VinF Hybrid Inference #1: Update the API

common/api-review/vertexai.api.md

@@ -326,6 +326,7 @@ export interface GenerativeContentBlob {
 export class GenerativeModel extends VertexAIModel {
     constructor(vertexAI: VertexAI, modelParams: ModelParams, requestOptions?: RequestOptions);
     countTokens(request: CountTokensRequest | string | Array<string | Part>): Promise<CountTokensResponse>;
+    static DEFAULT_HYBRID_IN_CLOUD_MODEL: string;
     generateContent(request: GenerateContentRequest | string | Array<string | Part>): Promise<GenerateContentResult>;
     generateContentStream(request: GenerateContentRequest | string | Array<string | Part>): Promise<GenerateContentStreamResult>;
     // (undocumented)
@@ -344,7 +345,7 @@ export class GenerativeModel extends VertexAIModel {
 }
 
 // @public
-export function getGenerativeModel(vertexAI: VertexAI, modelParams: ModelParams, requestOptions?: RequestOptions): GenerativeModel;
+export function getGenerativeModel(vertexAI: VertexAI, modelParams: ModelParams | HybridParams, requestOptions?: RequestOptions): GenerativeModel;
 
 // @beta
 export function getImagenModel(vertexAI: VertexAI, modelParams: ImagenModelParams, requestOptions?: RequestOptions): ImagenModel;
@@ -416,6 +417,13 @@ export enum HarmSeverity {
     HARM_SEVERITY_NEGLIGIBLE = "HARM_SEVERITY_NEGLIGIBLE"
 }
 
+// @public
+export interface HybridParams {
+    inCloudParams?: ModelParams;
+    mode: InferenceMode;
+    onDeviceParams?: AILanguageModelCreateOptionsWithSystemPrompt;
+}
+
 // @beta
 export enum ImagenAspectRatio {
     LANDSCAPE_16x9 = "16:9",
@@ -500,6 +508,9 @@ export interface ImagenSafetySettings {
     safetyFilterLevel?: ImagenSafetyFilterLevel;
 }
 
+// @public
+export type InferenceMode = 'prefer_on_device' | 'only_on_device' | 'only_in_cloud';
+
 // @public
 export interface InlineDataPart {
     // (undocumented)

docs-devsite/_toc.yaml

@@ -536,6 +536,8 @@ toc:
     path: /docs/reference/js/vertexai.groundingattribution.md
   - title: GroundingMetadata
     path: /docs/reference/js/vertexai.groundingmetadata.md
+  - title: HybridParams
+    path: /docs/reference/js/vertexai.hybridparams.md
   - title: ImagenGCSImage
     path: /docs/reference/js/vertexai.imagengcsimage.md
   - title: ImagenGenerationConfig

docs-devsite/vertexai.generativemodel.md

@@ -29,6 +29,7 @@ export declare class GenerativeModel extends VertexAIModel
 
 |  Property | Modifiers | Type | Description |
 |  --- | --- | --- | --- |
+|  [DEFAULT\_HYBRID\_IN\_CLOUD\_MODEL](./vertexai.generativemodel.md#generativemodeldefault_hybrid_in_cloud_model) | <code>static</code> | string | Defines the name of the default in-cloud model to use for hybrid inference. |
 |  [generationConfig](./vertexai.generativemodel.md#generativemodelgenerationconfig) |  | [GenerationConfig](./vertexai.generationconfig.md#generationconfig_interface) |  |
 |  [requestOptions](./vertexai.generativemodel.md#generativemodelrequestoptions) |  | [RequestOptions](./vertexai.requestoptions.md#requestoptions_interface) |  |
 |  [safetySettings](./vertexai.generativemodel.md#generativemodelsafetysettings) |  | [SafetySetting](./vertexai.safetysetting.md#safetysetting_interface)<!-- -->\[\] |  |
@@ -63,6 +64,16 @@ constructor(vertexAI: VertexAI, modelParams: ModelParams, requestOptions?: Reque
 |  modelParams | [ModelParams](./vertexai.modelparams.md#modelparams_interface) |  |
 |  requestOptions | [RequestOptions](./vertexai.requestoptions.md#requestoptions_interface) |  |
 
+## GenerativeModel.DEFAULT\_HYBRID\_IN\_CLOUD\_MODEL
+
+Defines the name of the default in-cloud model to use for hybrid inference.
+
+<b>Signature:</b>
+
+```typescript
+static DEFAULT_HYBRID_IN_CLOUD_MODEL: string;
+```
+
 ## GenerativeModel.generationConfig
 
 <b>Signature:</b>

docs-devsite/vertexai.hybridparams.md

@@ -0,0 +1,57 @@
+Project: /docs/reference/js/_project.yaml
+Book: /docs/reference/_book.yaml
+page_type: reference
+
+{% comment %}
+DO NOT EDIT THIS FILE!
+This is generated by the JS SDK team, and any local changes will be
+overwritten. Changes should be made in the source code at
+https://github.com/firebase/firebase-js-sdk
+{% endcomment %}
+
+# HybridParams interface
+Toggles hybrid inference.
+
+<b>Signature:</b>
+
+```typescript
+export interface HybridParams 
+```
+
+## Properties
+
+|  Property | Type | Description |
+|  --- | --- | --- |
+|  [inCloudParams](./vertexai.hybridparams.md#hybridparamsincloudparams) | [ModelParams](./vertexai.modelparams.md#modelparams_interface) | Optional. Specifies advanced params for in-cloud inference. |
+|  [mode](./vertexai.hybridparams.md#hybridparamsmode) | [InferenceMode](./vertexai.md#inferencemode) | Specifies on-device or in-cloud inference. Defaults to prefer on-device. |
+|  [onDeviceParams](./vertexai.hybridparams.md#hybridparamsondeviceparams) | AILanguageModelCreateOptionsWithSystemPrompt | Optional. Specifies advanced params for on-device inference. |
+
+## HybridParams.inCloudParams
+
+Optional. Specifies advanced params for in-cloud inference.
+
+<b>Signature:</b>
+
+```typescript
+inCloudParams?: ModelParams;
+```
+
+## HybridParams.mode
+
+Specifies on-device or in-cloud inference. Defaults to prefer on-device.
+
+<b>Signature:</b>
+
+```typescript
+mode: InferenceMode;
+```
+
+## HybridParams.onDeviceParams
+
+Optional. Specifies advanced params for on-device inference.
+
+<b>Signature:</b>
+
+```typescript
+onDeviceParams?: AILanguageModelCreateOptionsWithSystemPrompt;
+```

docs-devsite/vertexai.md

@@ -19,7 +19,7 @@ The Vertex AI in Firebase Web SDK.
 |  <b>function(app, ...)</b> |
 |  [getVertexAI(app, options)](./vertexai.md#getvertexai_04094cf) | Returns a <code>[VertexAI](./vertexai.vertexai.md#vertexai_interface)</code> instance for the given app. |
 |  <b>function(vertexAI, ...)</b> |
-|  [getGenerativeModel(vertexAI, modelParams, requestOptions)](./vertexai.md#getgenerativemodel_e3037c9) | Returns a <code>[GenerativeModel](./vertexai.generativemodel.md#generativemodel_class)</code> class with methods for inference and other functionality. |
+|  [getGenerativeModel(vertexAI, modelParams, requestOptions)](./vertexai.md#getgenerativemodel_8dbc150) | Returns a <code>[GenerativeModel](./vertexai.generativemodel.md#generativemodel_class)</code> class with methods for inference and other functionality. |
 |  [getImagenModel(vertexAI, modelParams, requestOptions)](./vertexai.md#getimagenmodel_812c375) | <b><i>(Public Preview)</i></b> Returns an <code>[ImagenModel](./vertexai.imagenmodel.md#imagenmodel_class)</code> class with methods for using Imagen.<!-- -->Only Imagen 3 models (named <code>imagen-3.0-*</code>) are supported. |
 
 ## Classes
@@ -91,6 +91,7 @@ The Vertex AI in Firebase Web SDK.
 |  [GenerativeContentBlob](./vertexai.generativecontentblob.md#generativecontentblob_interface) | Interface for sending an image. |
 |  [GroundingAttribution](./vertexai.groundingattribution.md#groundingattribution_interface) |  |
 |  [GroundingMetadata](./vertexai.groundingmetadata.md#groundingmetadata_interface) | Metadata returned to client when grounding is enabled. |
+|  [HybridParams](./vertexai.hybridparams.md#hybridparams_interface) | Toggles hybrid inference. |
 |  [ImagenGCSImage](./vertexai.imagengcsimage.md#imagengcsimage_interface) | An image generated by Imagen, stored in a Cloud Storage for Firebase bucket.<!-- -->This feature is not available yet. |
 |  [ImagenGenerationConfig](./vertexai.imagengenerationconfig.md#imagengenerationconfig_interface) | <b><i>(Public Preview)</i></b> Configuration options for generating images with Imagen.<!-- -->See the [documentation](http://firebase.google.com/docs/vertex-ai/generate-images-imagen) for more details. |
 |  [ImagenGenerationResponse](./vertexai.imagengenerationresponse.md#imagengenerationresponse_interface) | <b><i>(Public Preview)</i></b> The response from a request to generate images with Imagen. |
@@ -99,10 +100,10 @@ The Vertex AI in Firebase Web SDK.
 |  [ImagenSafetySettings](./vertexai.imagensafetysettings.md#imagensafetysettings_interface) | <b><i>(Public Preview)</i></b> Settings for controlling the aggressiveness of filtering out sensitive content.<!-- -->See the [documentation](http://firebase.google.com/docs/vertex-ai/generate-images) for more details. |
 |  [InlineDataPart](./vertexai.inlinedatapart.md#inlinedatapart_interface) | Content part interface if the part represents an image. |
 |  [ModalityTokenCount](./vertexai.modalitytokencount.md#modalitytokencount_interface) | Represents token counting info for a single modality. |
-|  [ModelParams](./vertexai.modelparams.md#modelparams_interface) | Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_e3037c9)</code>. |
+|  [ModelParams](./vertexai.modelparams.md#modelparams_interface) | Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_8dbc150)</code>. |
 |  [ObjectSchemaInterface](./vertexai.objectschemainterface.md#objectschemainterface_interface) | Interface for <code>[ObjectSchema](./vertexai.objectschema.md#objectschema_class)</code> class. |
 |  [PromptFeedback](./vertexai.promptfeedback.md#promptfeedback_interface) | If the prompt was blocked, this will be populated with <code>blockReason</code> and the relevant <code>safetyRatings</code>. |
-|  [RequestOptions](./vertexai.requestoptions.md#requestoptions_interface) | Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_e3037c9)</code>. |
+|  [RequestOptions](./vertexai.requestoptions.md#requestoptions_interface) | Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_8dbc150)</code>. |
 |  [RetrievedContextAttribution](./vertexai.retrievedcontextattribution.md#retrievedcontextattribution_interface) |  |
 |  [SafetyRating](./vertexai.safetyrating.md#safetyrating_interface) | A safety rating associated with a <code>[GenerateContentCandidate](./vertexai.generatecontentcandidate.md#generatecontentcandidate_interface)</code> |
 |  [SafetySetting](./vertexai.safetysetting.md#safetysetting_interface) | Safety setting that can be sent as part of request parameters. |
@@ -130,6 +131,7 @@ The Vertex AI in Firebase Web SDK.
 
 |  Type Alias | Description |
 |  --- | --- |
+|  [InferenceMode](./vertexai.md#inferencemode) | Determines whether inference happens on-device or in-cloud. |
 |  [Part](./vertexai.md#part) | Content part - includes text, image/video, or function call/response part types. |
 |  [Role](./vertexai.md#role) | Role is the producer of the content. |
 |  [Tool](./vertexai.md#tool) | Defines a tool that model can call to access external knowledge. |
@@ -160,22 +162,22 @@ export declare function getVertexAI(app?: FirebaseApp, options?: VertexAIOptions
 
 ## function(vertexAI, ...)
 
-### getGenerativeModel(vertexAI, modelParams, requestOptions) {:#getgenerativemodel_e3037c9}
+### getGenerativeModel(vertexAI, modelParams, requestOptions) {:#getgenerativemodel_8dbc150}
 
 Returns a <code>[GenerativeModel](./vertexai.generativemodel.md#generativemodel_class)</code> class with methods for inference and other functionality.
 
 <b>Signature:</b>
 
 ```typescript
-export declare function getGenerativeModel(vertexAI: VertexAI, modelParams: ModelParams, requestOptions?: RequestOptions): GenerativeModel;
+export declare function getGenerativeModel(vertexAI: VertexAI, modelParams: ModelParams | HybridParams, requestOptions?: RequestOptions): GenerativeModel;
 ```
 
 #### Parameters
 
 |  Parameter | Type | Description |
 |  --- | --- | --- |
 |  vertexAI | [VertexAI](./vertexai.vertexai.md#vertexai_interface) |  |
-|  modelParams | [ModelParams](./vertexai.modelparams.md#modelparams_interface) |  |
+|  modelParams | [ModelParams](./vertexai.modelparams.md#modelparams_interface) \| [HybridParams](./vertexai.hybridparams.md#hybridparams_interface) |  |
 |  requestOptions | [RequestOptions](./vertexai.requestoptions.md#requestoptions_interface) |  |
 
 <b>Returns:</b>
@@ -223,6 +225,16 @@ Possible roles.
 POSSIBLE_ROLES: readonly ["user", "model", "function", "system"]
 ```
 
+## InferenceMode
+
+Determines whether inference happens on-device or in-cloud.
+
+<b>Signature:</b>
+
+```typescript
+export type InferenceMode = 'prefer_on_device' | 'only_on_device' | 'only_in_cloud';
+```
+
 ## Part
 
 Content part - includes text, image/video, or function call/response part types.

docs-devsite/vertexai.modelparams.md

@@ -10,7 +10,7 @@ https://github.com/firebase/firebase-js-sdk
 {% endcomment %}
 
 # ModelParams interface
-Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_e3037c9)</code>.
+Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_8dbc150)</code>.
 
 <b>Signature:</b>
 

docs-devsite/vertexai.requestoptions.md

@@ -10,7 +10,7 @@ https://github.com/firebase/firebase-js-sdk
 {% endcomment %}
 
 # RequestOptions interface
-Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_e3037c9)</code>.
+Params passed to <code>[getGenerativeModel()](./vertexai.md#getgenerativemodel_8dbc150)</code>.
 
 <b>Signature:</b>
 

package.json

@@ -56,11 +56,16 @@
     "type": "git",
     "url": "git+https://github.com/firebase/firebase-js-sdk.git"
   },
-  "workspaces": [
-    "packages/*",
-    "integration/*",
-    "repo-scripts/*"
-  ],
+  "workspaces": {
+    "packages": [
+      "packages/*",
+      "integration/*",
+      "repo-scripts/*"
+    ],
+    "nohoist": [
+      "**/vertexai/@types/dom-chromium-ai"
+    ]
+  },
   "devDependencies": {
     "@babel/core": "7.26.8",
     "@babel/plugin-transform-modules-commonjs": "7.26.3",

packages/vertexai/package.json

@@ -58,6 +58,7 @@
   "devDependencies": {
     "@firebase/app": "0.11.3",
     "@rollup/plugin-json": "6.1.0",
+    "@types/dom-chromium-ai": "0.0.6",
     "rollup": "2.79.2",
     "rollup-plugin-replace": "2.2.0",
     "rollup-plugin-typescript2": "0.36.0",
@@ -78,4 +79,4 @@
     ],
     "reportDir": "./coverage/node"
   }
-}
+}

packages/vertexai/src/api.test.ts

@@ -101,6 +101,21 @@ describe('Top level API', () => {
     expect(genModel).to.be.an.instanceOf(GenerativeModel);
     expect(genModel.model).to.equal('publishers/google/models/my-model');
   });
+  it('getGenerativeModel with HybridParams sets a default model', () => {
+    const genModel = getGenerativeModel(fakeVertexAI, {
+      mode: 'only_on_device'
+    });
+    expect(genModel.model).to.equal(
+      `publishers/google/models/${GenerativeModel.DEFAULT_HYBRID_IN_CLOUD_MODEL}`
+    );
+  });
+  it('getGenerativeModel with HybridParams honors a model override', () => {
+    const genModel = getGenerativeModel(fakeVertexAI, {
+      mode: 'only_in_cloud',
+      inCloudParams: { model: 'my-model' }
+    });
+    expect(genModel.model).to.equal('publishers/google/models/my-model');
+  });
   it('getImagenModel throws if no model is provided', () => {
     try {
       getImagenModel(fakeVertexAI, {} as ImagenModelParams);

packages/vertexai/src/api.ts

@@ -23,6 +23,7 @@ import { VertexAIService } from './service';
 import { VertexAI, VertexAIOptions } from './public-types';
 import {
   ImagenModelParams,
+  HybridParams,
   ModelParams,
   RequestOptions,
   VertexAIErrorCode
@@ -70,16 +71,27 @@ export function getVertexAI(
  */
 export function getGenerativeModel(
   vertexAI: VertexAI,
-  modelParams: ModelParams,
+  modelParams: ModelParams | HybridParams,
   requestOptions?: RequestOptions
 ): GenerativeModel {
-  if (!modelParams.model) {
+  // Uses the existence of HybridParams.mode to clarify the type of the modelParams input.
+  const hybridParams = modelParams as HybridParams;
+  let inCloudParams: ModelParams;
+  if (hybridParams.mode) {
+    inCloudParams = hybridParams.inCloudParams || {
+      model: GenerativeModel.DEFAULT_HYBRID_IN_CLOUD_MODEL
+    };
+  } else {
+    inCloudParams = modelParams as ModelParams;
+  }
+
+  if (!inCloudParams.model) {
     throw new VertexAIError(
       VertexAIErrorCode.NO_MODEL,
       `Must provide a model name. Example: getGenerativeModel({ model: 'my-model-name' })`
     );
   }
-  return new GenerativeModel(vertexAI, modelParams, requestOptions);
+  return new GenerativeModel(vertexAI, inCloudParams, requestOptions);
 }
 
 /**

packages/vertexai/src/models/generative-model.ts

@@ -49,6 +49,10 @@ import { VertexAIModel } from './vertexai-model';
  * @public
  */
 export class GenerativeModel extends VertexAIModel {
+  /**
+   * Defines the name of the default in-cloud model to use for hybrid inference.
+   */
+  static DEFAULT_HYBRID_IN_CLOUD_MODEL = 'gemini-2.0-flash-lite';
   generationConfig: GenerationConfig;
   safetySettings: SafetySetting[];
   requestOptions?: RequestOptions;

packages/vertexai/src/types/requests.ts

@@ -213,3 +213,28 @@ export interface FunctionCallingConfig {
   mode?: FunctionCallingMode;
   allowedFunctionNames?: string[];
 }
+
+/**
+ * Toggles hybrid inference.
+ */
+export interface HybridParams {
+  /**
+   * Specifies on-device or in-cloud inference. Defaults to prefer on-device.
+   */
+  mode: InferenceMode;
+  /**
+   * Optional. Specifies advanced params for on-device inference.
+   */
+  onDeviceParams?: AILanguageModelCreateOptionsWithSystemPrompt;
+  /**
+   * Optional. Specifies advanced params for in-cloud inference.
+   */
+  inCloudParams?: ModelParams;
+}
+/**
+ * Determines whether inference happens on-device or in-cloud.
+ */
+export type InferenceMode =
+  | 'prefer_on_device'
+  | 'only_on_device'
+  | 'only_in_cloud';

repo-scripts/changelog-generator/tsconfig.json

@@ -3,7 +3,8 @@
     "strict": true,
     "outDir": "dist",
     "lib": [
-      "ESNext"
+      "ESNext",
+      "dom"
     ],
     "module": "CommonJS",
     "moduleResolution": "node",

scripts/release/utils/workspace.ts

@@ -27,8 +27,10 @@ const writeFile = promisify(_writeFile);
 
 const {
   workspaces: rawWorkspaces
-}: { workspaces: string[] } = require(`${root}/package.json`);
-const workspaces = rawWorkspaces.map(workspace => `${root}/${workspace}`);
+}: { workspaces: { packages: string[] } } = require(`${root}/package.json`);
+const workspaces = rawWorkspaces.packages.map(
+  workspace => `${root}/${workspace}`
+);
 
 export function mapWorkspaceToPackages(
   workspaces: string[]