from_state_dict = self.replace(torch.nn.LayerNorm, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(layer_norm_load_from_state_dict, *args, **kwargs))
+        group_norm_load_from_state_dict = self.replace(torch.nn.GroupNorm, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(group_norm_load_from_state_dict, *args, **kwargs))
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.restore()