Official weights to LoRA

networks/extract_control_net_lora.py

@@ -0,0 +1,206 @@
+# extract approximating LoRA by svd from SD 1.5 vs ControlNet
+# https://github.com/lllyasviel/ControlNet/blob/main/tool_transfer_control.py
+#
+# The code is based on https://github.com/cloneofsimo/lora/blob/develop/lora_diffusion/cli_svd.py
+# Thanks to cloneofsimo!
+
+import argparse
+import os
+import torch
+from safetensors.torch import load_file, save_file
+from tqdm import tqdm
+from diffusers import UNet2DConditionModel
+
+import library.model_util as model_util
+import control_net_lora
+
+
+CLAMP_QUANTILE = 0.99
+MIN_DIFF = 1e-6
+
+
+def save_to_file(file_name, model, state_dict, dtype):
+  if dtype is not None:
+    for key in list(state_dict.keys()):
+      if type(state_dict[key]) == torch.Tensor:
+        state_dict[key] = state_dict[key].to(dtype)
+
+  if os.path.splitext(file_name)[1] == '.safetensors':
+    save_file(model, file_name)
+  else:
+    torch.save(model, file_name)
+
+
+def svd(args):
+  def str_to_dtype(p):
+    if p == 'float':
+      return torch.float
+    if p == 'fp16':
+      return torch.float16
+    if p == 'bf16':
+      return torch.bfloat16
+    return None
+
+  save_dtype = str_to_dtype(args.save_precision)
+
+  # Diffusersのキーに変換するため、original sdとcontrol sdからU-Netに重みを読み込む ###############
+
+  # original sdをDiffusersに読み込む
+  print(f"loading original SD model : {args.model_org}")
+  org_text_encoder, _, org_unet = model_util.load_models_from_stable_diffusion_checkpoint(args.v2, args.model_org)
+
+  org_sd = torch.load(args.model_org, map_location='cpu')
+  if 'state_dict' in org_sd:
+    org_sd = org_sd['state_dict']
+
+  # control sdからキー変換しつつU-Netに対応する部分のみ取り出す
+  print(f"loading control SD model : {args.model_tuned}")
+
+  ctrl_sd = torch.load(args.model_tuned, map_location='cpu')
+  ctrl_unet_sd = org_sd                      # あらかじめloadしておくことでcontrol sdにない部分はoriginal sdと同じにする
+  for key in list(ctrl_sd.keys()):
+    if key.startswith("control_"):
+      unet_key = "model.diffusion_" + key[len("control_"):]
+      if unet_key not in ctrl_unet_sd:               # zero conv
+        continue
+      ctrl_unet_sd[unet_key] = ctrl_sd[key]
+
+  unet_config = model_util.create_unet_diffusers_config(False)
+  ctrl_unet_sd_du = model_util.convert_ldm_unet_checkpoint(False, ctrl_unet_sd, unet_config)
+
+  # load weights to U-Net
+  ctrl_unet = UNet2DConditionModel(**unet_config)
+  info = ctrl_unet.load_state_dict(ctrl_unet_sd_du)
+  print("loading control u-net:", info)
+
+  # LoRAに対応する部分のU-Netの重みを読み込む #################################
+
+  org_unet_sd_du = org_unet.state_dict()
+
+  diffs = {}
+  for (org_name, org_module), (ctrl_name, ctrl_module) in zip(org_unet.named_modules(), ctrl_unet.named_modules()):
+    if org_module.__class__.__name__ != "Linear" and org_module.__class__.__name__ != "Conv2d":
+      continue
+    assert org_name == ctrl_name
+
+    lora_name = control_net_lora.ControlLoRANetwork.LORA_PREFIX_UNET + '.' + org_name  # + '.' + child_name
+    lora_name = lora_name.replace('.', '_')
+
+    diff = ctrl_module.weight - org_module.weight
+    diff = diff.float()
+
+    if torch.max(torch.abs(diff)) < 1e-5:
+      # print(f"weights are same: {lora_name}")
+      continue
+    print(lora_name)
+
+    if args.device:
+      diff = diff.to(args.device)
+
+    diffs[lora_name] = diff
+
+  # make LoRA with svd
+  print("calculating by svd")
+  rank = args.dim
+  ctrl_lora_sd = {}
+  with torch.no_grad():
+    for lora_name, mat in tqdm(list(diffs.items())):
+      conv2d = (len(mat.size()) == 4)
+      kernel_size = None if not conv2d else mat.size()[2:]
+
+      if not conv2d or kernel_size == (1, 1):
+        if conv2d:
+          mat = mat.squeeze()
+
+        U, S, Vh = torch.linalg.svd(mat)
+
+        U = U[:, :rank]
+        S = S[:rank]
+        U = U @ torch.diag(S)
+
+        Vh = Vh[:rank, :]
+
+        dist = torch.cat([U.flatten(), Vh.flatten()])
+        hi_val = torch.quantile(dist, CLAMP_QUANTILE)
+        low_val = -hi_val
+
+        U = U.clamp(low_val, hi_val)
+        Vh = Vh.clamp(low_val, hi_val)
+
+        if conv2d:
+          U = U.unsqueeze(2).unsqueeze(3)
+          Vh = Vh.unsqueeze(2).unsqueeze(3)
+      else:
+        # conv2d kernel != (1,1)
+        in_channels = mat.size()[1]
+        current_rank = min(rank, in_channels, mat.size()[0])
+        if current_rank != rank:
+          print(f"channels of conv2d is too small. rank is changed to {current_rank} @ {lora_name}: {mat.size()}")
+
+        mat = mat.flatten(start_dim=1)
+
+        U, S, Vh = torch.linalg.svd(mat)
+
+        U = U[:, :current_rank]
+        S = S[:current_rank]
+        U = U @ torch.diag(S)
+
+        Vh = Vh[:current_rank, :]
+
+        dist = torch.cat([U.flatten(), Vh.flatten()])
+        hi_val = torch.quantile(dist, CLAMP_QUANTILE)
+        low_val = -hi_val
+
+        U = U.clamp(low_val, hi_val)
+        Vh = Vh.clamp(low_val, hi_val)
+
+        # U is (out_channels, rank) with 1x1 conv. So,
+        U = U.reshape(U.shape[0], U.shape[1], 1, 1)
+        # V is (rank, in_channels * kernel_size1 * kernel_size2)
+        # now reshape:
+        Vh = Vh.reshape(Vh.shape[0], in_channels, *kernel_size)
+
+      ctrl_lora_sd[lora_name + ".lora_up.weight"] = U
+      ctrl_lora_sd[lora_name + ".lora_down.weight"] = Vh
+      ctrl_lora_sd[lora_name + ".alpha"] = torch.tensor(current_rank)
+
+  # create LoRA from sd
+  lora_network = control_net_lora.ControlLoRANetwork(org_unet, ctrl_lora_sd, 1.0)
+  lora_network.apply_to()
+
+  for key, value in ctrl_sd.items():
+    if 'zero_convs' in key or 'input_hint_block' in key or 'middle_block_out' in key:
+      ctrl_lora_sd[key] = value
+
+  info = lora_network.load_state_dict(ctrl_lora_sd)
+  print(f"loading control lora sd: {info}")
+
+  dir_name = os.path.dirname(args.save_to)
+  if dir_name and not os.path.exists(dir_name):
+    os.makedirs(dir_name, exist_ok=True)
+
+  # # minimum metadata
+  # metadata = {"ss_network_dim": str(args.dim), "ss_network_alpha": str(args.dim)}
+
+  # lora_network.save_weights(args.save_to, save_dtype, metadata)
+  save_file(ctrl_lora_sd, args.save_to)
+  print(f"LoRA weights are saved to: {args.save_to}")
+
+
+if __name__ == '__main__':
+  parser = argparse.ArgumentParser()
+  parser.add_argument("--v2", action='store_true',
+                      help='load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む')
+  parser.add_argument("--save_precision", type=str, default=None,
+                      choices=[None, "float", "fp16", "bf16"], help="precision in saving, same to merging if omitted / 保存時に精度を変更して保存する、省略時はfloat")
+  parser.add_argument("--model_org", type=str, default=None,
+                      help="Stable Diffusion original model: ckpt or safetensors file / 元モデル、ckptまたはsafetensors")
+  parser.add_argument("--model_tuned", type=str, default=None,
+                      help="Stable Diffusion tuned model, LoRA is difference of `original to tuned`: ckpt or safetensors file / 派生モデル（生成されるLoRAは元→派生の差分になります）、ckptまたはsafetensors")
+  parser.add_argument("--save_to", type=str, default=None,
+                      help="destination file name: ckpt or safetensors file / 保存先のファイル名、ckptまたはsafetensors")
+  parser.add_argument("--dim", type=int, default=4, help="dimension (rank) of LoRA (default 4) / LoRAの次元数（rank）（デフォルト4）")
+  parser.add_argument("--device", type=str, default=None, help="device to use, cuda for GPU / 計算を行うデバイス、cuda でGPUを使う")
+
+  args = parser.parse_args()
+  svd(args)