Add example config for qwen image edit

config/examples/train_lora_qwen_image_24gb.yaml

@@ -39,7 +39,7 @@ config:
       train:
         batch_size: 1
         # caching text embeddings is required for 24GB
-        cache_text_embeddings: false
+        cache_text_embeddings: true
         
         steps: 2000  # total number of steps to train 500 - 4000 is a good range
         gradient_accumulation: 1

config/examples/train_lora_qwen_image_edit_32gb.yaml

@@ -0,0 +1,102 @@
+---
+job: extension
+config:
+  # this name will be the folder and filename name
+  name: "my_first_qwen_image_edit_lora_v1"
+  process:
+    - type: 'sd_trainer'
+      # root folder to save training sessions/samples/weights
+      training_folder: "output"
+      # uncomment to see performance stats in the terminal every N steps
+#      performance_log_every: 1000
+      device: cuda:0
+      # if a trigger word is specified, it will be added to captions of training data if it does not already exist
+      # alternatively, in your captions you can add [trigger] and it will be replaced with the trigger word
+      # Trigger words will not work when caching text embeddings
+#      trigger_word: "p3r5on"
+      network:
+        type: "lora"
+        linear: 16
+        linear_alpha: 16
+      save:
+        dtype: float16 # precision to save
+        save_every: 250 # save every this many steps
+        max_step_saves_to_keep: 4 # how many intermittent saves to keep
+      datasets:
+        # datasets are a folder of images. captions need to be txt files with the same name as the image
+        # for instance image2.jpg and image2.txt. Only jpg, jpeg, and png are supported currently
+        # images will automatically be resized and bucketed into the resolution specified
+        # on windows, escape back slashes with another backslash so
+        # "C:\\path\\to\\images\\folder"
+        - folder_path: "/path/to/images/folder"
+          control_path: "/path/to/control/images/folder"
+          caption_ext: "txt"
+          # default_caption: "a person" # if caching text embeddings, if you don't have captions, this will get cached
+          caption_dropout_rate: 0.05  # will drop out the caption 5% of time
+          resolution: [ 512, 768, 1024 ]  # qwen image enjoys multiple resolutions
+      train:
+        batch_size: 1
+        # caching text embeddings is required for 32GB
+        cache_text_embeddings: true
+        
+        steps: 3000  # total number of steps to train 500 - 4000 is a good range
+        gradient_accumulation: 1
+        timestep_type: "weighted"
+        train_unet: true
+        train_text_encoder: false  # probably won't work with qwen image
+        gradient_checkpointing: true  # need the on unless you have a ton of vram
+        noise_scheduler: "flowmatch" # for training only
+        optimizer: "adamw8bit"
+        lr: 1e-4
+        # uncomment this to skip the pre training sample
+#        skip_first_sample: true
+        # uncomment to completely disable sampling
+#        disable_sampling: true
+        dtype: bf16
+      model:
+        # huggingface model name or path
+        name_or_path: "Qwen/Qwen-Image-Edit"
+        arch: "qwen_image_edit"
+        quantize: true
+        # qtype_te: "qfloat8" Default float8 qquantization
+        # to use the ARA use the | pipe to point to hf path, or a local path if you have one.
+        # 3bit is required for 32GB
+        qtype: "uint3|qwen_image_edit_torchao_uint3.safetensors"
+        quantize_te: true
+        qtype_te: "qfloat8"
+        low_vram: true
+      sample:
+        sampler: "flowmatch" # must match train.noise_scheduler
+        sample_every: 250 # sample every this many steps
+        width: 1024
+        height: 1024
+        samples:
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+          - prompt: "do the thing to it"
+            ctrl_img: "/path/to/control/image.jpg"
+        neg: ""
+        seed: 42
+        walk_seed: true
+        guidance_scale: 3
+        sample_steps: 25
+# you can add any additional meta info here. [name] is replaced with config name at top
+meta:
+  name: "[name]"
+  version: '1.0'

toolkit/dataloader_mixins.py

@@ -1865,7 +1865,9 @@ class TextEmbeddingCachingMixin:
                         self.sd.set_device_state_preset('cache_text_encoder')
                         did_move = True
                         
-                    if file_item.encode_control_in_text_embeddings and file_item.control_path is not None:
+                    if file_item.encode_control_in_text_embeddings:
+                        if file_item.control_path is None:
+                            raise Exception(f"Could not find a control image for {file_item.path} which is needed for this model")
                         # load the control image and feed it into the text encoder
                         ctrl_img = Image.open(file_item.control_path).convert("RGB")
                         # convert to 0 to 1 tensor

version.py

@@ -1 +1 @@
-VERSION = "0.5.4"
+VERSION = "0.5.5"