Merge branch 'sd3' of https://github.com/kohya-ss/sd-scripts into sd3

.github/workflows/tests.yml

@@ -0,0 +1,42 @@
+
+name: Python package
+
+on: [push]
+
+jobs:
+  build:
+
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-latest]
+        python-version: ["3.10"]
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.x'
+
+      - name: Install dependencies
+        run: python -m pip install --upgrade pip setuptools wheel
+
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.x'
+          cache: 'pip' # caching pip dependencies
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install dadaptation==3.2 torch==2.4.0 torchvision==0.19.0 accelerate==0.33.0
+          pip install -r requirements.txt
+
+      - name: Test with pytest
+        run: |
+          pip install pytest
+          pytest
+

.github/workflows/typos.yml

@@ -1,9 +1,11 @@
 ---
-# yamllint disable rule:line-length
 name: Typos
 
-on:  # yamllint disable-line rule:truthy
+on: 
   push:
+    branches:
+      - main
+      - dev
   pull_request:
     types:
       - opened
@@ -18,4 +20,4 @@ jobs:
       - uses: actions/checkout@v4
 
       - name: typos-action
-        uses: crate-ci/typos@v1.24.3
+        uses: crate-ci/typos@v1.28.1

fine_tune.py

@@ -380,9 +380,7 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
 
                 # Sample noise, sample a random timestep for each image, and add noise to the latents,
                 # with noise offset and/or multires noise if specified
-                noise, noisy_latents, timesteps, huber_c = train_util.get_noise_noisy_latents_and_timesteps(
-                    args, noise_scheduler, latents
-                )
+                noise, noisy_latents, timesteps = train_util.get_noise_noisy_latents_and_timesteps(args, noise_scheduler, latents)
 
                 # Predict the noise residual
                 with accelerator.autocast():
@@ -394,11 +392,10 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
                 else:
                     target = noise
 
+                huber_c = train_util.get_huber_threshold_if_needed(args, timesteps, noise_scheduler)
                 if args.min_snr_gamma or args.scale_v_pred_loss_like_noise_pred or args.debiased_estimation_loss:
                     # do not mean over batch dimension for snr weight or scale v-pred loss
-                    loss = train_util.conditional_loss(
-                        noise_pred.float(), target.float(), reduction="none", loss_type=args.loss_type, huber_c=huber_c
-                    )
+                    loss = train_util.conditional_loss(noise_pred.float(), target.float(), args.loss_type, "none", huber_c)
                     loss = loss.mean([1, 2, 3])
 
                     if args.min_snr_gamma:
@@ -410,9 +407,7 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
 
                     loss = loss.mean()  # mean over batch dimension
                 else:
-                    loss = train_util.conditional_loss(
-                        noise_pred.float(), target.float(), reduction="mean", loss_type=args.loss_type, huber_c=huber_c
-                    )
+                    loss = train_util.conditional_loss(noise_pred.float(), target.float(), args.loss_type, "mean", huber_c)
 
                 accelerator.backward(loss)
                 if accelerator.sync_gradients and args.max_grad_norm != 0.0: