Update extract LoRA and add sdpa

README.md

@@ -494,4 +494,6 @@ If you come across a `FileNotFoundError`, it is likely due to an installation is
   - Add missing LR number of cycles and LR power to Dreambooth and TI scripts
   - Fix issue with conv_block_dims and conv_block_alphas
   - Fix 0 noise offset issue
-  - Implement Stop training button on LoRA
+  - Implement Stop training button on LoRA
+  - Add support to extract LoRA from SDXL finetuned models
+  - Add support for PagedAdamW8bit and PagedLion8bit optimizer. Those require a new version of bitsandbytes so success on some systems may vary. I had to uninstall all my nvidia drivers and othe cuda toolkit install, delete all cuda variable references and re-install cuda toolkit v11.8.0 to get things to work... so not super easy.

examples/finetune_latent.ps1

@@ -0,0 +1,11 @@
+# Command 1: merge_captions_to_metadata.py
+$captionExtension = "--caption_extension=.txt"
+$sourceDir1 = "d:\test\1_1960-1969"
+$targetFile1 = "d:\test\1_1960-1969/meta_cap.json"
+
+# Command 2: prepare_buckets_latents.py
+$targetLatentFile = "d:\test\1_1960-1969/meta_lat.json"
+$modelFile = "E:\models\sdxl\sd_xl_base_0.9.safetensors"
+
+./venv/Scripts/python.exe finetune/merge_captions_to_metadata.py $captionExtension $sourceDir1 $targetFile1 --full_path
+./venv/Scripts/python.exe finetune/prepare_buckets_latents.py $sourceDir1 $targetFile1 $targetLatentFile $modelFile --batch_size=4 --max_resolution=1024,1024 --min_bucket_reso=64 --max_bucket_reso=2048 --mixed_precision=bf16 --full_path

finetune_gui.py

@@ -89,6 +89,7 @@ def save_configuration(
     caption_extension,
     # use_8bit_adam,
     xformers,
+    sdpa,
     clip_skip,
     save_state,
     resume,
@@ -209,6 +210,7 @@ def open_configuration(
     caption_extension,
     # use_8bit_adam,
     xformers,
+    sdpa,
     clip_skip,
     save_state,
     resume,
@@ -326,6 +328,7 @@ def train_model(
     caption_extension,
     # use_8bit_adam,
     xformers,
+    sdpa,
     clip_skip,
     save_state,
     resume,
@@ -575,6 +578,7 @@ def train_model(
         gradient_checkpointing=gradient_checkpointing,
         full_fp16=full_fp16,
         xformers=xformers,
+        spda=sdpa,
         # use_8bit_adam=use_8bit_adam,
         keep_tokens=keep_tokens,
         persistent_data_loader_workers=persistent_data_loader_workers,
@@ -866,6 +870,7 @@ def finetune_tab(headless=False):
             source_model.save_model_as,
             basic_training.caption_extension,
             advanced_training.xformers,
+            advanced_training.sdpa,
             advanced_training.clip_skip,
             advanced_training.save_state,
             advanced_training.resume,

library/class_advanced_training.py

@@ -101,13 +101,14 @@ def noise_offset_type_change(noise_offset_type):
             # use_8bit_adam = gr.Checkbox(
             #     label='Use 8bit adam', value=False, visible=False
             # )
-            self.xformers = gr.Checkbox(label='Use xformers', value=True)
+            self.xformers = gr.Checkbox(label='Use xformers', value=True, info='Use xformers for CrossAttention')
             self.color_aug = gr.Checkbox(label='Color augmentation', value=False)
             self.flip_aug = gr.Checkbox(label='Flip augmentation', value=False)
             self.min_snr_gamma = gr.Slider(
                 label='Min SNR gamma', value=0, minimum=0, maximum=20, step=1
             )
         with gr.Row():
+            self.sdpa = gr.Checkbox(label='Use sdpa', value=False, info='Use sdpa for CrossAttention')
             self.bucket_no_upscale = gr.Checkbox(
                 label="Don't upscale bucket resolution", value=True
             )

library/common_gui.py

@@ -735,6 +735,10 @@ def run_cmd_advanced_training(**kwargs):
     xformers = kwargs.get('xformers')
     if xformers:
         run_cmd += ' --xformers'
+
+    sdpa = kwargs.get('sdpa')
+    if sdpa:
+        run_cmd += ' --sdpa'
 
     persistent_data_loader_workers = kwargs.get('persistent_data_loader_workers')
     if persistent_data_loader_workers:
@@ -856,3 +860,18 @@ def check_duplicate_filenames(folder_path, image_extension = ['.gif', '.png', '.
                         print(f"Current file: {full_path}")
                 else:
                     filenames[filename] = full_path
+
+def is_file_writable(file_path):
+    if not os.path.exists(file_path):
+        # print(f"File '{file_path}' does not exist.")
+        return True
+
+    try:
+        log.warning(f"File '{file_path}' already exist... it will be overwritten...")
+        # Check if the file can be opened in write mode (which implies it's not open by another process)
+        with open(file_path, 'a'):
+            pass
+        return True
+    except IOError:
+        log.warning(f"File '{file_path}' can't be written to...")
+        return False

library/extract_lora_gui.py

@@ -4,8 +4,8 @@
 import os
 from .common_gui import (
     get_saveasfilename_path,
-    get_any_file_path,
     get_file_path,
+    is_file_writable
 )
 
 from library.custom_logging import setup_logging
@@ -27,25 +27,31 @@ def extract_lora(
     save_precision,
     dim,
     v2,
+    sdxl,
     conv_dim,
+    clamp_quantile,
+    min_diff,
     device,
 ):
     # Check for caption_text_input
     if model_tuned == '':
-        msgbox('Invalid finetuned model file')
+        log.info('Invalid finetuned model file')
         return
 
     if model_org == '':
-        msgbox('Invalid base model file')
+        log.info('Invalid base model file')
         return
 
     # Check if source model exist
     if not os.path.isfile(model_tuned):
-        msgbox('The provided finetuned model is not a file')
+        log.info('The provided finetuned model is not a file')
         return
 
     if not os.path.isfile(model_org):
-        msgbox('The provided base model is not a file')
+        log.info('The provided base model is not a file')
+        return
+
+    if not is_file_writable(save_to):
         return
 
     run_cmd = (
@@ -61,6 +67,10 @@ def extract_lora(
         run_cmd += f' --conv_dim {conv_dim}'
     if v2:
         run_cmd += f' --v2'
+    if sdxl:
+        run_cmd += f' --sdxl'
+    run_cmd += f' --clamp_quantile {clamp_quantile}'
+    run_cmd += f' --min_diff {min_diff}'
 
     log.info(run_cmd)
 
@@ -160,7 +170,19 @@ def gradio_extract_lora_tab(headless=False):
                 step=1,
                 interactive=True,
             )
+            clamp_quantile = gr.Number(
+                label='Clamp Quantile',
+                value=1,
+                interactive=True,
+            )
+            min_diff = gr.Number(
+                label='Minimum difference',
+                value=0.01,
+                interactive=True,
+            )
+        with gr.Row():
             v2 = gr.Checkbox(label='v2', value=False, interactive=True)
+            sdxl = gr.Checkbox(label='SDXL', value=False, interactive=True)
             device = gr.Dropdown(
                 label='Device',
                 choices=[
@@ -182,7 +204,10 @@ def gradio_extract_lora_tab(headless=False):
                 save_precision,
                 dim,
                 v2,
+                sdxl,
                 conv_dim,
+                clamp_quantile,
+                min_diff,
                 device,
             ],
             show_progress=False,

networks/extract_lora_from_models.py

@@ -12,10 +12,8 @@
 import library.sdxl_model_util as sdxl_model_util
 import lora
 
-
-CLAMP_QUANTILE = 0.99
-MIN_DIFF = 1e-4
-
+# CLAMP_QUANTILE = 1
+# MIN_DIFF = 1e-2
 
 def save_to_file(file_name, model, state_dict, dtype):
     if dtype is not None:
@@ -91,9 +89,9 @@ def str_to_dtype(p):
         diff = module_t.weight - module_o.weight
 
         # Text Encoder might be same
-        if not text_encoder_different and torch.max(torch.abs(diff)) > MIN_DIFF:
+        if not text_encoder_different and torch.max(torch.abs(diff)) > args.min_diff:
             text_encoder_different = True
-            print(f"Text encoder is different. {torch.max(torch.abs(diff))} > {MIN_DIFF}")
+            print(f"Text encoder is different. {torch.max(torch.abs(diff))} > {args.min_diff}")
 
         diff = diff.float()
         diffs[lora_name] = diff
@@ -149,7 +147,7 @@ def str_to_dtype(p):
             Vh = Vh[:rank, :]
 
             dist = torch.cat([U.flatten(), Vh.flatten()])
-            hi_val = torch.quantile(dist, CLAMP_QUANTILE)
+            hi_val = torch.quantile(dist, args.clamp_quantile)
             low_val = -hi_val
 
             U = U.clamp(low_val, hi_val)
@@ -243,6 +241,18 @@ def setup_parser() -> argparse.ArgumentParser:
         help="dimension (rank) of LoRA for Conv2d-3x3 (default None, disabled) / LoRAのConv2d-3x3の次元数（rank）（デフォルトNone、適用なし）",
     )
     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")
+    parser.add_argument(
+        "--clamp_quantile",
+        type=float,
+        default=1,
+        help="Quantile clamping value, float, (0-1). Defailt = 1",
+    )
+    parser.add_argument(
+        "--min_diff",
+        type=float,
+        default=1,
+        help="Minimum difference betwen finetuned model and base to consider them different enough to extract, float, (0-1). Defailt = 0.01",
+    )
 
     return parser