Enable model to be a remote argument, bypassing uploading it to DCP's server

Author: RyanSawe

README.md

@@ -13,14 +13,15 @@ Create a `model.json` file and add your model's information to it, for example
 {
   "name": "mnist-example",
   "version": "0.0.1",
+  "modelDownload": false,
   "model": "MNIST.onnx",
   "preprocess": "preprocess.py",
   "postprocess": "postprocess.py",
   "packages": ["numpy", "opencv-python"]
 }
 ```
 
-Where `model`, `preprocess`, and `postprocess` are all the paths to the model, preprocess, and postprocess files respectively. 
+Where `model`, `preprocess`, and `postprocess` are all the paths to the model, preprocess, and postprocess files respectively. If your `model.json` has `modelDownload` set to a URL instead of false you do not need to upload your model, instead you must have a web server hosting the model that workers can fetch from. See `example-server.js` for an example server. DistributiveWorkers intended to work on these slices must have the URL added to their allowOrigins object. 
 
 Packages are all of the python libraries your pre and post processing scripts require. The list of supported packages is:
 ```

example/example-server.js

@@ -0,0 +1,55 @@
+/**
+ * @file   example-server.js
+ *         Example web server for hosting model to be fetched by workers
+ *
+ * @author Ryan Saweczko, [email protected]
+ */
+
+const express = require('express');
+const path = require('path');
+const fs = require('fs');
+const cors = require('cors');
+
+var modelInfo;
+try
+{
+  modelInfo = JSON.parse(fs.readFileSync('model.json', { encoding: 'utf-8' }));
+}
+catch (error)
+{
+  console.error('Unable to find model.json file in directory, exiting');
+  process.exit(1);
+}
+
+const app = express();
+const port = 3500;
+app.use(cors());
+
+const fileDirectory = path.join(__dirname);
+
+app.get('/:filename', (req, res) => {
+  const { filename } = req.params;
+  const file = path.join(fileDirectory, filename);
+  if (filename === modelInfo.model)
+  {
+    const content = fs.readFileSync(file).toString('base64');
+    res.send(content);
+  }
+  else if ([modelInfo.preprocess, modelInfo.postprocess].includes(filename))
+  {
+    res.sendFile(file, (err) => {
+      if (err)
+        res.status(500).send('Error sending the file.');
+    });
+  }
+  else
+  {
+    console.error(`request for invalid file: ${filename}`);
+    res.status(404).send('Invalid file');
+  }
+});
+
+// Start server
+app.listen(port, () => {
+  console.log(`Server is running at http://localhost:${port}`);
+});

example/model-remote.json

@@ -0,0 +1,9 @@
+{
+  "name": "mnist-example",
+  "version": "0.0.1",
+  "modelDownload": "http://localhost:3500",
+  "model": "MNIST.onnx",
+  "preprocess": "preprocess.py",
+  "postprocess": "postprocess.py",
+  "packages": ["numpy", "opencv-python"]
+}

example/model.json

@@ -1,6 +1,7 @@
 {
   "name": "mnist-example",
   "version": "0.0.1",
+  "modelDownload": false,
   "model": "MNIST.onnx",
   "preprocess": "preprocess.py",
   "postprocess": "postprocess.py",

inference.js

@@ -13,18 +13,28 @@ const { workFunction } = require('./workFunction');
 /**
  * Deploy work function for model inferencing via DCP
  */
-async function deploy(inputSet, modelName, computeGroup, output, webgpu)
+async function deploy(inputSet, model, computeGroup, output, webgpu)
 {
 	const compute = require('dcp/compute');
 
-  const labels = { modelName, projectID: Date.now(), debug: false, webgpu };
-	let job = compute.for(inputSet, workFunction, [labels]);
+  const labels = { modelName: model.name, projectID: Date.now(), debug: false, webgpu };
+  const args = [labels];
+  if (model.modelDownload)
+  {
+    const url = new URL(model.modelDownload);
+    args.push(new URL(`/${model.preprocess}`, url));
+    args.push(new URL(`/${model.postprocess}`, url));
+    args.push(model.packages);
+    args.push(new URL(`/${model.model}`, url));
+  }
+	let job = compute.for(inputSet, workFunction, args);
 
-  job.public.name = `DCP Inferencing: ${modelName}`;
+  job.public.name = `DCP Inferencing: ${model.name}`;
   job.requires('onnxruntime-dcp/dcp-wasm.js');
   job.requires('onnxruntime-dcp/dcp-ort.js');
   job.requires('pyodide-core/pyodide-core.js');
-  job.requires(`${modelName}/module.js`);
+  if (!model.modelDownload)
+    job.requires(`${model.name}/module.js`);
 
   if (computeGroup)
     job.computeGroups    = [computeGroup];
@@ -151,7 +161,7 @@ if (require.main === module)
   }
 
   require('dcp-client').init().then(() => {
-    deploy(inputSet, modelInfo.name, computeGroup, outputFile, webgpu);
+    deploy(inputSet, modelInfo, computeGroup, outputFile, webgpu);
   });
 }