Merge pull request #91 from invoke-ai/fix-static-asset-paths

Include static files required by the UI in the python package.

README.md

@@ -44,7 +44,7 @@ pip install -e ".[test]" --extra-index-url https://download.pytorch.org/whl/cu12
 
 Run training via the CLI with type-checked YAML configuration files for maximum control:
 ```bash
-invoke-train --cfg-file configs/sd_lora_pokemon_1x8gb.yaml
+invoke-train --cfg-file src/invoke_training/sample_configs/sd_lora_pokemon_1x8gb.yaml
 ```
 
 ### GUI

docs/get-started/quick-start-cli.md

@@ -15,12 +15,12 @@ This training process has been tested on an NVIDIA GPU with 8GB of VRAM.
 Follow the [`invoke-training` installation instructions](./installation.md).
 
 ### 2. Config file
-We will use the [configs/sd_lora_pokemon_1x8gb.yaml](https://github.com/invoke-ai/invoke-training/blob/main/configs/sd_lora_pokemon_1x8gb.yaml) (SD v1.5, 8GB VRAM) configuration file for this tutorial. This configuration file controls all of the parameters of the training process and has been pre-configured for this tutorial. If you're curious, you can learn more about the configuration file format in the [full pipeline tutorials](../tutorials/index.md) or the [configuration reference](../reference/config/index.md) docs.
+We will use the [src/invoke_training/sample_configs/sd_lora_pokemon_1x8gb.yaml](https://github.com/invoke-ai/invoke-training/blob/main/src/invoke_training/sample_configs/sd_lora_pokemon_1x8gb.yaml) (SD v1.5, 8GB VRAM) configuration file for this tutorial. This configuration file controls all of the parameters of the training process and has been pre-configured for this tutorial. If you're curious, you can learn more about the configuration file format in the [full pipeline tutorials](../tutorials/index.md) or the [configuration reference](../reference/config/index.md) docs.
 
 ### 3. Start training!
 Start the training pipeline:
 ```bash
-invoke-train --cfg-file configs/sd_lora_pokemon_1x8gb.yaml
+invoke-train --cfg-file src/invoke_training/sample_configs/sd_lora_pokemon_1x8gb.yaml
 ```
 
 ### 4. Monitor training

docs/reference/config/pipelines/sdxl_textual_inversion.md

@@ -1,9 +1,9 @@
 # `SdxlTextualInversionConfig`
 
-Below is a sample yaml config file for Textual Inversion SDXL training ([raw file](https://github.com/invoke-ai/invoke-training/blob/main/configs/sdxl_textual_inversion_gnome_1x24gb.yaml)). All of the configuration fields are explained in detail on this page.
+Below is a sample yaml config file for Textual Inversion SDXL training ([raw file](https://github.com/invoke-ai/invoke-training/blob/main/src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml)). All of the configuration fields are explained in detail on this page.
 
 ```yaml title="sdxl_textual_inversion_gnome_1x24gb.yaml"
---8<-- "configs/sdxl_textual_inversion_gnome_1x24gb.yaml"
+--8<-- "src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml"
 ```
 
 <!-- To control the member order, we first list out the members whose order we care about, then we list the rest. -->

docs/tutorials/stable_diffusion/dpo_lora_sd.md

@@ -12,7 +12,7 @@ The Diffusion-DPO paper does full model fine-tuning on the [pickapic_v2](https:/
 
 Run this experiment with the following command:
 ```bash
-invoke-train -c configs/_experimental/sd_dpo_lora_pickapic_1x24gb.yaml
+invoke-train -c src/invoke_training/sample_configs/_experimental/sd_dpo_lora_pickapic_1x24gb.yaml
 ```
 
 Here is a cherry-picked example of a prompt for which this training process was clearly beneficial.
@@ -36,7 +36,7 @@ Note: The steps listed below are pretty rough. They are included primarily for r
 ### 1. Train a style LoRA
 
 ```bash
-invoke-train -c configs/sd_lora_pokemon_1x8gb.yaml
+invoke-train -c src/invoke_training/sample_configs/sd_lora_pokemon_1x8gb.yaml
 ```
 
 ### 2. Generate images
@@ -80,13 +80,13 @@ Launch the gradio UI for selecting image pair preferences.
 
 ```bash
 # Note: rank_images.py accepts a full training pipeline config, but only uses the dataset configuration.
-python src/invoke_training/scripts/_experimental/rank_images.py -c configs/_experimental/sd_dpo_lora_refinement_pokemon_1x24gb.yaml
+python src/invoke_training/scripts/_experimental/rank_images.py -c src/invoke_training/sample_configs/_experimental/sd_dpo_lora_refinement_pokemon_1x24gb.yaml
 ```
 
 After completing the pair annotations, click "Save Metadata" and move the resultant metadata file to your image data directory (e.g. `output/pokemon_pairs/metadata.jsonl`).
 
 ### 4. Run Diffusion-DPO
 
 ```bash
-invoke-train -c configs/_experimental/sd_dpo_lora_refinement_pokemon_1x24gb.yaml
+invoke-train -c src/invoke_training/sample_configs/_experimental/sd_dpo_lora_refinement_pokemon_1x24gb.yaml
 ```

docs/tutorials/stable_diffusion/textual_inversion_sdxl.md

@@ -22,12 +22,12 @@ Here are a few tips for preparing a Textual Inversion dataset:
 
 Below is the training configuration that we'll use for this tutorial.
 
-Raw config file: [configs/sdxl_textual_inversion_gnome_1x24gb.yaml](https://github.com/invoke-ai/invoke-training/blob/main/configs/sdxl_textual_inversion_gnome_1x24gb.yaml).
+Raw config file: [src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml](https://github.com/invoke-ai/invoke-training/blob/main/src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml).
 
 Full config reference docs: [Textual Inversion SDXL Config](../../reference/config/pipelines/sdxl_textual_inversion.md)
 
 ```yaml title="sdxl_textual_inversion_gnome_1x24gb.yaml"
---8<-- "configs/sdxl_textual_inversion_gnome_1x24gb.yaml"
+--8<-- "src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml"
 ```
 
 ## 3 - Start Training
@@ -37,7 +37,7 @@ Full config reference docs: [Textual Inversion SDXL Config](../../reference/conf
 Launch the Textual Inversion training pipeline:
 ```bash
 # From inside the invoke-training/ source directory:
-invoke-train -c configs/sdxl_textual_inversion_gnome_1x24gb.yaml
+invoke-train -c src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml
 ```
 
 Training takes ~40 mins on an NVIDIA RTX 4090.

pyproject.toml

@@ -57,6 +57,10 @@ dependencies = [
 "Homepage" = "https://github.com/invoke-ai/invoke-training"
 "Discord" = "https://discord.gg/ZmtBAhwWhy"
 
+[tool.setuptools.package-data]
+"invoke_training.sample_configs" = ["**/*.yaml"]
+"invoke_training.assets" = ["*.png"]
+
 [tool.ruff]
 src = ["src"]
 select = ["E", "F", "W", "C9", "N8", "I"]

src/invoke_training/ui/utils.py

@@ -8,14 +8,14 @@
 
 
 def get_config_dir_path() -> Path:
-    p = Path(__file__).parent.parent.parent.parent / "configs"
+    p = Path(__file__).parent.parent / "sample_configs"
     if not p.exists():
         raise FileNotFoundError(f"Config directory not found: '{p}'")
     return p
 
 
 def get_assets_dir_path() -> Path:
-    p = Path(__file__).parent.parent.parent.parent / "assets"
+    p = Path(__file__).parent.parent / "assets"
     if not p.exists():
         pass
     return p

tests/invoke_training/config/pipelines/test_pipeline_config.py

@@ -8,9 +8,9 @@
 
 
 def test_pipeline_config():
-    """Test that all sample pipeline configs can be parse as PipelineConfigs."""
+    """Test that all sample pipeline configs can be parsed as PipelineConfigs."""
     cur_file = Path(__file__)
-    config_dir = cur_file.parent.parent.parent.parent.parent / "configs"
+    config_dir = cur_file.parent.parent.parent.parent.parent / "src/invoke_training/sample_configs"
     config_files = glob.glob(str(config_dir) + "/**/*.yaml", recursive=True)
 
     assert len(config_files) > 0