Merge ctk_copy: LoKr support, Scaled OFT, ETA fix, row counter fix; keep ctk_abstraction for UI conflicts

- Apply LoKr and Scaled OFT additions to BaseLoraTabView/LoraTabController
- Apply ETA session-tracking fix to TrainUIController
- Apply local row counter fix to BaseTrainingTabView.__create_text_encoder_frame
- Keep both supported_caption_extensions and json_path_modifier in path_util

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: dxqb

.pre-commit-config.yaml

@@ -14,7 +14,7 @@ repos:
     - id: check-yaml
 
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.15.12
+  rev: v0.15.14
   hooks:
     # Run the Ruff linter, but not the formatter.
     - id: ruff

modules/cloud/BaseCloud.py

@@ -2,6 +2,7 @@
 from abc import ABCMeta, abstractmethod
 from pathlib import Path
 
+from modules.util import path_util
 from modules.util.callbacks.TrainCallbacks import TrainCallbacks
 from modules.util.commands.TrainCommands import TrainCommands
 from modules.util.config.CloudConfig import CloudConfig
@@ -54,6 +55,7 @@ def upload_config(self,commands : TrainCommands=None):
             if hasattr(add_embedding,"local_model_name"):
                 self.file_sync.sync_up(local=Path(add_embedding.local_model_name),remote=Path(add_embedding.model_name))
 
+        concept_extensions = path_util.supported_image_extensions() | path_util.supported_video_extensions() | path_util.supported_caption_extensions()
         for concept in self.config.concepts:
             print(f"uploading concept {concept.name}...")
             if commands and commands.get_stop_command():
@@ -63,11 +65,16 @@ def upload_config(self,commands : TrainCommands=None):
                 self.file_sync.sync_up_dir(
                     local=Path(concept.local_path),
                     remote=Path(concept.path),
-                    recursive=concept.include_subdirectories)
+                    recursive=concept.include_subdirectories,
+                    skip_hidden=True,
+                    allowed_extensions=concept_extensions)
 
             if hasattr(concept.text,"local_prompt_path"):
                 self.file_sync.sync_up_file(local=Path(concept.text.local_prompt_path),remote=Path(concept.text.prompt_path))
 
+            if hasattr(concept.text,"local_tag_dropout_special_tags"):
+                self.file_sync.sync_up_file(local=Path(concept.text.local_tag_dropout_special_tags),remote=Path(concept.text.tag_dropout_special_tags))
+
     @staticmethod
     def _filter_download(config : CloudConfig,path : Path):
         if 'samples' in path.parts:

modules/cloud/BaseFileSync.py

@@ -25,7 +25,8 @@ def sync_up_file(self,local : Path,remote : Path):
         pass
 
     @abstractmethod
-    def sync_up_dir(self,local : Path,remote: Path,recursive: bool):
+    def sync_up_dir(self,local : Path,remote: Path,recursive: bool,sync_info=None,
+                    skip_hidden: bool=False, allowed_extensions: set[str] | None=None):
         pass
 
     @abstractmethod

modules/cloud/BaseSSHFileSync.py

@@ -46,19 +46,26 @@ def sync_up_file(self,local : Path,remote : Path):
         self.upload_file(local_file=local,remote_file=remote)
 
 
-    def sync_up_dir(self,local : Path,remote: Path,recursive: bool,sync_info=None):
+    def sync_up_dir(self,local : Path,remote: Path,recursive: bool,sync_info=None,
+                    skip_hidden: bool=False, allowed_extensions: set[str] | None=None):
         if sync_info is None:
             sync_info=self.__get_sync_info(remote)
         self.sync_connection.open()
         self.sync_connection.run(f'mkdir -p {shlex.quote(remote.as_posix())}',in_stream=False)
         files=[]
         for local_entry in local.iterdir():
             if local_entry.is_file():
+                if allowed_extensions is not None and local_entry.suffix.lower() not in allowed_extensions:
+                    continue
                 remote_entry=remote/local_entry.name
                 if self.__needs_upload(local=local_entry,remote=remote_entry,sync_info=sync_info):
                     files.append(local_entry)
             elif recursive and local_entry.is_dir():
-                self.sync_up_dir(local=local_entry,remote=remote/local_entry.name,recursive=True,sync_info=sync_info)
+                if skip_hidden and local_entry.name.startswith('.'):
+                    continue
+                self.sync_up_dir(local=local_entry,remote=remote/local_entry.name,recursive=True,
+                                 sync_info=sync_info,skip_hidden=skip_hidden,
+                                 allowed_extensions=allowed_extensions)
 
         self.upload_files(local_files=files,remote_dir=remote)
 

modules/dataLoader/FluxBaseDataLoader.py

@@ -41,7 +41,7 @@ def _preparation_modules(self, config: TrainConfig, model: FluxModel):
         shuffle_mask_channels = ShuffleFluxFillMaskChannels(in_name='mask', out_name='latent_mask')
         encode_conditioning_image = EncodeVAE(in_name='conditioning_image', out_name='latent_conditioning_image_distribution', vae=model.vae, autocast_contexts=[model.autocast_context], dtype=model.train_dtype.torch_dtype())
         conditioning_image_sample = SampleVAEDistribution(in_name='latent_conditioning_image_distribution', out_name='latent_conditioning_image', mode='mean')
-        tokenize_prompt_1 = Tokenize(in_name='prompt_1', tokens_out_name='tokens_1', mask_out_name='tokens_mask_1', tokenizer=model.tokenizer_1, max_token_length=model.tokenizer_1.model_max_length)
+        tokenize_prompt_1 = Tokenize(in_name='prompt_1', tokens_out_name='tokens_1', mask_out_name='tokens_mask_1', tokenizer=model.tokenizer_1, max_token_length=77)
         tokenize_prompt_2 = Tokenize(in_name='prompt_2', tokens_out_name='tokens_2', mask_out_name='tokens_mask_2', tokenizer=model.tokenizer_2, max_token_length=config.text_encoder_2_sequence_length)
         encode_prompt_1 = EncodeClipText(in_name='tokens_1', tokens_attention_mask_in_name=None, hidden_state_out_name='text_encoder_1_hidden_state', pooled_out_name='text_encoder_1_pooled_state',
                                          add_layer_norm=False, text_encoder=model.text_encoder_1, hidden_state_output_index=-(2 + config.text_encoder_layer_skip), autocast_contexts=[model.autocast_context],

modules/model/ChromaModel.py

@@ -208,7 +208,7 @@ def encode_text(
         )
 
         # apply dropout FIXME not a real dropout
-        if text_encoder_dropout_probability is not None:
+        if text_encoder_dropout_probability is not None and text_encoder_dropout_probability > 0.0:
             dropout_text_encoder_mask = (torch.tensor(
                 [rand.random() > text_encoder_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()

modules/model/FluxModel.py

@@ -287,13 +287,13 @@ def encode_text(
         )
 
         # apply dropout
-        if text_encoder_1_dropout_probability is not None:
+        if text_encoder_1_dropout_probability is not None and text_encoder_1_dropout_probability > 0.0:
             dropout_text_encoder_1_mask = (torch.tensor(
                 [rand.random() > text_encoder_1_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()
             pooled_text_encoder_1_output = pooled_text_encoder_1_output * dropout_text_encoder_1_mask[:, None]
 
-        if text_encoder_2_dropout_probability is not None:
+        if text_encoder_2_dropout_probability is not None and text_encoder_2_dropout_probability > 0.0:
             dropout_text_encoder_2_mask = (torch.tensor(
                 [rand.random() > text_encoder_2_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()

modules/model/HiDreamModel.py

@@ -442,26 +442,26 @@ def encode_text(
         # )
 
         # apply dropout
-        if text_encoder_1_dropout_probability is not None:
+        if text_encoder_1_dropout_probability is not None and text_encoder_1_dropout_probability > 0.0:
             dropout_text_encoder_1_mask = (torch.tensor(
                 [rand.random() > text_encoder_1_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()
             pooled_text_encoder_1_output = pooled_text_encoder_1_output * dropout_text_encoder_1_mask[:, None]
 
-        if text_encoder_2_dropout_probability is not None:
+        if text_encoder_2_dropout_probability is not None and text_encoder_2_dropout_probability > 0.0:
             dropout_text_encoder_2_mask = (torch.tensor(
                 [rand.random() > text_encoder_2_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()
             pooled_text_encoder_2_output = pooled_text_encoder_2_output * dropout_text_encoder_2_mask[:, None]
 
-        if text_encoder_3_dropout_probability is not None:
+        if text_encoder_3_dropout_probability is not None and text_encoder_3_dropout_probability > 0.0:
             dropout_text_encoder_3_mask = (torch.tensor(
                 [rand.random() > text_encoder_3_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()
             text_encoder_3_output = text_encoder_3_output * dropout_text_encoder_3_mask[:, None, None]
 
         text_encoder_4_output = torch.stack(text_encoder_4_output, dim=0)
-        if text_encoder_4_dropout_probability is not None:
+        if text_encoder_4_dropout_probability is not None and text_encoder_4_dropout_probability > 0.0:
             dropout_text_encoder_4_mask = (torch.tensor(
                 [rand.random() > text_encoder_4_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()

modules/model/HunyuanVideoModel.py

@@ -300,13 +300,13 @@ def encode_text(
         )
 
         # apply dropout
-        if text_encoder_1_dropout_probability is not None:
+        if text_encoder_1_dropout_probability is not None and text_encoder_1_dropout_probability > 0.0:
             dropout_text_encoder_1_mask = (torch.tensor(
                 [rand.random() > text_encoder_1_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()
             text_encoder_1_output = text_encoder_1_output * dropout_text_encoder_1_mask[:, None, None]
 
-        if text_encoder_2_dropout_probability is not None:
+        if text_encoder_2_dropout_probability is not None and text_encoder_2_dropout_probability > 0.0:
             dropout_text_encoder_2_mask = (torch.tensor(
                 [rand.random() > text_encoder_2_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()

modules/model/PixArtAlphaModel.py

@@ -211,7 +211,7 @@ def encode_text(
         )
 
         # apply dropout
-        if text_encoder_dropout_probability is not None:
+        if text_encoder_dropout_probability is not None and text_encoder_dropout_probability > 0.0:
             dropout_text_encoder_mask = (torch.tensor(
                 [rand.random() > text_encoder_dropout_probability for _ in range(batch_size)],
                 device=train_device)).float()