chore: start moving stuff off parcel

backends/tract/tests/session.rs

@@ -16,7 +16,7 @@ fn mnist_5() -> ort::Result<()> {
 
 	let mut session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level3)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")
 		.expect("Could not download model from file");
 
 	// Load image and resize to model's shape, converting to RGB format

examples/async-gpt2-api/examples/async-gpt2-api.rs

@@ -38,7 +38,7 @@ async fn main() -> anyhow::Result<()> {
 	let session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level1)?
 		.with_intra_threads(4)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/gpt2.onnx")?;
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/gpt2.onnx")?;
 
 	// Load the tokenizer and encode the prompt into a sequence of tokens.
 	let tokenizer = Tokenizer::from_file(Path::new(env!("CARGO_MANIFEST_DIR")).join("data").join("tokenizer.json")).unwrap();

examples/cudarc/src/main.rs

@@ -20,7 +20,7 @@ fn main() -> anyhow::Result<()> {
 		.commit()?;
 
 	let mut session =
-		Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/modnet_photographic_portrait_matting.onnx")?;
+		Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/modnet_photographic_portrait_matting.onnx")?;
 
 	let original_img = image::open(Path::new(env!("CARGO_MANIFEST_DIR")).join("data").join("photo.jpg")).unwrap();
 	let (img_width, img_height) = (original_img.width(), original_img.height());

examples/gpt2/examples/gpt2.rs

@@ -38,7 +38,7 @@ fn main() -> ort::Result<()> {
 	let mut session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level1)?
 		.with_intra_threads(1)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/gpt2.onnx")?;
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/gpt2.onnx")?;
 
 	// Load the tokenizer and encode the prompt into a sequence of tokens.
 	let tokenizer = Tokenizer::from_file(Path::new(env!("CARGO_MANIFEST_DIR")).join("data").join("tokenizer.json")).unwrap();

examples/modnet/examples/modnet.rs

@@ -16,7 +16,7 @@ fn main() -> ort::Result<()> {
 		.commit()?;
 
 	let mut session =
-		Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/modnet_photographic_portrait_matting.onnx")?;
+		Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/modnet_photographic_portrait_matting.onnx")?;
 
 	let original_img = image::open(Path::new(env!("CARGO_MANIFEST_DIR")).join("data").join("photo.jpg")).unwrap();
 	let (img_width, img_height) = (original_img.width(), original_img.height());

examples/sentence-transformers/examples/semantic-similarity.rs

@@ -28,7 +28,7 @@ fn main() -> ort::Result<()> {
 	let mut session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level1)?
 		.with_intra_threads(1)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/all-MiniLM-L6-v2.onnx")?;
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/all-MiniLM-L6-v2.onnx")?;
 
 	// Load the tokenizer and encode the text.
 	let tokenizer = Tokenizer::from_file(Path::new(env!("CARGO_MANIFEST_DIR")).join("data").join("tokenizer.json")).unwrap();

examples/training/README.md

@@ -10,7 +10,7 @@ pip install -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/O
 
 We're installing the CPU version of the `onnxruntime-training` & `torch` packages because we only need to use Python to *create* the initial graph which will be used for training. Run `python tools/train-data/mini-clm.py` from the root directory of the `ort` repo to create the training artifacts.
 
-Next, we need to convert our dataset into tokens to feed the model. This can be achieved by downloading the `oshicats-v2.jsonl` file from the OshiChats v2 dataset and running `cargo run -p example-training --example pretokenize -- ~/oshichats-v2.jsonl`, or if you (rightfully) don't wish to waste 30 GB worth of disk space and bandwidth on brainrot, you may download a [1 MB pre-tokenized subset of the dataset](https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_data/dataset.bin). Make sure `dataset.bin` is in the root of the `ort` repo.
+Next, we need to convert our dataset into tokens to feed the model. This can be achieved by downloading the `oshicats-v2.jsonl` file from the OshiChats v2 dataset and running `cargo run -p example-training --example pretokenize -- ~/oshichats-v2.jsonl`, or if you (rightfully) don't wish to waste 30 GB worth of disk space and bandwidth on brainrot, you may download a [1 MB pre-tokenized subset of the dataset](https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/train-clm-dataset.bin) (make sure `train-clm-dataset.bin` is in the root of the `ort` repo).
 
 Finally, we can train our model! Run `cargo run -p example-training --example train-clm` to start training. If you have an NVIDIA GPU, add `--features cuda` to enable CUDA, though it's not required and you can train directly on CPU instead. **This will use ~8 GB of (V)RAM!** You can lower the memory usage by adjusting the `BATCH_SIZE` and `SEQUENCE_LENGTH` constants in `train-clm.rs`, though note that changing the batch size may require adjustments to the learning rate.
 

examples/training/examples/pretokenize.rs

@@ -12,7 +12,7 @@ const MAX_TOKENS: usize = 500_000;
 
 fn main() {
 	let input = env::args().nth(1).expect("provide input jsonl");
-	let output = env::args().nth(2).unwrap_or_else(|| "dataset.bin".into());
+	let output = env::args().nth(2).unwrap_or_else(|| "train-clm-dataset.bin".into());
 
 	let input = BufReader::new(File::open(input).unwrap());
 	let mut output = BufWriter::new(File::create(output).unwrap());

examples/training/examples/train-clm-simple.rs

@@ -61,7 +61,7 @@ fn main() -> ort::Result<()> {
 	)
 	.unwrap();
 
-	let mut dataset = File::open("dataset.bin").unwrap();
+	let mut dataset = File::open("train-clm-dataset.bin").unwrap();
 	let file_size = dataset.metadata().unwrap().len();
 	let num_tokens = (file_size / 2) as usize; // 16-bit tokens
 	let mut rng = rand::thread_rng();

examples/training/examples/train-clm.rs

@@ -48,7 +48,7 @@ fn main() -> ort::Result<()> {
 	let mut optimizer = trainer.optimizer();
 	optimizer.set_lr(7e-5)?;
 
-	let mut dataset = File::open("dataset.bin").unwrap();
+	let mut dataset = File::open("train-clm-dataset.bin").unwrap();
 	let file_size = dataset.metadata().unwrap().len();
 	let num_tokens = (file_size / 2) as usize; // 16-bit tokens
 	let mut rng = rand::thread_rng();

examples/yolov8/examples/yolov8.rs

@@ -29,7 +29,7 @@ fn union(box1: &BoundingBox, box2: &BoundingBox) -> f32 {
 	((box1.x2 - box1.x1) * (box1.y2 - box1.y1)) + ((box2.x2 - box2.x1) * (box2.y2 - box2.y1)) - intersection(box1, box2)
 }
 
-const YOLOV8M_URL: &str = "https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/yolov8m.onnx";
+const YOLOV8M_URL: &str = "https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/yolov8m.onnx";
 
 #[rustfmt::skip]
 const YOLOV8_CLASS_LABELS: [&str; 80] = [

src/io_binding.rs

@@ -257,7 +257,7 @@ mod tests {
 	#[test]
 	#[cfg(all(feature = "ndarray", feature = "fetch-models"))]
 	fn test_mnist_input_bound() -> Result<()> {
-		let mut session = Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")?;
+		let mut session = Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")?;
 
 		let array = get_image();
 
@@ -275,7 +275,7 @@ mod tests {
 	#[test]
 	#[cfg(all(feature = "ndarray", feature = "fetch-models"))]
 	fn test_mnist_input_output_bound() -> Result<()> {
-		let mut session = Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")?;
+		let mut session = Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")?;
 
 		let array = get_image();
 
@@ -295,7 +295,7 @@ mod tests {
 	#[test]
 	#[cfg(all(feature = "ndarray", feature = "fetch-models"))]
 	fn test_send_iobinding() -> Result<()> {
-		let mut session = Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")?;
+		let mut session = Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")?;
 
 		let array = get_image();
 
@@ -320,7 +320,7 @@ mod tests {
 	#[test]
 	#[cfg(all(feature = "ndarray", feature = "fetch-models"))]
 	fn test_mnist_clear_bounds() -> Result<()> {
-		let mut session = Session::builder()?.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")?;
+		let mut session = Session::builder()?.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")?;
 
 		let array = get_image();
 

tests/mnist.rs

@@ -18,7 +18,7 @@ fn mnist_5() -> ort::Result<()> {
 	let mut session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level1)?
 		.with_intra_threads(1)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/mnist.onnx")
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/mnist.onnx")
 		.expect("Could not download model from file");
 
 	let input0_shape = {

tests/squeezenet.rs

@@ -23,7 +23,7 @@ fn squeezenet_mushroom() -> ort::Result<()> {
 	let mut session = Session::builder()?
 		.with_optimization_level(GraphOptimizationLevel::Level1)?
 		.with_intra_threads(1)?
-		.commit_from_url("https://parcel.pyke.io/v2/cdn/assetdelivery/ortrsv2/ex_models/squeezenet.onnx")
+		.commit_from_url("https://cdn.pyke.io/0/pyke:ort-rs/[email protected]/squeezenet.onnx")
 		.expect("Could not download model from file");
 
 	let class_labels = get_imagenet_labels()?;