You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+24-2Lines changed: 24 additions & 2 deletions
Original file line number
Diff line number
Diff line change
@@ -154,7 +154,18 @@ The majority of scripts is licensed under ASL 2.0 (including codes from Diffuser
154
154
-`--fused_optimizer_groups` cannot be used with `--fused_backward_pass`. When using AdaFactor, the memory usage is slightly larger than with Fused optimizer. PyTorch 2.1 or later is required.
155
155
- Mechanism: While Fused optimizer performs backward/step for individual parameters within the optimizer, optimizer groups reduce memory usage by grouping parameters and creating multiple optimizers to perform backward/step for each group. Fused optimizer requires implementation on the optimizer side, while optimizer groups are implemented only on the training script side.
156
156
157
-
- Fixed some bugs when using DeepSpeed. Related [#1247]
157
+
- LoRA+ is supported. PR [#1233](https://github.com/kohya-ss/sd-scripts/pull/1233) Thanks to rockerBOO!
158
+
- LoRA+ is a method to improve training speed by increasing the learning rate of the UP side (LoRA-B) of LoRA. Specify the multiple. The original paper recommends 16, but adjust as needed. Please see the PR for details.
159
+
- Specify `loraplus_lr_ratio` with `--network_args`. Example: `--network_args "loraplus_lr_ratio=16"`
160
+
-`loraplus_unet_lr_ratio` and `loraplus_lr_ratio` can be specified separately for U-Net and Text Encoder.
161
+
- Example: `--network_args "loraplus_unet_lr_ratio=16" "loraplus_text_encoder_lr_ratio=4"` or `--network_args "loraplus_lr_ratio=16" "loraplus_text_encoder_lr_ratio=4"` etc.
162
+
-`network_module``networks.lora` and `networks.dylora` are available.
163
+
164
+
- LoRA training in SDXL now supports block-wise learning rates and block-wise dim (rank). PR [#1331](https://github.com/kohya-ss/sd-scripts/pull/1331)
165
+
- Specify the learning rate and dim (rank) for each block.
166
+
- See [Block-wise learning rates in LoRA](./docs/train_network_README-ja.md#階層別学習率) for details (Japanese only).
167
+
168
+
- Fixed some bugs when using DeepSpeed. Related [#1247](https://github.com/kohya-ss/sd-scripts/pull/1247)
0 commit comments