@firebase/ai 2.4.0-canary.261508183 → 2.4.0-canary.44d9891f9

package/dist/esm/src/methods/chrome-adapter.d.ts

@@ -25,11 +25,11 @@ import { LanguageModel } from '../types/language-model';
 export declare class ChromeAdapterImpl implements ChromeAdapter {
     languageModelProvider: LanguageModel;
     mode: InferenceMode;
-    onDeviceParams: OnDeviceParams;
     static SUPPORTED_MIME_TYPES: string[];
     private isDownloading;
     private downloadPromise;
     private oldSession;
+    onDeviceParams: OnDeviceParams;
     constructor(languageModelProvider: LanguageModel, mode: InferenceMode, onDeviceParams?: OnDeviceParams);
     /**
      * Checks if a given request can be made on-device.

package/dist/esm/src/requests/hybrid-helpers.d.ts

@@ -14,7 +14,11 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-import { GenerateContentRequest, ChromeAdapter } from '../types';
+import { GenerateContentRequest, ChromeAdapter, InferenceSource } from '../types';
+interface CallResult<Response> {
+    response: Response;
+    inferenceSource: InferenceSource;
+}
 /**
  * Dispatches a request to the appropriate backend (on-device or in-cloud)
  * based on the inference mode.
@@ -25,4 +29,5 @@ import { GenerateContentRequest, ChromeAdapter } from '../types';
  * @param inCloudCall - The function to call for in-cloud inference.
  * @returns The response from the backend.
  */
-export declare function callCloudOrDevice<Response>(request: GenerateContentRequest, chromeAdapter: ChromeAdapter | undefined, onDeviceCall: () => Promise<Response>, inCloudCall: () => Promise<Response>): Promise<Response>;
+export declare function callCloudOrDevice<Response>(request: GenerateContentRequest, chromeAdapter: ChromeAdapter | undefined, onDeviceCall: () => Promise<Response>, inCloudCall: () => Promise<Response>): Promise<CallResult<Response>>;
+export {};

package/dist/esm/src/requests/response-helpers.d.ts

@@ -14,12 +14,12 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-import { EnhancedGenerateContentResponse, FunctionCall, GenerateContentResponse, ImagenGCSImage, ImagenInlineImage, InlineDataPart, Part } from '../types';
+import { EnhancedGenerateContentResponse, FunctionCall, GenerateContentResponse, ImagenGCSImage, ImagenInlineImage, InlineDataPart, Part, InferenceSource } from '../types';
 /**
  * Creates an EnhancedGenerateContentResponse object that has helper functions and
  * other modifications that improve usability.
  */
-export declare function createEnhancedContentResponse(response: GenerateContentResponse): EnhancedGenerateContentResponse;
+export declare function createEnhancedContentResponse(response: GenerateContentResponse, inferenceSource?: InferenceSource): EnhancedGenerateContentResponse;
 /**
  * Adds convenience helper methods to a response object, including stream
  * chunks (as long as each chunk is a complete GenerateContentResponse JSON).

package/dist/esm/src/requests/stream-reader.d.ts

@@ -16,6 +16,7 @@
  */
 import { GenerateContentResponse, GenerateContentStreamResult } from '../types';
 import { ApiSettings } from '../types/internal';
+import { InferenceSource } from '../public-types';
 /**
  * Process a response.body stream from the backend and return an
  * iterator that provides one complete GenerateContentResponse at a time
@@ -24,7 +25,7 @@ import { ApiSettings } from '../types/internal';
  *
  * @param response - Response from a fetch call
  */
-export declare function processStream(response: Response, apiSettings: ApiSettings): GenerateContentStreamResult;
+export declare function processStream(response: Response, apiSettings: ApiSettings, inferenceSource?: InferenceSource): GenerateContentStreamResult;
 /**
  * Reads a raw stream from the fetch response and join incomplete
  * chunks, returning a new stream that provides a single complete

package/dist/esm/src/types/enums.d.ts

@@ -349,6 +349,21 @@ export declare const InferenceMode: {
  * @beta
  */
 export type InferenceMode = (typeof InferenceMode)[keyof typeof InferenceMode];
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export declare const InferenceSource: {
+    readonly ON_DEVICE: "on_device";
+    readonly IN_CLOUD: "in_cloud";
+};
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export type InferenceSource = (typeof InferenceSource)[keyof typeof InferenceSource];
 /**
  * Represents the result of the code execution.
  *

package/dist/esm/src/types/responses.d.ts

@@ -15,7 +15,7 @@
  * limitations under the License.
  */
 import { Content, FunctionCall, InlineDataPart } from './content';
-import { BlockReason, FinishReason, HarmCategory, HarmProbability, HarmSeverity, Modality } from './enums';
+import { BlockReason, FinishReason, HarmCategory, HarmProbability, HarmSeverity, InferenceSource, Modality } from './enums';
 /**
  * Result object returned from {@link GenerativeModel.generateContent} call.
  *
@@ -76,6 +76,12 @@ export interface EnhancedGenerateContentResponse extends GenerateContentResponse
      * set to `true`.
      */
     thoughtSummary: () => string | undefined;
+    /**
+     * Indicates whether inference happened on-device or in-cloud.
+     *
+     * @beta
+     */
+    inferenceSource?: InferenceSource;
 }
 /**
  * Individual response from {@link GenerativeModel.generateContent} and

package/dist/index.cjs.js

@@ -8,7 +8,7 @@ var util = require('@firebase/util');
 var logger$1 = require('@firebase/logger');
 
 var name = "@firebase/ai";
-var version = "2.4.0-canary.261508183";
+var version = "2.4.0-canary.44d9891f9";
 
 /**
  * @license
@@ -383,6 +383,15 @@ const InferenceMode = {
     'ONLY_IN_CLOUD': 'only_in_cloud',
     'PREFER_IN_CLOUD': 'prefer_in_cloud'
 };
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+const InferenceSource = {
+    'ON_DEVICE': 'on_device',
+    'IN_CLOUD': 'in_cloud'
+};
 /**
  * Represents the result of the code execution.
  *
@@ -892,22 +901,35 @@ var Availability;
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
+// Defaults to support image inputs for convenience.
+const defaultExpectedInputs = [{ type: 'image' }];
 /**
  * Defines an inference "backend" that uses Chrome's on-device model,
  * and encapsulates logic for detecting when on-device inference is
  * possible.
  */
 class ChromeAdapterImpl {
-    constructor(languageModelProvider, mode, onDeviceParams = {
-        createOptions: {
-            // Defaults to support image inputs for convenience.
-            expectedInputs: [{ type: 'image' }]
-        }
-    }) {
+    constructor(languageModelProvider, mode, onDeviceParams) {
         this.languageModelProvider = languageModelProvider;
         this.mode = mode;
-        this.onDeviceParams = onDeviceParams;
         this.isDownloading = false;
+        this.onDeviceParams = {
+            createOptions: {
+                expectedInputs: defaultExpectedInputs
+            }
+        };
+        if (onDeviceParams) {
+            this.onDeviceParams = onDeviceParams;
+            if (!this.onDeviceParams.createOptions) {
+                this.onDeviceParams.createOptions = {
+                    expectedInputs: defaultExpectedInputs
+                };
+            }
+            else if (!this.onDeviceParams.createOptions.expectedInputs) {
+                this.onDeviceParams.createOptions.expectedInputs =
+                    defaultExpectedInputs;
+            }
+        }
     }
     /**
      * Checks if a given request can be made on-device.
@@ -1596,7 +1618,7 @@ function hasValidCandidates(response) {
  * Creates an EnhancedGenerateContentResponse object that has helper functions and
  * other modifications that improve usability.
  */
-function createEnhancedContentResponse(response) {
+function createEnhancedContentResponse(response, inferenceSource = InferenceSource.IN_CLOUD) {
     /**
      * The Vertex AI backend omits default values.
      * This causes the `index` property to be omitted from the first candidate in the
@@ -1607,6 +1629,7 @@ function createEnhancedContentResponse(response) {
         response.candidates[0].index = 0;
     }
     const responseWithHelpers = addHelpers(response);
+    responseWithHelpers.inferenceSource = inferenceSource;
     return responseWithHelpers;
 }
 /**
@@ -1983,16 +2006,16 @@ const responseLineRE = /^data\: (.*)(?:\n\n|\r\r|\r\n\r\n)/;
  *
  * @param response - Response from a fetch call
  */
-function processStream(response, apiSettings) {
+function processStream(response, apiSettings, inferenceSource) {
     const inputStream = response.body.pipeThrough(new TextDecoderStream('utf8', { fatal: true }));
     const responseStream = getResponseStream(inputStream);
     const [stream1, stream2] = responseStream.tee();
     return {
-        stream: generateResponseSequence(stream1, apiSettings),
-        response: getResponsePromise(stream2, apiSettings)
+        stream: generateResponseSequence(stream1, apiSettings, inferenceSource),
+        response: getResponsePromise(stream2, apiSettings, inferenceSource)
     };
 }
-async function getResponsePromise(stream, apiSettings) {
+async function getResponsePromise(stream, apiSettings, inferenceSource) {
     const allResponses = [];
     const reader = stream.getReader();
     while (true) {
@@ -2002,12 +2025,12 @@ async function getResponsePromise(stream, apiSettings) {
             if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
                 generateContentResponse = mapGenerateContentResponse(generateContentResponse);
             }
-            return createEnhancedContentResponse(generateContentResponse);
+            return createEnhancedContentResponse(generateContentResponse, inferenceSource);
         }
         allResponses.push(value);
     }
 }
-async function* generateResponseSequence(stream, apiSettings) {
+async function* generateResponseSequence(stream, apiSettings, inferenceSource) {
     const reader = stream.getReader();
     while (true) {
         const { value, done } = await reader.read();
@@ -2016,10 +2039,10 @@ async function* generateResponseSequence(stream, apiSettings) {
         }
         let enhancedResponse;
         if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
-            enhancedResponse = createEnhancedContentResponse(mapGenerateContentResponse(value));
+            enhancedResponse = createEnhancedContentResponse(mapGenerateContentResponse(value), inferenceSource);
         }
         else {
-            enhancedResponse = createEnhancedContentResponse(value);
+            enhancedResponse = createEnhancedContentResponse(value, inferenceSource);
         }
         const firstCandidate = enhancedResponse.candidates?.[0];
         // Don't yield a response with no useful data for the developer.
@@ -2189,31 +2212,52 @@ const errorsCausingFallback = [
  */
 async function callCloudOrDevice(request, chromeAdapter, onDeviceCall, inCloudCall) {
     if (!chromeAdapter) {
-        return inCloudCall();
+        return {
+            response: await inCloudCall(),
+            inferenceSource: InferenceSource.IN_CLOUD
+        };
     }
     switch (chromeAdapter.mode) {
         case InferenceMode.ONLY_ON_DEVICE:
             if (await chromeAdapter.isAvailable(request)) {
-                return onDeviceCall();
+                return {
+                    response: await onDeviceCall(),
+                    inferenceSource: InferenceSource.ON_DEVICE
+                };
             }
             throw new AIError(AIErrorCode.UNSUPPORTED, 'Inference mode is ONLY_ON_DEVICE, but an on-device model is not available.');
         case InferenceMode.ONLY_IN_CLOUD:
-            return inCloudCall();
+            return {
+                response: await inCloudCall(),
+                inferenceSource: InferenceSource.IN_CLOUD
+            };
         case InferenceMode.PREFER_IN_CLOUD:
             try {
-                return await inCloudCall();
+                return {
+                    response: await inCloudCall(),
+                    inferenceSource: InferenceSource.IN_CLOUD
+                };
             }
             catch (e) {
                 if (e instanceof AIError && errorsCausingFallback.includes(e.code)) {
-                    return onDeviceCall();
+                    return {
+                        response: await onDeviceCall(),
+                        inferenceSource: InferenceSource.ON_DEVICE
+                    };
                 }
                 throw e;
             }
         case InferenceMode.PREFER_ON_DEVICE:
             if (await chromeAdapter.isAvailable(request)) {
-                return onDeviceCall();
+                return {
+                    response: await onDeviceCall(),
+                    inferenceSource: InferenceSource.ON_DEVICE
+                };
             }
-            return inCloudCall();
+            return {
+                response: await inCloudCall(),
+                inferenceSource: InferenceSource.IN_CLOUD
+            };
         default:
             throw new AIError(AIErrorCode.ERROR, `Unexpected infererence mode: ${chromeAdapter.mode}`);
     }
@@ -2243,8 +2287,8 @@ async function generateContentStreamOnCloud(apiSettings, model, params, requestO
     /* stream */ true, JSON.stringify(params), requestOptions);
 }
 async function generateContentStream(apiSettings, model, params, chromeAdapter, requestOptions) {
-    const response = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContentStream(params), () => generateContentStreamOnCloud(apiSettings, model, params, requestOptions));
-    return processStream(response, apiSettings); // TODO: Map streaming responses
+    const callResult = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContentStream(params), () => generateContentStreamOnCloud(apiSettings, model, params, requestOptions));
+    return processStream(callResult.response, apiSettings); // TODO: Map streaming responses
 }
 async function generateContentOnCloud(apiSettings, model, params, requestOptions) {
     if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
@@ -2254,9 +2298,9 @@ async function generateContentOnCloud(apiSettings, model, params, requestOptions
     /* stream */ false, JSON.stringify(params), requestOptions);
 }
 async function generateContent(apiSettings, model, params, chromeAdapter, requestOptions) {
-    const response = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContent(params), () => generateContentOnCloud(apiSettings, model, params, requestOptions));
-    const generateContentResponse = await processGenerateContentResponse(response, apiSettings);
-    const enhancedResponse = createEnhancedContentResponse(generateContentResponse);
+    const callResult = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContent(params), () => generateContentOnCloud(apiSettings, model, params, requestOptions));
+    const generateContentResponse = await processGenerateContentResponse(callResult.response, apiSettings);
+    const enhancedResponse = createEnhancedContentResponse(generateContentResponse, callResult.inferenceSource);
     return {
         response: enhancedResponse
     };
@@ -4138,6 +4182,7 @@ exports.ImagenModel = ImagenModel;
 exports.ImagenPersonFilterLevel = ImagenPersonFilterLevel;
 exports.ImagenSafetyFilterLevel = ImagenSafetyFilterLevel;
 exports.InferenceMode = InferenceMode;
+exports.InferenceSource = InferenceSource;
 exports.IntegerSchema = IntegerSchema;
 exports.Language = Language;
 exports.LiveGenerativeModel = LiveGenerativeModel;