Update README and comments in examples code

README.md

@@ -50,14 +50,17 @@ classification example:
 git clone https://github.com/robertknight/rten.git
 cd rten
 
+# Install dependencies for model conversion
+pip install -e rten-convert
+
 # Install dependencies for Python scripts
 pip install -r tools/requirements.txt
 
 # Export an ONNX model. We're using resnet-50, a classic image classification model.
 python -m tools.export-timm-model timm/resnet50.a1_in1k
 
 # Convert model to this library's format
-tools/convert-onnx.py resnet50.a1_in1k.onnx resnet50.rten
+rten-convert resnet50.a1_in1k.onnx resnet50.rten
 
 # Run image classification example. Replace `image.png` with your own image.
 cargo run -p rten-examples --release --bin imagenet mobilenet resnet50.rten image.png
@@ -76,8 +79,8 @@ run:
 
 ```sh
 git clone https://github.com/robertknight/rten.git
-pip install -r rten/tools/requirements.txt
-rten/tools/convert-onnx.py your-model.onnx output.rten
+pip install -e rten/rten-convert
+rten-convert your-model.onnx output.rten
 ```
 
 The RTen model format does not yet guarantee long-term backwards compatibility,
@@ -111,7 +114,7 @@ a JavaScript project are:
     or [Hugging Face](https://huggingface.co/docs/transformers/serialization).
 2.  If the model is not already in ONNX format, convert it to ONNX. PyTorch
     users can use [torch.onnx](https://pytorch.org/docs/stable/onnx.html) for this.
-3.  Use the `tools/convert-onnx.py` script in this repository to convert the model
+3.  Use the `rten-convert` package in this repository to convert the model
     to the optimized format RTen uses. See the section above on converting models.
 4.  In your JavaScript code, fetch the WebAssembly binary and initialize RTen
     using the `init` function.

docs/adding-operators.md

@@ -19,7 +19,7 @@ In detail, the process is:
    attributes from that operator.
 3. Run `make` to generate updated Rust and Python code to read the updated
    FlatBuffers schema
-4. If the new operator has attributes, edit `tools/convert-onnx.py` to read
+4. If the new operator has attributes, edit `rten-convert/rten_convert/converter.py` and reinstall rten-convert to read
    the attributes from ONNX and convert to this library's model format
 5. Define the implementation of the new operator in Rust. This is a struct
    that implements the `Operator` trait.

js-examples/image-classification/README.md

@@ -14,7 +14,9 @@ kind of object.
 
    ```sh
    curl -L https://github.com/onnx/models/raw/main/vision/classification/mobilenet/model/mobilenetv2-10.onnx -o mobilenet.onnx
-   ../../tools/convert-onnx.py mobilenet.onnx mobilenet.rten
+
+   pip install -e ../../rten-convert
+   rten-convert mobilenet.onnx mobilenet.rten
    ```
 4. Follow either of the subsections below to run the example in Node or the
    browser

rten-cli/src/main.rs

@@ -293,7 +293,8 @@ fn print_input_output_list(model: &Model, node_ids: &[NodeId]) {
 /// generated inputs.
 ///
 /// ```
-/// tools/convert-onnx.py model.onnx output.rten
+/// pip install -e rten-convert
+/// rten-convert model.onnx output.rten
 /// cargo run -p rten-cli --release output.rten
 /// ```
 ///

rten-examples/README.md

@@ -16,11 +16,12 @@ The general steps to run an example are:
    Face](https://huggingface.co/docs/optimum/exporters/onnx/overview),
    the [ONNX Model Zoo](https://github.com/onnx/models) or pre-created ONNX
    models by the model authors.
-2. Convert the ONNX model to this library's format using the `convert-onnx.py`
-   script:
+2. Convert the ONNX model to this library's format using the `rten-convert`
+   package:
 
    ```sh
-   $ ../tools/convert-onnx.py <onnx_model> <output_model>
+   $ pip install -e rten-convert
+   $ rten-convert <onnx_model> <output_model>
    ```
 
 3. Run the example using:

rten-examples/src/bert_qa.rs

@@ -206,7 +206,8 @@ fn extract_nbest_answers<'a>(
 ///
 /// ```
 /// optimum-cli export onnx --model distilbert-base-cased-distilled-squad distilbert
-/// tools/convert-onnx.py distilbert/model.onnx distilbert/distilbert.rten
+/// pip install -e rten-convert
+/// rten-convert distilbert/model.onnx distilbert/distilbert.rten
 /// ```
 ///
 /// Then run the example with:

rten-examples/src/deeplab.rs

@@ -95,7 +95,8 @@ const PASCAL_VOC_LABELS: [(&str, Rgb); 21] = [
 ///
 /// ```
 /// python examples/export-deeplab.py
-/// tools/convert-onnx.py deeplab.onnx deeplab.rten
+/// pip install -e rten-convert
+/// rten-convert deeplab.onnx deeplab.rten
 /// ```
 ///
 /// Run this program on an image with:

rten-examples/src/depth_anything.rs

@@ -66,7 +66,8 @@ Args:
 /// After downloading the model, it can be run on an image using:
 ///
 /// ```
-/// tools/convert-onnx.py depth_anything.onnx
+/// pip install -e rten-convert
+/// rten-convert depth_anything.onnx
 /// cargo run --release --bin depth_anything depth_anything.rten image.jpg
 /// ```
 ///

rten-examples/src/detr.rs

@@ -261,7 +261,8 @@ const LABELS: &[&str] = &[
 ///
 /// ```
 /// optimum-cli export onnx --model facebook/detr-resnet-50 detr
-/// tools/convert-onnx.py detr/model.onnx detr.rten
+/// pip install -e rten-convert
+/// rten-convert detr/model.onnx detr.rten
 /// ```
 ///
 /// This model also works with YOLOS. Use `hustvl/yolos-tiny` or

rten-examples/src/jina_similarity.rs

@@ -176,7 +176,8 @@ fn embed_sentence_batch(
 /// Convert the model using:
 ///
 /// ```text
-/// tools/convert-onnx.py jina-embed.onnx jina-embed.rten
+/// pip install -e rten-convert
+/// rten-convert jina-embed.onnx jina-embed.rten
 /// ```
 ///
 /// Then run the example with:

rten-examples/src/wav2vec2.rs

@@ -78,7 +78,8 @@ fn read_wav_file(path: &str) -> Result<Vec<f32>, hound::Error> {
 ///
 /// ```
 /// optimum-cli export onnx --model facebook/wav2vec2-base-960h wav2vec2
-/// tools/convert-onnx.py wav2vec2/model.onnx wav2vec2.rten
+/// pip install -e rten-convert
+/// rten-convert wav2vec2/model.onnx wav2vec2.rten
 /// ```
 ///
 /// To record a .wav file and test this app:

rten-examples/src/yolo.rs

@@ -80,7 +80,8 @@ fn resource_path(path: &str) -> PathBuf {
 /// ```
 /// pip install ultralytics
 /// yolo mode=export model=yolov8s.pt format=onnx
-/// tools/convert-onnx.py yolov8s.onnx yolov8.rten
+/// pip install -e rten-convert
+/// rten-convert yolov8s.onnx yolov8.rten
 /// ```
 ///
 /// Run this program on an image: