Revert unet config, add option to convert script

library/model_util.py

@@ -22,7 +22,7 @@ UNET_PARAMS_OUT_CHANNELS = 4
 UNET_PARAMS_NUM_RES_BLOCKS = 2
 UNET_PARAMS_CONTEXT_DIM = 768
 UNET_PARAMS_NUM_HEADS = 8
-UNET_PARAMS_USE_LINEAR_PROJECTION = False
+# UNET_PARAMS_USE_LINEAR_PROJECTION = False
 
 VAE_PARAMS_Z_CHANNELS = 4
 VAE_PARAMS_RESOLUTION = 256
@@ -35,7 +35,7 @@ VAE_PARAMS_NUM_RES_BLOCKS = 2
 # V2
 V2_UNET_PARAMS_ATTENTION_HEAD_DIM = [5, 10, 20, 20]
 V2_UNET_PARAMS_CONTEXT_DIM = 1024
-V2_UNET_PARAMS_USE_LINEAR_PROJECTION = True
+# V2_UNET_PARAMS_USE_LINEAR_PROJECTION = True
 
 # Diffusersの設定を読み込むための参照モデル
 DIFFUSERS_REF_MODEL_ID_V1 = "runwayml/stable-diffusion-v1-5"
@@ -209,13 +209,13 @@ def conv_attn_to_linear(checkpoint):
                 checkpoint[key] = checkpoint[key][:, :, 0]
 
 
-# def linear_transformer_to_conv(checkpoint):
+def linear_transformer_to_conv(checkpoint):
-#     keys = list(checkpoint.keys())
+    keys = list(checkpoint.keys())
-#     tf_keys = ["proj_in.weight", "proj_out.weight"]
+    tf_keys = ["proj_in.weight", "proj_out.weight"]
-#     for key in keys:
+    for key in keys:
-#         if ".".join(key.split(".")[-2:]) in tf_keys:
+        if ".".join(key.split(".")[-2:]) in tf_keys:
-#             if checkpoint[key].ndim == 2:
+            if checkpoint[key].ndim == 2:
-#                 checkpoint[key] = checkpoint[key].unsqueeze(2).unsqueeze(2)
+                checkpoint[key] = checkpoint[key].unsqueeze(2).unsqueeze(2)
 
 
 def convert_ldm_unet_checkpoint(v2, checkpoint, config):
@@ -359,9 +359,10 @@ def convert_ldm_unet_checkpoint(v2, checkpoint, config):
 
                 new_checkpoint[new_path] = unet_state_dict[old_path]
 
-    # SDのv2では1*1のconv2dがlinearに変わっているが、Diffusers側も同じなので、変換不要
+    # SDのv2では1*1のconv2dがlinearに変わっている
-    # if v2:
+    # 誤って Diffusers 側を conv2d のままにしてしまったので、変換必要
-    #     linear_transformer_to_conv(new_checkpoint)
+    if v2 and not config.get('use_linear_projection', False):
+        linear_transformer_to_conv(new_checkpoint)
 
     return new_checkpoint
 
@@ -470,7 +471,7 @@ def convert_ldm_vae_checkpoint(checkpoint, config):
     return new_checkpoint
 
 
-def create_unet_diffusers_config(v2):
+def create_unet_diffusers_config(v2, use_linear_projection_in_v2=False):
     """
     Creates a config for the diffusers based on the config of the LDM model.
     """
@@ -502,8 +503,10 @@ def create_unet_diffusers_config(v2):
         layers_per_block=UNET_PARAMS_NUM_RES_BLOCKS,
         cross_attention_dim=UNET_PARAMS_CONTEXT_DIM if not v2 else V2_UNET_PARAMS_CONTEXT_DIM,
         attention_head_dim=UNET_PARAMS_NUM_HEADS if not v2 else V2_UNET_PARAMS_ATTENTION_HEAD_DIM,
-        use_linear_projection=UNET_PARAMS_USE_LINEAR_PROJECTION if not v2 else V2_UNET_PARAMS_USE_LINEAR_PROJECTION,
+        # use_linear_projection=UNET_PARAMS_USE_LINEAR_PROJECTION if not v2 else V2_UNET_PARAMS_USE_LINEAR_PROJECTION,
     )
+    if v2 and use_linear_projection_in_v2:
+        config["use_linear_projection"] = True
 
     return config
 
@@ -849,11 +852,11 @@ def load_checkpoint_with_text_encoder_conversion(ckpt_path, device="cpu"):
 
 
 # TODO dtype指定の動作が怪しいので確認する text_encoderを指定形式で作れるか未確認
-def load_models_from_stable_diffusion_checkpoint(v2, ckpt_path, device="cpu", dtype=None):
+def load_models_from_stable_diffusion_checkpoint(v2, ckpt_path, device="cpu", dtype=None, unet_use_linear_projection_in_v2=False):
     _, state_dict = load_checkpoint_with_text_encoder_conversion(ckpt_path, device)
 
     # Convert the UNet2DConditionModel model.
-    unet_config = create_unet_diffusers_config(v2)
+    unet_config = create_unet_diffusers_config(v2, unet_use_linear_projection_in_v2)
     converted_unet_checkpoint = convert_ldm_unet_checkpoint(v2, state_dict, unet_config)
 
     unet = UNet2DConditionModel(**unet_config).to(device)

tools/convert_diffusers20_original_sd.py

@@ -34,7 +34,7 @@ def convert(args):
 
     if is_load_ckpt:
         v2_model = args.v2
-        text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(v2_model, args.model_to_load)
+        text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(v2_model, args.model_to_load, unet_use_linear_projection_in_v2=args.unet_use_linear_projection)
     else:
         pipe = StableDiffusionPipeline.from_pretrained(
             args.model_to_load, torch_dtype=load_dtype, tokenizer=None, safety_checker=None
@@ -76,6 +76,9 @@ def setup_parser() -> argparse.ArgumentParser:
     parser.add_argument(
         "--v2", action="store_true", help="load v2.0 model (v1 or v2 is required to load checkpoint) / 2.0のモデルを読み込む"
     )
+    parser.add_argument(
+        "--unet_use_linear_projection", action="store_true", help="When saving v2 model as Diffusers, set U-Net config to `use_linear_projection=true` (to match stabilityai's model) / Diffusers形式でv2モデルを保存するときにU-Netの設定を`use_linear_projection=true`にする（stabilityaiのモデルと合わせる）"
+    )
     parser.add_argument(
         "--fp16",
         action="store_true",