feat: onnx runtime upgrade (#478)

* chore(sb_ai): update `Cargo.toml`

* chore(sb_ai): update `ONNXRUNTIME_VERSION`

* stamp: solving `ort` api breaking changes

- `ndarray_linalg`: Since `ort` deps to `ndarray=0.16`, decided to
remove this extra dependency. Normalization can be done in a simple
small funtion instead of deriving it from an external crate.

* stamp: converting `from_ort_tensor` without clonning

using the `rc-9` apis to get the same results from older `rc-2`

* test(sb_ai): removing snapshots from onnx `v1.19.2`

Since `ONNX`version has been upgraded, the snapshots need to be
recreated

* stamp: cleaning

* test(sb_ai): adding `macos_aarch64` snapshots with onnx `v1.20`

* feat: apply `normalize` without external deps

- Since `ndarray_linalg` has been removed we need to manually apply L2
normalization

* test(sb_ai): ensuring `mean_pool` and `normalize`

- Applying `dot product` between two embeddings to ensures that
`mean_pool` and `normalize` is working properly

* chore: add aarch64 (linux) snapshots for integration tests

* chore: add x86_64 (macos) snapshots for integration tests

* chore: add x86_64 (linux) snapshots for integration tests

* feat: add `Tensor<String>` support

Co-authored-by: Nyannyacha <[email protected]>
Co-authored-by: kallebysantos <[email protected]>

* test(sb_ai): adding `Tensor<string>` integration test

* fix(sb_ai): adding 'bool' alias to `JsTensorData`

* fix(sb_ai): parsing input tensors correctly

* test(sb_ai): updating `extract_tensor_ref` to match `JsTensorData` enum

* fix(ai): reflecting module rename

* stamp: enabling cache adapter on `ort-rust-backend` example

* stamp: applying format

* chore: update `Cargo.lock`

---------

Co-authored-by: Nyannyacha <[email protected]>
Co-authored-by: EC2 Default User <[email protected]>

Author: 3 people

.env

@@ -1,4 +1,4 @@
-ONNXRUNTIME_VERSION=1.19.2
+ONNXRUNTIME_VERSION=1.20.1
 DENO_VERSION=2.1.4
 EDGE_RUNTIME_PORT=9998
-AI_INFERENCE_API_HOST=http://localhost:11434
+AI_INFERENCE_API_HOST=http://localhost:11434

Cargo.lock

@@ -0,0 +1,32 @@
+import { assertExists, assertInstanceOf } from "jsr:@std/assert";
+
+// @ts-ignore: No types for 'onnxruntime' symbol
+const { Tensor, InferenceSession } = globalThis[Symbol.for("onnxruntime")];
+
+const modelUrl =
+  "https://huggingface.co/pirocheto/phishing-url-detection/resolve/main/model.onnx?download=true";
+
+const encoder = new TextEncoder();
+const modelUrlBuffer = encoder.encode(modelUrl);
+const session = await InferenceSession.create(modelUrlBuffer);
+
+Deno.serve(async (_req: Request) => {
+  const urls = [
+    "https://clubedemilhagem.com/home.php",
+    "http://www.medicalnewstoday.com/articles/188939.php",
+    "https://magalu-crediarioluiza.com/Produto_20203/produto.php?sku=1",
+  ];
+
+  const inputs = new Tensor("string", urls, [urls.length]);
+  const output = await session.run({ inputs });
+
+  // just need to make sure that model output something
+  // we don't care about the predicts values
+  assertExists(output.label);
+  assertExists(output.probabilities);
+
+  assertInstanceOf(output.label, Tensor);
+  assertInstanceOf(output.probabilities, Tensor);
+
+  return new Response();
+});