import * as GCP from "@alchemy.run/gcp";
import * as Effect from "effect/Effect";
import { type Probe, serveProbes } from "../../bindingHost.ts";
import { region, runVersionLifecycle } from "../common.ts";
export const Classifier = GCP.ML.Model("Classifier", {
description: "binding probe",
regions: [region],
});
/** Needs a trained SavedModel at GCP_TEST_ML_DEPLOYMENT_URI. */
export const V1 = Effect.gen(function* () {
const model = yield* Classifier;
return yield* GCP.ML.ModelsVersion("V1", {
model: model.name,
deploymentUri: process.env.GCP_TEST_ML_DEPLOYMENT_URI ?? "",
runtimeVersion: "2.11",
pythonVersion: "3.7",
framework: "TENSORFLOW",
});
});
/**
* Effect-native Cloud Run service exercising every AI Platform (legacy ML)
* binding as its own runtime service account. The version (and its
* GetVersion binding) is only declared when `runVersionLifecycle` is set.
* Deployed from {@link ../Bindings.test.ts}.
*/
export default class MlBindingsHost extends GCP.Function<MlBindingsHost>()(
"MlBindingsHost",
{ main: import.meta.url, invokerIamDisabled: true },
Effect.gen(function* () {
const model = yield* Classifier;
const getModel = yield* GCP.ML.GetModel(model);
const predict = yield* GCP.ML.Predict(model);
const versionRoutes: Record<string, Probe> = runVersionLifecycle
? { getVersion: (yield* GCP.ML.GetVersion(yield* V1))() }
: {};
return {
fetch: serveProbes({
getModel: getModel(),
predict: predict({
body: {
httpBody: {
contentType: "application/json",
data: btoa(JSON.stringify({ instances: [{ f1: 1 }] })),
},
},
}),
...versionRoutes,
}),
};
}).pipe(
Effect.provide(GCP.ML.GetModelHttp),
Effect.provide(GCP.ML.PredictHttp),
Effect.provide(GCP.ML.GetVersionHttp),
),
) {}