OPENGIT/stable-diffusion-webui
1
0
Fork 0
mirror of https://github.com/AUTOMATIC1111/stable-diffusion-webui.git synced 2025-01-31 02:32:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: LyCORIS/kohya OFT network support

Browse Source
This commit is contained in:
v0xie 2023-11-15 18:28:48 -08:00
parent d6d0b22e66
commit eb667e715a
1 changed files with 26 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

108
extensions-builtin/Lora/network_oft.py
View File

@ -11,8 +11,8 @@ class ModuleTypeOFT(network.ModuleType):
return None return None
# adapted from kohya-ss' implementation https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py # Supports both kohya-ss' implementation of COFT https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py
# and KohakuBlueleaf's implementation https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/diag_oft.py # and KohakuBlueleaf's implementation of OFT/COFT https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/diag_oft.py
class NetworkModuleOFT(network.NetworkModule): class NetworkModuleOFT(network.NetworkModule):
def __init__(self, net: network.Network, weights: network.NetworkWeights): def __init__(self, net: network.Network, weights: network.NetworkWeights):
@ -25,117 +25,61 @@ class NetworkModuleOFT(network.NetworkModule):
if "oft_blocks" in weights.w.keys(): if "oft_blocks" in weights.w.keys():
self.is_kohya = True self.is_kohya = True
self.oft_blocks = weights.w["oft_blocks"] # (num_blocks, block_size, block_size) self.oft_blocks = weights.w["oft_blocks"] # (num_blocks, block_size, block_size)
self.alpha = weights.w["alpha"] self.alpha = weights.w["alpha"] # alpha is constraint
self.dim = self.oft_blocks.shape[0] # lora dim self.dim = self.oft_blocks.shape[0] # lora dim
#self.oft_blocks = rearrange(self.oft_blocks, 'k m ... -> (k m) ...') # LyCORIS
elif "oft_diag" in weights.w.keys(): elif "oft_diag" in weights.w.keys():
self.is_kohya = False self.is_kohya = False
self.oft_blocks = weights.w["oft_diag"] # (num_blocks, block_size, block_size) self.oft_blocks = weights.w["oft_diag"]
# self.alpha is unused
# alpha is rank if alpha is 0 or None self.dim = self.oft_blocks.shape[1] # (num_blocks, block_size, block_size)
if self.alpha is None:
pass
self.dim = self.oft_blocks.shape[1] # FIXME: almost certainly incorrect, assumes tensor is shape [*, m, n]
else:
raise ValueError("oft_blocks or oft_diag must be in weights dict")
is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear] is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear]
is_conv = type(self.sd_module) in [torch.nn.Conv2d] is_conv = type(self.sd_module) in [torch.nn.Conv2d]
is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] # unsupported
if is_linear: if is_linear:
self.out_dim = self.sd_module.out_features self.out_dim = self.sd_module.out_features
elif is_other_linear:
self.out_dim = self.sd_module.embed_dim
elif is_conv: elif is_conv:
self.out_dim = self.sd_module.out_channels self.out_dim = self.sd_module.out_channels
else: elif is_other_linear:
raise ValueError("sd_module must be Linear or Conv") self.out_dim = self.sd_module.embed_dim
if self.is_kohya: if self.is_kohya:
self.constraint = self.alpha * self.out_dim self.constraint = self.alpha * self.out_dim
self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) self.num_blocks = self.dim
self.block_size = self.out_dim // self.dim
else: else:
self.constraint = None self.constraint = None
self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) self.block_size, self.num_blocks = factorization(self.out_dim, self.dim)
def merge_weight(self, R_weight, org_weight): def calc_updown_kb(self, orig_weight, multiplier):
R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype)
if org_weight.dim() == 4: oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) # ensure skew-symmetric orthogonal matrix
weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight)
else:
weight = torch.einsum("oi, op -> pi", org_weight, R_weight)
return weight
def get_weight(self, oft_blocks, multiplier=None): R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype)
if self.constraint is not None: R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device)
constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype)
block_Q = oft_blocks - oft_blocks.transpose(1, 2) # This errors out for MultiheadAttention, might need to be handled up-stream
norm_Q = torch.norm(block_Q.flatten()) merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size)
if self.constraint is not None: merged_weight = torch.einsum(
new_norm_Q = torch.clamp(norm_Q, max=constraint) 'k n m, k n ... -> k m ...',
else: R,
new_norm_Q = norm_Q merged_weight
block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) )
m_I = torch.eye(self.num_blocks, device=oft_blocks.device).unsqueeze(0).repeat(self.block_size, 1, 1) merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...')
#m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1)
block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse())
block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I
R = torch.block_diag(*block_R_weighted)
return R
def calc_updown_kohya(self, orig_weight, multiplier):
R = self.get_weight(self.oft_blocks, multiplier)
merged_weight = self.merge_weight(R, orig_weight)
updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight
output_shape = orig_weight.shape output_shape = orig_weight.shape
orig_weight = orig_weight
return self.finalize_updown(updown, orig_weight, output_shape)
def calc_updown_kb(self, orig_weight, multiplier):
is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention]
if not is_other_linear:
oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype)
# ensure skew-symmetric matrix
oft_blocks = oft_blocks - oft_blocks.transpose(1, 2)
R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype)
R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device)
merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size)
merged_weight = torch.einsum(
'k n m, k n ... -> k m ...',
R,
merged_weight
)
merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...')
updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight
output_shape = orig_weight.shape
else:
# FIXME: skip MultiheadAttention for now
#up = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype)
updown = torch.zeros([orig_weight.shape[1], orig_weight.shape[1]], device=orig_weight.device, dtype=orig_weight.dtype)
output_shape = (orig_weight.shape[1], orig_weight.shape[1])
return self.finalize_updown(updown, orig_weight, output_shape) return self.finalize_updown(updown, orig_weight, output_shape)
def calc_updown(self, orig_weight): def calc_updown(self, orig_weight):
# if alpha is a very small number as in coft, calc_scale will return a almost zero number so we ignore it # if alpha is a very small number as in coft, calc_scale() will return a almost zero number so we ignore it
#multiplier = self.multiplier() * self.calc_scale()
multiplier = self.multiplier() multiplier = self.multiplier()
return self.calc_updown_kb(orig_weight, multiplier) return self.calc_updown_kb(orig_weight, multiplier)
# override to remove the multiplier/scale factor; it's already multiplied in get_weight # override to remove the multiplier/scale factor; it's already multiplied in get_weight
def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None):
#return super().finalize_updown(updown, orig_weight, output_shape, ex_bias)
if self.bias is not None: if self.bias is not None:
updown = updown.reshape(self.bias.shape) updown = updown.reshape(self.bias.shape)
updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype) updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype)