Merge branch 'dev' into sd3

README.md

@@ -461,6 +461,7 @@ The majority of scripts is licensed under ASL 2.0 (including codes from Diffuser
 
 ### Working in progress
 
+- `--v_parameterization` is available in `sdxl_train.py`. The results are unpredictable, so use with caution. PR [#1505](https://github.com/kohya-ss/sd-scripts/pull/1505) Thanks to liesened!
 - Fused optimizer is available for SDXL training. PR [#1259](https://github.com/kohya-ss/sd-scripts/pull/1259) Thanks to 2kpr!
   - The memory usage during training is significantly reduced by integrating the optimizer's backward pass with step. The training results are the same as before, but if you have plenty of memory, the speed will be slower.
   - Specify the `--fused_backward_pass` option in `sdxl_train.py`. At this time, only Adafactor is supported. Gradient accumulation is not available.

sdxl_train.py

@@ -700,6 +700,10 @@ def train(args):
                 with accelerator.autocast():
                     noise_pred = unet(noisy_latents, timesteps, text_embedding, vector_embedding)
 
+                if args.v_parameterization:
+                    # v-parameterization training
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
                     target = noise
 
                 if (
@@ -718,7 +722,7 @@ def train(args):
                     loss = loss.mean([1, 2, 3])
 
                     if args.min_snr_gamma:
-                        loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma)
+                        loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
                     if args.scale_v_pred_loss_like_noise_pred:
                         loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                     if args.v_pred_like_loss: