Final configuration: Link all 4 model repos via README.md

README.md

@@ -1,10 +1,19 @@
 ---
-license: apache-2.0
-title: wan lora trainer
+title: Wan LoRA Trainer
 sdk: docker
+# This links all four required model repositories.
+# Each one will be mounted as a separate folder inside the Space.
 models:
-- Wan-AI/Wan2.1-I2V-14B-720P
+- wan-video/wan2.1-i2v-14B-fp8-720p
+- wan-video/wan2.1-vae
+- wan-video/wan2.1-clip-xlm-roberta
+- wan-video/wan2.1-t5-xxl
 ---
+
+# Wan 2.1 LoRA Trainer
+
+This Space runs the Wan 2.1 LoRA training script.
+The required models are linked via the repository configuration above.
 ---
 # Simple GUI for [Musubi Tuner](https://github.com/kohya-ss/musubi-tuner) (Wan 2.1 models only)
 
@@ -16,57 +25,4 @@ models:
 
 - To open the GUI just run `Start_Wan_GUI.bat`.
 - All settings can be saved and loaded using the "**Load Settings**" and "**Save Setting**" buttons.
-- More info about settings see in [Wan2.1 documentation](./docs/wan.md), [Advanced Configuration](./docs/advanced_config.md#fp8-quantization), [Dataset configuration guide](./dataset/dataset_config.md).
-
-
-![Preview](docs/Preview.png)
-
-
-
-
-
-# Miscellaneous
-
-
-## SageAttention Installation
-
-sdbsd has provided a Windows-compatible SageAttention implementation and pre-built wheels here:  https://github.com/sdbds/SageAttention-for-windows. After installing triton, if your Python, PyTorch, and CUDA versions match, you can download and install the pre-built wheel from the [Releases](https://github.com/sdbds/SageAttention-for-windows/releases) page. Thanks to sdbsd for this contribution.
-
-For reference, the build and installation instructions are as follows. You may need to update Microsoft Visual C++ Redistributable to the latest version.
-
-1. Download and install triton 3.1.0 wheel matching your Python version from [here](https://github.com/woct0rdho/triton-windows/releases/tag/v3.1.0-windows.post5).
-
-2. Install Microsoft Visual Studio 2022 or Build Tools for Visual Studio 2022, configured for C++ builds.
-
-3. Clone the SageAttention repository in your preferred directory:
-    ```shell
-    git clone https://github.com/thu-ml/SageAttention.git
-    ```
-
-    You can skip step 4 by using the sdbsd repository mentioned above by `git clone https://github.com/sdbds/SageAttention-for-windows.git`.
-
-4. Open `math.cuh` in the `SageAttention/csrc` folder and change `ushort` to `unsigned short` on lines 71 and 146, then save.
-
-5. Open `x64 Native Tools Command Prompt for VS 2022` from the Start menu under Visual Studio 2022.
-
-6. Activate your venv, navigate to the SageAttention folder, and run the following command. If you get a DISTUTILS not configured error, set `set DISTUTILS_USE_SDK=1` and try again:
-    ```shell
-    python setup.py install
-    ```
-
-This completes the SageAttention installation.
-
-### PyTorch version
-
-If you specify `torch` for `--attn_mode`, use PyTorch 2.5.1 or later (earlier versions may result in black videos).
-
-If you use an earlier version, use xformers or SageAttention.
-
-
-# License
-
-Code under the `hunyuan_model` directory is modified from [HunyuanVideo](https://github.com/Tencent/HunyuanVideo) and follows their license.
-
-Code under the `wan` directory is modified from [Wan2.1](https://github.com/Wan-Video/Wan2.1). The license is under the Apache License 2.0.
-
-Other code is under the Apache License 2.0. Some code is copied and modified from Diffusers.
+- More info about settings see in [Wan2.1 documentation](./docs/wan.md), [Advanced Configuration](./docs/advanced_config.md#fp8-quantization), [Dataset configuration guide](./dataset/dataset_config.md).

start.sh

@@ -1,36 +1,43 @@
 #!/bin/bash
-# --- Final, Simplified Startup Script (v10) ---
+# --- Final Definitive Startup Script (v11) ---
 
 # Exit immediately if a command exits with a non-zero status.
 set -e
 
 echo "--- Startup Script Initialized ---"
-echo "--- Models are mounted by the Space. No downloads needed. ---"
+echo "--- Models are mounted by the Space from multiple repositories. ---"
 
-# The model path is the name of the mounted repository.
-MODEL_DIR="/Wan2.1-I2V-14B-720P"
-
-# The output will be saved to the persistent /data directory.
-# The training script will create the output subfolder if needed.
+# --- Define the correct paths for each mounted model repository ---
+DIT_DIR="/wan2.1-i2v-14B-fp8-720p"
+VAE_DIR="/wan2.1-vae"
+CLIP_DIR="/wan2.1-clip-xlm-roberta"
+T5_DIR="/wan2.1-t5-xxl"
 OUTPUT_DIR="/data/output"
 
-echo "Using models from: $MODEL_DIR"
-echo "Saving output to: $OUTPUT_DIR"
+echo "DiT Path: $DIT_DIR"
+echo "VAE Path: $VAE_DIR"
+echo "CLIP Path: $CLIP_DIR"
+echo "T5 Path: $T5_DIR"
+echo "Output Path: $OUTPUT_DIR"
 
-# Verify that the main model file exists before starting training
-if [ ! -f "$MODEL_DIR/wan2.1_i2v_720p_14B_fp8_e4m3fn.safetensors" ]; then
-    echo "CRITICAL ERROR: Main model file not found. Check if the model repository is linked correctly in README.md. Exiting."
+# For robust verification, check for the existence of one file from each repo
+if [ ! -f "$DIT_DIR/wan2.1_i2v_720p_14B_fp8_e4m3fn.safetensors" ]; then
+    echo "CRITICAL ERROR: DiT model not found. Check README.md linking for 'wan-video/wan2.1-i2v-14B-fp8-720p'."
+    exit 1
+fi
+if [ ! -f "$VAE_DIR/Wan2.1_VAE.pth" ]; then
+    echo "CRITICAL ERROR: VAE model not found. Check README.md linking for 'wan-video/wan2.1-vae'."
     exit 1
 fi
 
-# Run the training command.
-echo "--- Starting training... ---"
+echo "All model repositories appear to be linked correctly. Starting training..."
+# Run the training command with the correct paths
 accelerate launch wan_train_network.py \
      --task "i2v-14B" \
-     --dit "$MODEL_DIR/wan2.1_i2v_720p_14B_fp8_e4m3fn.safetensors" \
-     --vae "$MODEL_DIR/Wan2.1_VAE.pth" \
-     --clip "$MODEL_DIR/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
-     --t5 "$MODEL_DIR/models_t5_umt5-xxl-enc-bf16.pth" \
+     --dit "$DIT_DIR/wan2.1_i2v_720p_14B_fp8_e4m3fn.safetensors" \
+     --vae "$VAE_DIR/Wan2.1_VAE.pth" \
+     --clip "$CLIP_DIR/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
+     --t5 "$T5_DIR/models_t5_umt5-xxl-enc-bf16.pth" \
      --dataset_config "dataset/testtoml.toml" \
      --output_dir "$OUTPUT_DIR" \
      --output_name "My_HF_Lora_v1" \