You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
114 lines
3.8 KiB
114 lines
3.8 KiB
import torch |
|
import comfy.model_management |
|
|
|
def cast_bias_weight(s, input): |
|
bias = None |
|
non_blocking = comfy.model_management.device_supports_non_blocking(input.device) |
|
if s.bias is not None: |
|
bias = s.bias.to(device=input.device, dtype=input.dtype, non_blocking=non_blocking) |
|
weight = s.weight.to(device=input.device, dtype=input.dtype, non_blocking=non_blocking) |
|
return weight, bias |
|
|
|
|
|
class disable_weight_init: |
|
class Linear(torch.nn.Linear): |
|
comfy_cast_weights = False |
|
def reset_parameters(self): |
|
return None |
|
|
|
def forward_comfy_cast_weights(self, input): |
|
weight, bias = cast_bias_weight(self, input) |
|
return torch.nn.functional.linear(input, weight, bias) |
|
|
|
def forward(self, *args, **kwargs): |
|
if self.comfy_cast_weights: |
|
return self.forward_comfy_cast_weights(*args, **kwargs) |
|
else: |
|
return super().forward(*args, **kwargs) |
|
|
|
class Conv2d(torch.nn.Conv2d): |
|
comfy_cast_weights = False |
|
def reset_parameters(self): |
|
return None |
|
|
|
def forward_comfy_cast_weights(self, input): |
|
weight, bias = cast_bias_weight(self, input) |
|
return self._conv_forward(input, weight, bias) |
|
|
|
def forward(self, *args, **kwargs): |
|
if self.comfy_cast_weights: |
|
return self.forward_comfy_cast_weights(*args, **kwargs) |
|
else: |
|
return super().forward(*args, **kwargs) |
|
|
|
class Conv3d(torch.nn.Conv3d): |
|
comfy_cast_weights = False |
|
def reset_parameters(self): |
|
return None |
|
|
|
def forward_comfy_cast_weights(self, input): |
|
weight, bias = cast_bias_weight(self, input) |
|
return self._conv_forward(input, weight, bias) |
|
|
|
def forward(self, *args, **kwargs): |
|
if self.comfy_cast_weights: |
|
return self.forward_comfy_cast_weights(*args, **kwargs) |
|
else: |
|
return super().forward(*args, **kwargs) |
|
|
|
class GroupNorm(torch.nn.GroupNorm): |
|
comfy_cast_weights = False |
|
def reset_parameters(self): |
|
return None |
|
|
|
def forward_comfy_cast_weights(self, input): |
|
weight, bias = cast_bias_weight(self, input) |
|
return torch.nn.functional.group_norm(input, self.num_groups, weight, bias, self.eps) |
|
|
|
def forward(self, *args, **kwargs): |
|
if self.comfy_cast_weights: |
|
return self.forward_comfy_cast_weights(*args, **kwargs) |
|
else: |
|
return super().forward(*args, **kwargs) |
|
|
|
|
|
class LayerNorm(torch.nn.LayerNorm): |
|
comfy_cast_weights = False |
|
def reset_parameters(self): |
|
return None |
|
|
|
def forward_comfy_cast_weights(self, input): |
|
weight, bias = cast_bias_weight(self, input) |
|
return torch.nn.functional.layer_norm(input, self.normalized_shape, weight, bias, self.eps) |
|
|
|
def forward(self, *args, **kwargs): |
|
if self.comfy_cast_weights: |
|
return self.forward_comfy_cast_weights(*args, **kwargs) |
|
else: |
|
return super().forward(*args, **kwargs) |
|
|
|
@classmethod |
|
def conv_nd(s, dims, *args, **kwargs): |
|
if dims == 2: |
|
return s.Conv2d(*args, **kwargs) |
|
elif dims == 3: |
|
return s.Conv3d(*args, **kwargs) |
|
else: |
|
raise ValueError(f"unsupported dimensions: {dims}") |
|
|
|
|
|
class manual_cast(disable_weight_init): |
|
class Linear(disable_weight_init.Linear): |
|
comfy_cast_weights = True |
|
|
|
class Conv2d(disable_weight_init.Conv2d): |
|
comfy_cast_weights = True |
|
|
|
class Conv3d(disable_weight_init.Conv3d): |
|
comfy_cast_weights = True |
|
|
|
class GroupNorm(disable_weight_init.GroupNorm): |
|
comfy_cast_weights = True |
|
|
|
class LayerNorm(disable_weight_init.LayerNorm): |
|
comfy_cast_weights = True
|
|
|