mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-20 07:18:14 +00:00
flux
This commit is contained in:
Artiprocher
@@ -285,6 +285,34 @@ flux_series = [
|
||||
"model_class": "diffsynth.models.flux_text_encoder_t5.FluxTextEncoderT5",
|
||||
"state_dict_converter": "diffsynth.utils.state_dict_converters.flux_text_encoder_t5.FluxTextEncoderT5StateDictConverter",
|
||||
},
|
||||
{
|
||||
# Example: ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="ae.safetensors")
|
||||
"model_hash": "21ea55f476dfc4fd135587abb59dfe5d",
|
||||
"model_name": "flux_vae_encoder",
|
||||
"model_class": "diffsynth.models.flux_vae.FluxVAEEncoder",
|
||||
"state_dict_converter": "diffsynth.utils.state_dict_converters.flux_vae.FluxVAEEncoderStateDictConverter",
|
||||
},
|
||||
{
|
||||
# Example: ModelConfig(model_id="black-forest-labs/FLUX.1-dev", origin_file_pattern="ae.safetensors")
|
||||
"model_hash": "21ea55f476dfc4fd135587abb59dfe5d",
|
||||
"model_name": "flux_vae_decoder",
|
||||
"model_class": "diffsynth.models.flux_vae.FluxVAEDecoder",
|
||||
"state_dict_converter": "diffsynth.utils.state_dict_converters.flux_vae.FluxVAEDecoderStateDictConverter",
|
||||
},
|
||||
{
|
||||
# Example: ModelConfig(model_id="ostris/Flex.2-preview", origin_file_pattern="Flex.2-preview.safetensors")
|
||||
"model_hash": "d02f41c13549fa5093d3521f62a5570a",
|
||||
"model_name": "flux_dit",
|
||||
"model_class": "diffsynth.models.flux_dit.FluxDiT",
|
||||
"extra_kwargs": {'input_dim': 196, 'num_blocks': 8},
|
||||
"state_dict_converter": "diffsynth.utils.state_dict_converters.flux_dit.FluxDiTStateDictConverter",
|
||||
},
|
||||
{
|
||||
# Example: ModelConfig(model_id="DiffSynth-Studio/AttriCtrl-FLUX.1-Dev", origin_file_pattern="models/brightness.safetensors")
|
||||
"model_hash": "0629116fce1472503a66992f96f3eb1a",
|
||||
"model_name": "flux_value_controller",
|
||||
"model_class": "diffsynth.models.flux_value_control.SingleValueEncoder",
|
||||
}
|
||||
]
|
||||
|
||||
MODEL_CONFIGS = qwen_image_series + wan_series + flux_series
|
||||
|
||||
@@ -1,9 +1,38 @@
|
||||
from .svd_image_encoder import SVDImageEncoder
|
||||
from .sd3_dit import RMSNorm
|
||||
from transformers import CLIPImageProcessor
|
||||
from .general_modules import RMSNorm
|
||||
from transformers import SiglipVisionModel, SiglipVisionConfig
|
||||
import torch
|
||||
|
||||
|
||||
class SiglipVisionModelSO400M(SiglipVisionModel):
|
||||
def __init__(self):
|
||||
config = SiglipVisionConfig(**{
|
||||
"architectures": [
|
||||
"SiglipModel"
|
||||
],
|
||||
"initializer_factor": 1.0,
|
||||
"model_type": "siglip",
|
||||
"text_config": {
|
||||
"hidden_size": 1152,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "siglip_text_model",
|
||||
"num_attention_heads": 16,
|
||||
"num_hidden_layers": 27
|
||||
},
|
||||
"torch_dtype": "float32",
|
||||
"transformers_version": "4.37.0.dev0",
|
||||
"vision_config": {
|
||||
"hidden_size": 1152,
|
||||
"image_size": 384,
|
||||
"intermediate_size": 4304,
|
||||
"model_type": "siglip_vision_model",
|
||||
"num_attention_heads": 16,
|
||||
"num_hidden_layers": 27,
|
||||
"patch_size": 14
|
||||
}
|
||||
})
|
||||
super().__init__(config)
|
||||
|
||||
|
||||
class MLPProjModel(torch.nn.Module):
|
||||
def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, num_tokens=4):
|
||||
super().__init__()
|
||||
|
||||
@@ -106,7 +106,7 @@ class TileWorker:
|
||||
return model_output
|
||||
|
||||
|
||||
class Attention(torch.nn.Module):
|
||||
class ConvAttention(torch.nn.Module):
|
||||
|
||||
def __init__(self, q_dim, num_heads, head_dim, kv_dim=None, bias_q=False, bias_kv=False, bias_out=False):
|
||||
super().__init__()
|
||||
@@ -115,10 +115,10 @@ class Attention(torch.nn.Module):
|
||||
self.num_heads = num_heads
|
||||
self.head_dim = head_dim
|
||||
|
||||
self.to_q = torch.nn.Linear(q_dim, dim_inner, bias=bias_q)
|
||||
self.to_k = torch.nn.Linear(kv_dim, dim_inner, bias=bias_kv)
|
||||
self.to_v = torch.nn.Linear(kv_dim, dim_inner, bias=bias_kv)
|
||||
self.to_out = torch.nn.Linear(dim_inner, q_dim, bias=bias_out)
|
||||
self.to_q = torch.nn.Conv2d(q_dim, dim_inner, kernel_size=(1, 1), bias=bias_q)
|
||||
self.to_k = torch.nn.Conv2d(kv_dim, dim_inner, kernel_size=(1, 1), bias=bias_kv)
|
||||
self.to_v = torch.nn.Conv2d(kv_dim, dim_inner, kernel_size=(1, 1), bias=bias_kv)
|
||||
self.to_out = torch.nn.Conv2d(dim_inner, q_dim, kernel_size=(1, 1), bias=bias_out)
|
||||
|
||||
def forward(self, hidden_states, encoder_hidden_states=None, attn_mask=None):
|
||||
if encoder_hidden_states is None:
|
||||
@@ -126,9 +126,14 @@ class Attention(torch.nn.Module):
|
||||
|
||||
batch_size = encoder_hidden_states.shape[0]
|
||||
|
||||
q = self.to_q(hidden_states)
|
||||
k = self.to_k(encoder_hidden_states)
|
||||
v = self.to_v(encoder_hidden_states)
|
||||
conv_input = rearrange(hidden_states, "B L C -> B C L 1")
|
||||
q = self.to_q(conv_input)
|
||||
q = rearrange(q[:, :, :, 0], "B C L -> B L C")
|
||||
conv_input = rearrange(encoder_hidden_states, "B L C -> B C L 1")
|
||||
k = self.to_k(conv_input)
|
||||
v = self.to_v(conv_input)
|
||||
k = rearrange(k[:, :, :, 0], "B C L -> B L C")
|
||||
v = rearrange(v[:, :, :, 0], "B C L -> B L C")
|
||||
|
||||
q = q.view(batch_size, -1, self.num_heads, self.head_dim).transpose(1, 2)
|
||||
k = k.view(batch_size, -1, self.num_heads, self.head_dim).transpose(1, 2)
|
||||
@@ -138,7 +143,9 @@ class Attention(torch.nn.Module):
|
||||
hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, self.num_heads * self.head_dim)
|
||||
hidden_states = hidden_states.to(q.dtype)
|
||||
|
||||
hidden_states = self.to_out(hidden_states)
|
||||
conv_input = rearrange(hidden_states, "B L C -> B C L 1")
|
||||
hidden_states = self.to_out(conv_input)
|
||||
hidden_states = rearrange(hidden_states[:, :, :, 0], "B C L -> B L C")
|
||||
|
||||
return hidden_states
|
||||
|
||||
@@ -152,7 +159,7 @@ class VAEAttentionBlock(torch.nn.Module):
|
||||
self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=eps, affine=True)
|
||||
|
||||
self.transformer_blocks = torch.nn.ModuleList([
|
||||
Attention(
|
||||
ConvAttention(
|
||||
inner_dim,
|
||||
num_attention_heads,
|
||||
attention_head_dim,
|
||||
@@ -236,7 +243,7 @@ class DownSampler(torch.nn.Module):
|
||||
return hidden_states, time_emb, text_emb, res_stack
|
||||
|
||||
|
||||
class SD3VAEDecoder(torch.nn.Module):
|
||||
class FluxVAEDecoder(torch.nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.scaling_factor = 0.3611
|
||||
@@ -308,7 +315,7 @@ class SD3VAEDecoder(torch.nn.Module):
|
||||
return hidden_states
|
||||
|
||||
|
||||
class SD3VAEEncoder(torch.nn.Module):
|
||||
class FluxVAEEncoder(torch.nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.scaling_factor = 0.3611
|
||||
|
||||
@@ -1,10 +1,12 @@
|
||||
import torch
|
||||
from diffsynth.models.svd_unet import TemporalTimesteps
|
||||
from .general_modules import TemporalTimesteps
|
||||
|
||||
|
||||
class MultiValueEncoder(torch.nn.Module):
|
||||
def __init__(self, encoders=()):
|
||||
super().__init__()
|
||||
if not isinstance(encoders, list):
|
||||
encoders = [encoders]
|
||||
self.encoders = torch.nn.ModuleList(encoders)
|
||||
|
||||
def __call__(self, values, dtype):
|
||||
|
||||
1163
diffsynth/pipelines/flux_image.py
Normal file
1163
diffsynth/pipelines/flux_image.py
Normal file
File diff suppressed because it is too large
Load Diff
204
diffsynth/utils/lora/flux.py
Normal file
204
diffsynth/utils/lora/flux.py
Normal file
@@ -0,0 +1,204 @@
|
||||
from .general import GeneralLoRALoader
|
||||
import torch, math
|
||||
|
||||
|
||||
class FluxLoRALoader(GeneralLoRALoader):
|
||||
def __init__(self, device="cpu", torch_dtype=torch.float32):
|
||||
super().__init__(device=device, torch_dtype=torch_dtype)
|
||||
|
||||
self.diffusers_rename_dict = {
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_k.lora_A.weight":"single_blocks.blockid.a_to_k.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_k.lora_B.weight":"single_blocks.blockid.a_to_k.lora_B.weight",
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_q.lora_A.weight":"single_blocks.blockid.a_to_q.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_q.lora_B.weight":"single_blocks.blockid.a_to_q.lora_B.weight",
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_v.lora_A.weight":"single_blocks.blockid.a_to_v.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.attn.to_v.lora_B.weight":"single_blocks.blockid.a_to_v.lora_B.weight",
|
||||
"transformer.single_transformer_blocks.blockid.norm.linear.lora_A.weight":"single_blocks.blockid.norm.linear.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.norm.linear.lora_B.weight":"single_blocks.blockid.norm.linear.lora_B.weight",
|
||||
"transformer.single_transformer_blocks.blockid.proj_mlp.lora_A.weight":"single_blocks.blockid.proj_in_besides_attn.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.proj_mlp.lora_B.weight":"single_blocks.blockid.proj_in_besides_attn.lora_B.weight",
|
||||
"transformer.single_transformer_blocks.blockid.proj_out.lora_A.weight":"single_blocks.blockid.proj_out.lora_A.weight",
|
||||
"transformer.single_transformer_blocks.blockid.proj_out.lora_B.weight":"single_blocks.blockid.proj_out.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_k_proj.lora_A.weight":"blocks.blockid.attn.b_to_k.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_k_proj.lora_B.weight":"blocks.blockid.attn.b_to_k.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_q_proj.lora_A.weight":"blocks.blockid.attn.b_to_q.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_q_proj.lora_B.weight":"blocks.blockid.attn.b_to_q.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_v_proj.lora_A.weight":"blocks.blockid.attn.b_to_v.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.add_v_proj.lora_B.weight":"blocks.blockid.attn.b_to_v.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_add_out.lora_A.weight":"blocks.blockid.attn.b_to_out.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_add_out.lora_B.weight":"blocks.blockid.attn.b_to_out.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_k.lora_A.weight":"blocks.blockid.attn.a_to_k.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_k.lora_B.weight":"blocks.blockid.attn.a_to_k.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_out.0.lora_A.weight":"blocks.blockid.attn.a_to_out.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_out.0.lora_B.weight":"blocks.blockid.attn.a_to_out.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_q.lora_A.weight":"blocks.blockid.attn.a_to_q.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_q.lora_B.weight":"blocks.blockid.attn.a_to_q.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_v.lora_A.weight":"blocks.blockid.attn.a_to_v.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.attn.to_v.lora_B.weight":"blocks.blockid.attn.a_to_v.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.ff.net.0.proj.lora_A.weight":"blocks.blockid.ff_a.0.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.ff.net.0.proj.lora_B.weight":"blocks.blockid.ff_a.0.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.ff.net.2.lora_A.weight":"blocks.blockid.ff_a.2.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.ff.net.2.lora_B.weight":"blocks.blockid.ff_a.2.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.ff_context.net.0.proj.lora_A.weight":"blocks.blockid.ff_b.0.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.ff_context.net.0.proj.lora_B.weight":"blocks.blockid.ff_b.0.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.ff_context.net.2.lora_A.weight":"blocks.blockid.ff_b.2.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.ff_context.net.2.lora_B.weight":"blocks.blockid.ff_b.2.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.norm1.linear.lora_A.weight":"blocks.blockid.norm1_a.linear.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.norm1.linear.lora_B.weight":"blocks.blockid.norm1_a.linear.lora_B.weight",
|
||||
"transformer.transformer_blocks.blockid.norm1_context.linear.lora_A.weight":"blocks.blockid.norm1_b.linear.lora_A.weight",
|
||||
"transformer.transformer_blocks.blockid.norm1_context.linear.lora_B.weight":"blocks.blockid.norm1_b.linear.lora_B.weight",
|
||||
}
|
||||
|
||||
self.civitai_rename_dict = {
|
||||
"lora_unet_double_blocks_blockid_img_mod_lin.lora_down.weight": "blocks.blockid.norm1_a.linear.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_img_mod_lin.lora_up.weight": "blocks.blockid.norm1_a.linear.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mod_lin.lora_down.weight": "blocks.blockid.norm1_b.linear.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mod_lin.lora_up.weight": "blocks.blockid.norm1_b.linear.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_img_attn_qkv.lora_down.weight": "blocks.blockid.attn.a_to_qkv.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_img_attn_qkv.lora_up.weight": "blocks.blockid.attn.a_to_qkv.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_attn_qkv.lora_down.weight": "blocks.blockid.attn.b_to_qkv.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_attn_qkv.lora_up.weight": "blocks.blockid.attn.b_to_qkv.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_img_attn_proj.lora_down.weight": "blocks.blockid.attn.a_to_out.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_img_attn_proj.lora_up.weight": "blocks.blockid.attn.a_to_out.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_attn_proj.lora_down.weight": "blocks.blockid.attn.b_to_out.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_attn_proj.lora_up.weight": "blocks.blockid.attn.b_to_out.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_img_mlp_0.lora_down.weight": "blocks.blockid.ff_a.0.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_img_mlp_0.lora_up.weight": "blocks.blockid.ff_a.0.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_img_mlp_2.lora_down.weight": "blocks.blockid.ff_a.2.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_img_mlp_2.lora_up.weight": "blocks.blockid.ff_a.2.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mlp_0.lora_down.weight": "blocks.blockid.ff_b.0.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mlp_0.lora_up.weight": "blocks.blockid.ff_b.0.lora_B.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mlp_2.lora_down.weight": "blocks.blockid.ff_b.2.lora_A.weight",
|
||||
"lora_unet_double_blocks_blockid_txt_mlp_2.lora_up.weight": "blocks.blockid.ff_b.2.lora_B.weight",
|
||||
"lora_unet_single_blocks_blockid_modulation_lin.lora_down.weight": "single_blocks.blockid.norm.linear.lora_A.weight",
|
||||
"lora_unet_single_blocks_blockid_modulation_lin.lora_up.weight": "single_blocks.blockid.norm.linear.lora_B.weight",
|
||||
"lora_unet_single_blocks_blockid_linear1.lora_down.weight": "single_blocks.blockid.to_qkv_mlp.lora_A.weight",
|
||||
"lora_unet_single_blocks_blockid_linear1.lora_up.weight": "single_blocks.blockid.to_qkv_mlp.lora_B.weight",
|
||||
"lora_unet_single_blocks_blockid_linear2.lora_down.weight": "single_blocks.blockid.proj_out.lora_A.weight",
|
||||
"lora_unet_single_blocks_blockid_linear2.lora_up.weight": "single_blocks.blockid.proj_out.lora_B.weight",
|
||||
}
|
||||
|
||||
def fuse_lora_to_base_model(self, model: torch.nn.Module, state_dict_lora, alpha=1.0):
|
||||
super().fuse_lora_to_base_model(model, state_dict_lora, alpha)
|
||||
|
||||
def convert_state_dict(self, state_dict):
|
||||
|
||||
def guess_block_id(name,model_resource):
|
||||
if model_resource == 'civitai':
|
||||
names = name.split("_")
|
||||
for i in names:
|
||||
if i.isdigit():
|
||||
return i, name.replace(f"_{i}_", "_blockid_")
|
||||
if model_resource == 'diffusers':
|
||||
names = name.split(".")
|
||||
for i in names:
|
||||
if i.isdigit():
|
||||
return i, name.replace(f"transformer_blocks.{i}.", "transformer_blocks.blockid.")
|
||||
return None, None
|
||||
|
||||
def guess_resource(state_dict):
|
||||
for k in state_dict:
|
||||
if "lora_unet_" in k:
|
||||
return 'civitai'
|
||||
elif k.startswith("transformer."):
|
||||
return 'diffusers'
|
||||
else:
|
||||
None
|
||||
|
||||
model_resource = guess_resource(state_dict)
|
||||
if model_resource is None:
|
||||
return state_dict
|
||||
|
||||
rename_dict = self.diffusers_rename_dict if model_resource == 'diffusers' else self.civitai_rename_dict
|
||||
def guess_alpha(state_dict):
|
||||
for name, param in state_dict.items():
|
||||
if ".alpha" in name:
|
||||
for suffix in [".lora_down.weight", ".lora_A.weight"]:
|
||||
name_ = name.replace(".alpha", suffix)
|
||||
if name_ in state_dict:
|
||||
lora_alpha = param.item() / state_dict[name_].shape[0]
|
||||
lora_alpha = math.sqrt(lora_alpha)
|
||||
return lora_alpha
|
||||
|
||||
return 1
|
||||
|
||||
alpha = guess_alpha(state_dict)
|
||||
|
||||
state_dict_ = {}
|
||||
for name, param in state_dict.items():
|
||||
block_id, source_name = guess_block_id(name,model_resource)
|
||||
if alpha != 1:
|
||||
param *= alpha
|
||||
if source_name in rename_dict:
|
||||
target_name = rename_dict[source_name]
|
||||
target_name = target_name.replace(".blockid.", f".{block_id}.")
|
||||
state_dict_[target_name] = param
|
||||
else:
|
||||
state_dict_[name] = param
|
||||
|
||||
if model_resource == 'diffusers':
|
||||
for name in list(state_dict_.keys()):
|
||||
if "single_blocks." in name and ".a_to_q." in name:
|
||||
mlp = state_dict_.get(name.replace(".a_to_q.", ".proj_in_besides_attn."), None)
|
||||
if mlp is None:
|
||||
dim = 4
|
||||
if 'lora_A' in name:
|
||||
dim = 1
|
||||
mlp = torch.zeros(dim * state_dict_[name].shape[0],
|
||||
*state_dict_[name].shape[1:],
|
||||
dtype=state_dict_[name].dtype)
|
||||
else:
|
||||
state_dict_.pop(name.replace(".a_to_q.", ".proj_in_besides_attn."))
|
||||
if 'lora_A' in name:
|
||||
param = torch.concat([
|
||||
state_dict_.pop(name),
|
||||
state_dict_.pop(name.replace(".a_to_q.", ".a_to_k.")),
|
||||
state_dict_.pop(name.replace(".a_to_q.", ".a_to_v.")),
|
||||
mlp,
|
||||
], dim=0)
|
||||
elif 'lora_B' in name:
|
||||
d, r = state_dict_[name].shape
|
||||
param = torch.zeros((3*d+mlp.shape[0], 3*r+mlp.shape[1]), dtype=state_dict_[name].dtype, device=state_dict_[name].device)
|
||||
param[:d, :r] = state_dict_.pop(name)
|
||||
param[d:2*d, r:2*r] = state_dict_.pop(name.replace(".a_to_q.", ".a_to_k."))
|
||||
param[2*d:3*d, 2*r:3*r] = state_dict_.pop(name.replace(".a_to_q.", ".a_to_v."))
|
||||
param[3*d:, 3*r:] = mlp
|
||||
else:
|
||||
param = torch.concat([
|
||||
state_dict_.pop(name),
|
||||
state_dict_.pop(name.replace(".a_to_q.", ".a_to_k.")),
|
||||
state_dict_.pop(name.replace(".a_to_q.", ".a_to_v.")),
|
||||
mlp,
|
||||
], dim=0)
|
||||
name_ = name.replace(".a_to_q.", ".to_qkv_mlp.")
|
||||
state_dict_[name_] = param
|
||||
for name in list(state_dict_.keys()):
|
||||
for component in ["a", "b"]:
|
||||
if f".{component}_to_q." in name:
|
||||
name_ = name.replace(f".{component}_to_q.", f".{component}_to_qkv.")
|
||||
concat_dim = 0
|
||||
if 'lora_A' in name:
|
||||
param = torch.concat([
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_q.")],
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_k.")],
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_v.")],
|
||||
], dim=0)
|
||||
elif 'lora_B' in name:
|
||||
origin = state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_q.")]
|
||||
d, r = origin.shape
|
||||
# print(d, r)
|
||||
param = torch.zeros((3*d, 3*r), dtype=origin.dtype, device=origin.device)
|
||||
param[:d, :r] = state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_q.")]
|
||||
param[d:2*d, r:2*r] = state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_k.")]
|
||||
param[2*d:3*d, 2*r:3*r] = state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_v.")]
|
||||
else:
|
||||
param = torch.concat([
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_q.")],
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_k.")],
|
||||
state_dict_[name.replace(f".{component}_to_q.", f".{component}_to_v.")],
|
||||
], dim=0)
|
||||
state_dict_[name_] = param
|
||||
state_dict_.pop(name.replace(f".{component}_to_q.", f".{component}_to_q."))
|
||||
state_dict_.pop(name.replace(f".{component}_to_q.", f".{component}_to_k."))
|
||||
state_dict_.pop(name.replace(f".{component}_to_q.", f".{component}_to_v."))
|
||||
return state_dict_
|
||||
264
diffsynth/utils/state_dict_converters/flux_vae.py
Normal file
264
diffsynth/utils/state_dict_converters/flux_vae.py
Normal file
@@ -0,0 +1,264 @@
|
||||
def FluxVAEEncoderStateDictConverter(state_dict):
|
||||
rename_dict = {
|
||||
"encoder.conv_in.bias": "conv_in.bias",
|
||||
"encoder.conv_in.weight": "conv_in.weight",
|
||||
"encoder.conv_out.bias": "conv_out.bias",
|
||||
"encoder.conv_out.weight": "conv_out.weight",
|
||||
"encoder.down.0.block.0.conv1.bias": "blocks.0.conv1.bias",
|
||||
"encoder.down.0.block.0.conv1.weight": "blocks.0.conv1.weight",
|
||||
"encoder.down.0.block.0.conv2.bias": "blocks.0.conv2.bias",
|
||||
"encoder.down.0.block.0.conv2.weight": "blocks.0.conv2.weight",
|
||||
"encoder.down.0.block.0.norm1.bias": "blocks.0.norm1.bias",
|
||||
"encoder.down.0.block.0.norm1.weight": "blocks.0.norm1.weight",
|
||||
"encoder.down.0.block.0.norm2.bias": "blocks.0.norm2.bias",
|
||||
"encoder.down.0.block.0.norm2.weight": "blocks.0.norm2.weight",
|
||||
"encoder.down.0.block.1.conv1.bias": "blocks.1.conv1.bias",
|
||||
"encoder.down.0.block.1.conv1.weight": "blocks.1.conv1.weight",
|
||||
"encoder.down.0.block.1.conv2.bias": "blocks.1.conv2.bias",
|
||||
"encoder.down.0.block.1.conv2.weight": "blocks.1.conv2.weight",
|
||||
"encoder.down.0.block.1.norm1.bias": "blocks.1.norm1.bias",
|
||||
"encoder.down.0.block.1.norm1.weight": "blocks.1.norm1.weight",
|
||||
"encoder.down.0.block.1.norm2.bias": "blocks.1.norm2.bias",
|
||||
"encoder.down.0.block.1.norm2.weight": "blocks.1.norm2.weight",
|
||||
"encoder.down.0.downsample.conv.bias": "blocks.2.conv.bias",
|
||||
"encoder.down.0.downsample.conv.weight": "blocks.2.conv.weight",
|
||||
"encoder.down.1.block.0.conv1.bias": "blocks.3.conv1.bias",
|
||||
"encoder.down.1.block.0.conv1.weight": "blocks.3.conv1.weight",
|
||||
"encoder.down.1.block.0.conv2.bias": "blocks.3.conv2.bias",
|
||||
"encoder.down.1.block.0.conv2.weight": "blocks.3.conv2.weight",
|
||||
"encoder.down.1.block.0.nin_shortcut.bias": "blocks.3.conv_shortcut.bias",
|
||||
"encoder.down.1.block.0.nin_shortcut.weight": "blocks.3.conv_shortcut.weight",
|
||||
"encoder.down.1.block.0.norm1.bias": "blocks.3.norm1.bias",
|
||||
"encoder.down.1.block.0.norm1.weight": "blocks.3.norm1.weight",
|
||||
"encoder.down.1.block.0.norm2.bias": "blocks.3.norm2.bias",
|
||||
"encoder.down.1.block.0.norm2.weight": "blocks.3.norm2.weight",
|
||||
"encoder.down.1.block.1.conv1.bias": "blocks.4.conv1.bias",
|
||||
"encoder.down.1.block.1.conv1.weight": "blocks.4.conv1.weight",
|
||||
"encoder.down.1.block.1.conv2.bias": "blocks.4.conv2.bias",
|
||||
"encoder.down.1.block.1.conv2.weight": "blocks.4.conv2.weight",
|
||||
"encoder.down.1.block.1.norm1.bias": "blocks.4.norm1.bias",
|
||||
"encoder.down.1.block.1.norm1.weight": "blocks.4.norm1.weight",
|
||||
"encoder.down.1.block.1.norm2.bias": "blocks.4.norm2.bias",
|
||||
"encoder.down.1.block.1.norm2.weight": "blocks.4.norm2.weight",
|
||||
"encoder.down.1.downsample.conv.bias": "blocks.5.conv.bias",
|
||||
"encoder.down.1.downsample.conv.weight": "blocks.5.conv.weight",
|
||||
"encoder.down.2.block.0.conv1.bias": "blocks.6.conv1.bias",
|
||||
"encoder.down.2.block.0.conv1.weight": "blocks.6.conv1.weight",
|
||||
"encoder.down.2.block.0.conv2.bias": "blocks.6.conv2.bias",
|
||||
"encoder.down.2.block.0.conv2.weight": "blocks.6.conv2.weight",
|
||||
"encoder.down.2.block.0.nin_shortcut.bias": "blocks.6.conv_shortcut.bias",
|
||||
"encoder.down.2.block.0.nin_shortcut.weight": "blocks.6.conv_shortcut.weight",
|
||||
"encoder.down.2.block.0.norm1.bias": "blocks.6.norm1.bias",
|
||||
"encoder.down.2.block.0.norm1.weight": "blocks.6.norm1.weight",
|
||||
"encoder.down.2.block.0.norm2.bias": "blocks.6.norm2.bias",
|
||||
"encoder.down.2.block.0.norm2.weight": "blocks.6.norm2.weight",
|
||||
"encoder.down.2.block.1.conv1.bias": "blocks.7.conv1.bias",
|
||||
"encoder.down.2.block.1.conv1.weight": "blocks.7.conv1.weight",
|
||||
"encoder.down.2.block.1.conv2.bias": "blocks.7.conv2.bias",
|
||||
"encoder.down.2.block.1.conv2.weight": "blocks.7.conv2.weight",
|
||||
"encoder.down.2.block.1.norm1.bias": "blocks.7.norm1.bias",
|
||||
"encoder.down.2.block.1.norm1.weight": "blocks.7.norm1.weight",
|
||||
"encoder.down.2.block.1.norm2.bias": "blocks.7.norm2.bias",
|
||||
"encoder.down.2.block.1.norm2.weight": "blocks.7.norm2.weight",
|
||||
"encoder.down.2.downsample.conv.bias": "blocks.8.conv.bias",
|
||||
"encoder.down.2.downsample.conv.weight": "blocks.8.conv.weight",
|
||||
"encoder.down.3.block.0.conv1.bias": "blocks.9.conv1.bias",
|
||||
"encoder.down.3.block.0.conv1.weight": "blocks.9.conv1.weight",
|
||||
"encoder.down.3.block.0.conv2.bias": "blocks.9.conv2.bias",
|
||||
"encoder.down.3.block.0.conv2.weight": "blocks.9.conv2.weight",
|
||||
"encoder.down.3.block.0.norm1.bias": "blocks.9.norm1.bias",
|
||||
"encoder.down.3.block.0.norm1.weight": "blocks.9.norm1.weight",
|
||||
"encoder.down.3.block.0.norm2.bias": "blocks.9.norm2.bias",
|
||||
"encoder.down.3.block.0.norm2.weight": "blocks.9.norm2.weight",
|
||||
"encoder.down.3.block.1.conv1.bias": "blocks.10.conv1.bias",
|
||||
"encoder.down.3.block.1.conv1.weight": "blocks.10.conv1.weight",
|
||||
"encoder.down.3.block.1.conv2.bias": "blocks.10.conv2.bias",
|
||||
"encoder.down.3.block.1.conv2.weight": "blocks.10.conv2.weight",
|
||||
"encoder.down.3.block.1.norm1.bias": "blocks.10.norm1.bias",
|
||||
"encoder.down.3.block.1.norm1.weight": "blocks.10.norm1.weight",
|
||||
"encoder.down.3.block.1.norm2.bias": "blocks.10.norm2.bias",
|
||||
"encoder.down.3.block.1.norm2.weight": "blocks.10.norm2.weight",
|
||||
"encoder.mid.attn_1.k.bias": "blocks.12.transformer_blocks.0.to_k.bias",
|
||||
"encoder.mid.attn_1.k.weight": "blocks.12.transformer_blocks.0.to_k.weight",
|
||||
"encoder.mid.attn_1.norm.bias": "blocks.12.norm.bias",
|
||||
"encoder.mid.attn_1.norm.weight": "blocks.12.norm.weight",
|
||||
"encoder.mid.attn_1.proj_out.bias": "blocks.12.transformer_blocks.0.to_out.bias",
|
||||
"encoder.mid.attn_1.proj_out.weight": "blocks.12.transformer_blocks.0.to_out.weight",
|
||||
"encoder.mid.attn_1.q.bias": "blocks.12.transformer_blocks.0.to_q.bias",
|
||||
"encoder.mid.attn_1.q.weight": "blocks.12.transformer_blocks.0.to_q.weight",
|
||||
"encoder.mid.attn_1.v.bias": "blocks.12.transformer_blocks.0.to_v.bias",
|
||||
"encoder.mid.attn_1.v.weight": "blocks.12.transformer_blocks.0.to_v.weight",
|
||||
"encoder.mid.block_1.conv1.bias": "blocks.11.conv1.bias",
|
||||
"encoder.mid.block_1.conv1.weight": "blocks.11.conv1.weight",
|
||||
"encoder.mid.block_1.conv2.bias": "blocks.11.conv2.bias",
|
||||
"encoder.mid.block_1.conv2.weight": "blocks.11.conv2.weight",
|
||||
"encoder.mid.block_1.norm1.bias": "blocks.11.norm1.bias",
|
||||
"encoder.mid.block_1.norm1.weight": "blocks.11.norm1.weight",
|
||||
"encoder.mid.block_1.norm2.bias": "blocks.11.norm2.bias",
|
||||
"encoder.mid.block_1.norm2.weight": "blocks.11.norm2.weight",
|
||||
"encoder.mid.block_2.conv1.bias": "blocks.13.conv1.bias",
|
||||
"encoder.mid.block_2.conv1.weight": "blocks.13.conv1.weight",
|
||||
"encoder.mid.block_2.conv2.bias": "blocks.13.conv2.bias",
|
||||
"encoder.mid.block_2.conv2.weight": "blocks.13.conv2.weight",
|
||||
"encoder.mid.block_2.norm1.bias": "blocks.13.norm1.bias",
|
||||
"encoder.mid.block_2.norm1.weight": "blocks.13.norm1.weight",
|
||||
"encoder.mid.block_2.norm2.bias": "blocks.13.norm2.bias",
|
||||
"encoder.mid.block_2.norm2.weight": "blocks.13.norm2.weight",
|
||||
"encoder.norm_out.bias": "conv_norm_out.bias",
|
||||
"encoder.norm_out.weight": "conv_norm_out.weight",
|
||||
}
|
||||
state_dict_ = {}
|
||||
for name in state_dict:
|
||||
if name in rename_dict:
|
||||
param = state_dict[name]
|
||||
state_dict_[rename_dict[name]] = param
|
||||
return state_dict_
|
||||
|
||||
|
||||
def FluxVAEDecoderStateDictConverter(state_dict):
|
||||
rename_dict = {
|
||||
"decoder.conv_in.bias": "conv_in.bias",
|
||||
"decoder.conv_in.weight": "conv_in.weight",
|
||||
"decoder.conv_out.bias": "conv_out.bias",
|
||||
"decoder.conv_out.weight": "conv_out.weight",
|
||||
"decoder.mid.attn_1.k.bias": "blocks.1.transformer_blocks.0.to_k.bias",
|
||||
"decoder.mid.attn_1.k.weight": "blocks.1.transformer_blocks.0.to_k.weight",
|
||||
"decoder.mid.attn_1.norm.bias": "blocks.1.norm.bias",
|
||||
"decoder.mid.attn_1.norm.weight": "blocks.1.norm.weight",
|
||||
"decoder.mid.attn_1.proj_out.bias": "blocks.1.transformer_blocks.0.to_out.bias",
|
||||
"decoder.mid.attn_1.proj_out.weight": "blocks.1.transformer_blocks.0.to_out.weight",
|
||||
"decoder.mid.attn_1.q.bias": "blocks.1.transformer_blocks.0.to_q.bias",
|
||||
"decoder.mid.attn_1.q.weight": "blocks.1.transformer_blocks.0.to_q.weight",
|
||||
"decoder.mid.attn_1.v.bias": "blocks.1.transformer_blocks.0.to_v.bias",
|
||||
"decoder.mid.attn_1.v.weight": "blocks.1.transformer_blocks.0.to_v.weight",
|
||||
"decoder.mid.block_1.conv1.bias": "blocks.0.conv1.bias",
|
||||
"decoder.mid.block_1.conv1.weight": "blocks.0.conv1.weight",
|
||||
"decoder.mid.block_1.conv2.bias": "blocks.0.conv2.bias",
|
||||
"decoder.mid.block_1.conv2.weight": "blocks.0.conv2.weight",
|
||||
"decoder.mid.block_1.norm1.bias": "blocks.0.norm1.bias",
|
||||
"decoder.mid.block_1.norm1.weight": "blocks.0.norm1.weight",
|
||||
"decoder.mid.block_1.norm2.bias": "blocks.0.norm2.bias",
|
||||
"decoder.mid.block_1.norm2.weight": "blocks.0.norm2.weight",
|
||||
"decoder.mid.block_2.conv1.bias": "blocks.2.conv1.bias",
|
||||
"decoder.mid.block_2.conv1.weight": "blocks.2.conv1.weight",
|
||||
"decoder.mid.block_2.conv2.bias": "blocks.2.conv2.bias",
|
||||
"decoder.mid.block_2.conv2.weight": "blocks.2.conv2.weight",
|
||||
"decoder.mid.block_2.norm1.bias": "blocks.2.norm1.bias",
|
||||
"decoder.mid.block_2.norm1.weight": "blocks.2.norm1.weight",
|
||||
"decoder.mid.block_2.norm2.bias": "blocks.2.norm2.bias",
|
||||
"decoder.mid.block_2.norm2.weight": "blocks.2.norm2.weight",
|
||||
"decoder.norm_out.bias": "conv_norm_out.bias",
|
||||
"decoder.norm_out.weight": "conv_norm_out.weight",
|
||||
"decoder.up.0.block.0.conv1.bias": "blocks.15.conv1.bias",
|
||||
"decoder.up.0.block.0.conv1.weight": "blocks.15.conv1.weight",
|
||||
"decoder.up.0.block.0.conv2.bias": "blocks.15.conv2.bias",
|
||||
"decoder.up.0.block.0.conv2.weight": "blocks.15.conv2.weight",
|
||||
"decoder.up.0.block.0.nin_shortcut.bias": "blocks.15.conv_shortcut.bias",
|
||||
"decoder.up.0.block.0.nin_shortcut.weight": "blocks.15.conv_shortcut.weight",
|
||||
"decoder.up.0.block.0.norm1.bias": "blocks.15.norm1.bias",
|
||||
"decoder.up.0.block.0.norm1.weight": "blocks.15.norm1.weight",
|
||||
"decoder.up.0.block.0.norm2.bias": "blocks.15.norm2.bias",
|
||||
"decoder.up.0.block.0.norm2.weight": "blocks.15.norm2.weight",
|
||||
"decoder.up.0.block.1.conv1.bias": "blocks.16.conv1.bias",
|
||||
"decoder.up.0.block.1.conv1.weight": "blocks.16.conv1.weight",
|
||||
"decoder.up.0.block.1.conv2.bias": "blocks.16.conv2.bias",
|
||||
"decoder.up.0.block.1.conv2.weight": "blocks.16.conv2.weight",
|
||||
"decoder.up.0.block.1.norm1.bias": "blocks.16.norm1.bias",
|
||||
"decoder.up.0.block.1.norm1.weight": "blocks.16.norm1.weight",
|
||||
"decoder.up.0.block.1.norm2.bias": "blocks.16.norm2.bias",
|
||||
"decoder.up.0.block.1.norm2.weight": "blocks.16.norm2.weight",
|
||||
"decoder.up.0.block.2.conv1.bias": "blocks.17.conv1.bias",
|
||||
"decoder.up.0.block.2.conv1.weight": "blocks.17.conv1.weight",
|
||||
"decoder.up.0.block.2.conv2.bias": "blocks.17.conv2.bias",
|
||||
"decoder.up.0.block.2.conv2.weight": "blocks.17.conv2.weight",
|
||||
"decoder.up.0.block.2.norm1.bias": "blocks.17.norm1.bias",
|
||||
"decoder.up.0.block.2.norm1.weight": "blocks.17.norm1.weight",
|
||||
"decoder.up.0.block.2.norm2.bias": "blocks.17.norm2.bias",
|
||||
"decoder.up.0.block.2.norm2.weight": "blocks.17.norm2.weight",
|
||||
"decoder.up.1.block.0.conv1.bias": "blocks.11.conv1.bias",
|
||||
"decoder.up.1.block.0.conv1.weight": "blocks.11.conv1.weight",
|
||||
"decoder.up.1.block.0.conv2.bias": "blocks.11.conv2.bias",
|
||||
"decoder.up.1.block.0.conv2.weight": "blocks.11.conv2.weight",
|
||||
"decoder.up.1.block.0.nin_shortcut.bias": "blocks.11.conv_shortcut.bias",
|
||||
"decoder.up.1.block.0.nin_shortcut.weight": "blocks.11.conv_shortcut.weight",
|
||||
"decoder.up.1.block.0.norm1.bias": "blocks.11.norm1.bias",
|
||||
"decoder.up.1.block.0.norm1.weight": "blocks.11.norm1.weight",
|
||||
"decoder.up.1.block.0.norm2.bias": "blocks.11.norm2.bias",
|
||||
"decoder.up.1.block.0.norm2.weight": "blocks.11.norm2.weight",
|
||||
"decoder.up.1.block.1.conv1.bias": "blocks.12.conv1.bias",
|
||||
"decoder.up.1.block.1.conv1.weight": "blocks.12.conv1.weight",
|
||||
"decoder.up.1.block.1.conv2.bias": "blocks.12.conv2.bias",
|
||||
"decoder.up.1.block.1.conv2.weight": "blocks.12.conv2.weight",
|
||||
"decoder.up.1.block.1.norm1.bias": "blocks.12.norm1.bias",
|
||||
"decoder.up.1.block.1.norm1.weight": "blocks.12.norm1.weight",
|
||||
"decoder.up.1.block.1.norm2.bias": "blocks.12.norm2.bias",
|
||||
"decoder.up.1.block.1.norm2.weight": "blocks.12.norm2.weight",
|
||||
"decoder.up.1.block.2.conv1.bias": "blocks.13.conv1.bias",
|
||||
"decoder.up.1.block.2.conv1.weight": "blocks.13.conv1.weight",
|
||||
"decoder.up.1.block.2.conv2.bias": "blocks.13.conv2.bias",
|
||||
"decoder.up.1.block.2.conv2.weight": "blocks.13.conv2.weight",
|
||||
"decoder.up.1.block.2.norm1.bias": "blocks.13.norm1.bias",
|
||||
"decoder.up.1.block.2.norm1.weight": "blocks.13.norm1.weight",
|
||||
"decoder.up.1.block.2.norm2.bias": "blocks.13.norm2.bias",
|
||||
"decoder.up.1.block.2.norm2.weight": "blocks.13.norm2.weight",
|
||||
"decoder.up.1.upsample.conv.bias": "blocks.14.conv.bias",
|
||||
"decoder.up.1.upsample.conv.weight": "blocks.14.conv.weight",
|
||||
"decoder.up.2.block.0.conv1.bias": "blocks.7.conv1.bias",
|
||||
"decoder.up.2.block.0.conv1.weight": "blocks.7.conv1.weight",
|
||||
"decoder.up.2.block.0.conv2.bias": "blocks.7.conv2.bias",
|
||||
"decoder.up.2.block.0.conv2.weight": "blocks.7.conv2.weight",
|
||||
"decoder.up.2.block.0.norm1.bias": "blocks.7.norm1.bias",
|
||||
"decoder.up.2.block.0.norm1.weight": "blocks.7.norm1.weight",
|
||||
"decoder.up.2.block.0.norm2.bias": "blocks.7.norm2.bias",
|
||||
"decoder.up.2.block.0.norm2.weight": "blocks.7.norm2.weight",
|
||||
"decoder.up.2.block.1.conv1.bias": "blocks.8.conv1.bias",
|
||||
"decoder.up.2.block.1.conv1.weight": "blocks.8.conv1.weight",
|
||||
"decoder.up.2.block.1.conv2.bias": "blocks.8.conv2.bias",
|
||||
"decoder.up.2.block.1.conv2.weight": "blocks.8.conv2.weight",
|
||||
"decoder.up.2.block.1.norm1.bias": "blocks.8.norm1.bias",
|
||||
"decoder.up.2.block.1.norm1.weight": "blocks.8.norm1.weight",
|
||||
"decoder.up.2.block.1.norm2.bias": "blocks.8.norm2.bias",
|
||||
"decoder.up.2.block.1.norm2.weight": "blocks.8.norm2.weight",
|
||||
"decoder.up.2.block.2.conv1.bias": "blocks.9.conv1.bias",
|
||||
"decoder.up.2.block.2.conv1.weight": "blocks.9.conv1.weight",
|
||||
"decoder.up.2.block.2.conv2.bias": "blocks.9.conv2.bias",
|
||||
"decoder.up.2.block.2.conv2.weight": "blocks.9.conv2.weight",
|
||||
"decoder.up.2.block.2.norm1.bias": "blocks.9.norm1.bias",
|
||||
"decoder.up.2.block.2.norm1.weight": "blocks.9.norm1.weight",
|
||||
"decoder.up.2.block.2.norm2.bias": "blocks.9.norm2.bias",
|
||||
"decoder.up.2.block.2.norm2.weight": "blocks.9.norm2.weight",
|
||||
"decoder.up.2.upsample.conv.bias": "blocks.10.conv.bias",
|
||||
"decoder.up.2.upsample.conv.weight": "blocks.10.conv.weight",
|
||||
"decoder.up.3.block.0.conv1.bias": "blocks.3.conv1.bias",
|
||||
"decoder.up.3.block.0.conv1.weight": "blocks.3.conv1.weight",
|
||||
"decoder.up.3.block.0.conv2.bias": "blocks.3.conv2.bias",
|
||||
"decoder.up.3.block.0.conv2.weight": "blocks.3.conv2.weight",
|
||||
"decoder.up.3.block.0.norm1.bias": "blocks.3.norm1.bias",
|
||||
"decoder.up.3.block.0.norm1.weight": "blocks.3.norm1.weight",
|
||||
"decoder.up.3.block.0.norm2.bias": "blocks.3.norm2.bias",
|
||||
"decoder.up.3.block.0.norm2.weight": "blocks.3.norm2.weight",
|
||||
"decoder.up.3.block.1.conv1.bias": "blocks.4.conv1.bias",
|
||||
"decoder.up.3.block.1.conv1.weight": "blocks.4.conv1.weight",
|
||||
"decoder.up.3.block.1.conv2.bias": "blocks.4.conv2.bias",
|
||||
"decoder.up.3.block.1.conv2.weight": "blocks.4.conv2.weight",
|
||||
"decoder.up.3.block.1.norm1.bias": "blocks.4.norm1.bias",
|
||||
"decoder.up.3.block.1.norm1.weight": "blocks.4.norm1.weight",
|
||||
"decoder.up.3.block.1.norm2.bias": "blocks.4.norm2.bias",
|
||||
"decoder.up.3.block.1.norm2.weight": "blocks.4.norm2.weight",
|
||||
"decoder.up.3.block.2.conv1.bias": "blocks.5.conv1.bias",
|
||||
"decoder.up.3.block.2.conv1.weight": "blocks.5.conv1.weight",
|
||||
"decoder.up.3.block.2.conv2.bias": "blocks.5.conv2.bias",
|
||||
"decoder.up.3.block.2.conv2.weight": "blocks.5.conv2.weight",
|
||||
"decoder.up.3.block.2.norm1.bias": "blocks.5.norm1.bias",
|
||||
"decoder.up.3.block.2.norm1.weight": "blocks.5.norm1.weight",
|
||||
"decoder.up.3.block.2.norm2.bias": "blocks.5.norm2.bias",
|
||||
"decoder.up.3.block.2.norm2.weight": "blocks.5.norm2.weight",
|
||||
"decoder.up.3.upsample.conv.bias": "blocks.6.conv.bias",
|
||||
"decoder.up.3.upsample.conv.weight": "blocks.6.conv.weight",
|
||||
}
|
||||
state_dict_ = {}
|
||||
for name in state_dict:
|
||||
if name in rename_dict:
|
||||
param = state_dict[name]
|
||||
state_dict_[rename_dict[name]] = param
|
||||
return state_dict_
|
||||
Reference in New Issue
Block a user