Automated PR - 2026-07-07

packages/ltx-trainer/AGENTS.md

@@ -94,11 +94,29 @@ packages/ltx-trainer/
 │   ├── audio_inpainting_lora.yaml # Audio inpainting
 │   ├── a2a_ic_lora.yaml          # Audio-to-audio IC-LoRA
 │   ├── av2av_ic_lora.yaml        # AV2AV IC-LoRA
+│   ├── README.md                 # Configs index + training-modes table (see note below)
 │   └── accelerate/               # FSDP, DDP configs
 ├── tests/                        # Pytest tests
 └── docs/                         # Documentation
 ```
 
+### Configs Directory Documentation
+
+The `configs/` directory is documented in two places that both contain a training-modes table linking to the
+`.yaml` files:
+
+- **[`configs/README.md`](configs/README.md)** — a short index with a training-modes table linking to each config
+  file in the directory (relative links like `./t2v_lora.yaml`).
+- **[`docs/training-modes.md`](docs/training-modes.md)** — the full training-modes guide, whose "Quick Reference"
+  table and per-mode sections link to the configs (relative links like `../configs/t2v_lora.yaml`).
+
+> **⚠️ When changing the contents of `configs/`:**
+>
+> - If you **rename, move, or delete** a `.yaml` config, update the links in **both** `configs/README.md` and
+>   `docs/training-modes.md` so no link is broken.
+> - If you **add** a new `.yaml` config, add a corresponding row (or config link) to the tables in **both**
+>   `configs/README.md` and `docs/training-modes.md`, and add it to the directory tree above.
+
 ### Key Architectural Patterns
 
 **Model Loading:**

packages/ltx-trainer/configs/README.md

@@ -0,0 +1,27 @@
+# Training Configs
+
+Example training configurations for the LTX-2 trainer. Each file is a ready-to-run config for one training mode,
+expressed through the unified **flexible** strategy (`name: "flexible"`). Pick the one closest to your use case and
+adjust paths, dataset, and hyperparameters.
+
+> 📖 For more information about using each training mode, see [Training Modes Guide](../docs/training-modes.md).
+
+## Training Modes
+
+| Mode                  | Video     | Audio     | Conditions          | Config |
+|-----------------------|-----------|-----------|---------------------|--------|
+| **T2V**               | Generated | Generated | —                   | [`t2v_lora.yaml`](./t2v_lora.yaml), [`t2v_lora_low_vram.yaml`](./t2v_lora_low_vram.yaml) (low VRAM) |
+| **I2V**               | Generated | Generated | `first_frame`       | [`i2v_lora.yaml`](./i2v_lora.yaml) |
+| **Video Extension**   | Generated | Generated | `prefix`/`suffix`   | [`video_extend_lora.yaml`](./video_extend_lora.yaml) (forward), [`video_suffix_lora.yaml`](./video_suffix_lora.yaml) (backward) |
+| **V2V IC-LoRA**       | Generated | —         | `reference`         | [`v2v_ic_lora.yaml`](./v2v_ic_lora.yaml) |
+| **A2V**               | Generated | Frozen    | —                   | [`a2v_lora.yaml`](./a2v_lora.yaml) |
+| **V2A (Foley)**       | Frozen    | Generated | —                   | [`v2a_lora.yaml`](./v2a_lora.yaml) |
+| **Video Inpainting**  | Generated | —         | `mask`              | [`video_inpainting_lora.yaml`](./video_inpainting_lora.yaml) |
+| **Video Outpainting** | Generated | —         | `spatial_crop`      | [`video_outpainting_lora.yaml`](./video_outpainting_lora.yaml) |
+| **T2A**               | —         | Generated | —                   | [`t2a_lora.yaml`](./t2a_lora.yaml) |
+| **Audio Extension**   | —         | Generated | `prefix`/`suffix`   | [`audio_extend_lora.yaml`](./audio_extend_lora.yaml) (forward), [`audio_suffix_lora.yaml`](./audio_suffix_lora.yaml) (backward) |
+| **Audio Inpainting**  | —         | Generated | `mask`              | [`audio_inpainting_lora.yaml`](./audio_inpainting_lora.yaml) |
+| **A2A IC-LoRA**       | —         | Generated | `reference`         | [`a2a_ic_lora.yaml`](./a2a_ic_lora.yaml) |
+| **AV2AV IC-LoRA**     | Generated | Generated | `reference` (both)  | [`av2av_ic_lora.yaml`](./av2av_ic_lora.yaml) |
+
+The [`accelerate/`](./accelerate) directory holds the Accelerate launch configs (FSDP, DDP) for multi-GPU training.

packages/ltx-trainer/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ltx-trainer"
-version = "1.1.6"
+version = "1.1.7"
 description = "LTX-2 training, democratized."
 readme = "README.md"
 authors = [
@@ -54,7 +54,7 @@ build-backend = "hatchling.build"
 
 
 [tool.ruff]
-target-version = "1.1.6"
+target-version = "1.1.7"
 line-length = 120
 # Restrict isort first-party detection to src/ so stray dirs (e.g. wandb/ run output)
 # next to pyproject.toml don't get classified as first-party packages. See ruff#10519.

packages/ltx-trainer/src/ltx_trainer/validation_runner.py

@@ -774,7 +774,11 @@ class ValidationRunner:
         stg_guider = STGGuider(cfg.stg_scale)
 
         stg_perturbation_config = (
-            self._build_stg_perturbation_config(cfg.stg_blocks, cfg.stg_mode) if stg_guider.enabled() else None
+            self._build_stg_perturbation_config(
+                cfg.stg_blocks, cfg.stg_mode, transformer.num_blocks, device, next(transformer.parameters()).dtype
+            )
+            if stg_guider.enabled()
+            else None
         )
 
         x0_model = X0Model(transformer)
@@ -1072,7 +1076,11 @@ class ValidationRunner:
 
     @staticmethod
     def _build_stg_perturbation_config(
-        stg_blocks: list[int] | None, stg_mode: Literal["stg_av", "stg_v"]
+        stg_blocks: list[int] | None,
+        stg_mode: Literal["stg_av", "stg_v"],
+        num_blocks: int,
+        device: torch.device,
+        dtype: torch.dtype,
     ) -> BatchedPerturbationConfig:
         """Build STG perturbation config that skips self-attention in the specified blocks."""
         perturbations: list[Perturbation] = [
@@ -1080,7 +1088,7 @@ class ValidationRunner:
         ]
         if stg_mode == "stg_av":
             perturbations.append(Perturbation(type=PerturbationType.SKIP_AUDIO_SELF_ATTN, blocks=stg_blocks))
-        return BatchedPerturbationConfig(perturbations=[PerturbationConfig(perturbations=perturbations)])
+        return BatchedPerturbationConfig([PerturbationConfig(perturbations=perturbations)], num_blocks, device, dtype)
 
     @staticmethod
     def _load_first_frame(media_path: Path) -> Tensor: