test: add tests

javascript/examples/model.ts

@@ -29,15 +29,39 @@ export const openaiAudioModel = new OpenAIModel(
     },
   },
 );
-export const anthropicModel = new AnthropicModel({
-  modelId: "claude-3-5-sonnet-20241022",
-  apiKey: process.env["ANTHROPIC_API_KEY"] as string,
-});
-export const googleModel = new GoogleModel({
-  modelId: "gemini-1.5-pro",
-  apiKey: process.env["GOOGLE_API_KEY"] as string,
-});
-export const cohereModel = new CohereModel({
-  modelId: "command-r-08-2024",
-  apiKey: process.env["CO_API_KEY"] as string,
-});
+export const anthropicModel = new AnthropicModel(
+  {
+    modelId: "claude-3-5-sonnet-20241022",
+    apiKey: process.env["ANTHROPIC_API_KEY"] as string,
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 3.0 / 1_000_000,
+      outputCostPerTextToken: 15.0 / 1_000_000,
+    },
+  },
+);
+export const googleModel = new GoogleModel(
+  {
+    modelId: "gemini-1.5-pro",
+    apiKey: process.env["GOOGLE_API_KEY"] as string,
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 1.25 / 1_000_000,
+      outputCostPerTextToken: 5.0 / 1_000_000,
+    },
+  },
+);
+export const cohereModel = new CohereModel(
+  {
+    modelId: "command-r-08-2024",
+    apiKey: process.env["CO_API_KEY"] as string,
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 0.16 / 1_000_000,
+      outputCostPerTextToken: 0.6 / 1_000_000,
+    },
+  },
+);

javascript/src/anthropic/anthropic.ts

@@ -141,7 +141,7 @@ export function convertToAnthropicMessages(
   options: AnthropicModelOptions,
 ): Anthropic.Messages.MessageParam[] {
   if (options.convertAudioPartsToTextParts) {
-    messages = messages.map((message) => convertAudioPartsToTextParts(message));
+    messages = messages.map(convertAudioPartsToTextParts);
   }
 
   return messages.map((message): Anthropic.Messages.MessageParam => {

javascript/src/cohere/cohere.ts

@@ -14,6 +14,7 @@ import {
   Tool,
   ToolCallPart,
 } from "../schemas/types.gen.js";
+import { convertAudioPartsToTextParts } from "../utils/message.utils.js";
 import { ContentDeltaAccumulator } from "../utils/stream.utils.js";
 import { calculateCost } from "../utils/usage.utils.js";
 import { CohereModelOptions } from "./types.js";
@@ -114,7 +115,7 @@ export function convertToCohereParams(
 
   return {
     model: options.modelId,
-    messages: convertToCohereMessages(input),
+    messages: convertToCohereMessages(input, options),
     ...(input.tools && { tools: input.tools.map(convertToCohereTool) }),
     ...samplingParams,
     ...(response_format && {
@@ -126,10 +127,14 @@ export function convertToCohereParams(
 
 export function convertToCohereMessages(
   input: Pick<CohereLanguageModelInput, "messages" | "systemPrompt">,
+  options: CohereModelOptions,
 ): Cohere.ChatMessageV2[] {
   const cohereMessages: Cohere.ChatMessageV2[] = [];
 
-  const messages = input.messages;
+  let messages = input.messages;
+  if (options.convertAudioPartsToTextParts) {
+    messages = messages.map(convertAudioPartsToTextParts);
+  }
 
   if (input.systemPrompt) {
     cohereMessages.push({

javascript/src/cohere/types.ts

@@ -1,4 +1,8 @@
 export interface CohereModelOptions {
   apiKey: string;
   modelId: string;
+  /**
+   * If the AudioPart has a transcript, convert it to an TextPart
+   */
+  convertAudioPartsToTextParts?: boolean;
 }

javascript/src/utils/stream.utils.ts

@@ -109,7 +109,7 @@ export class ContentDeltaAccumulator {
         case "tool-call":
           if (!delta.part.toolCallId || !delta.part.toolName) {
             throw new Error(
-              `missing toolCallId or toolName at index ${String(delta.index)}. toolCallId: ${delta.part.toolCallId || "undefined"}, toolName: ${delta.part.toolName || "undefined"}`,
+              `missing toolCallId or toolName at index ${String(delta.index)}. toolCallId: ${String(delta.part.toolCallId)}, toolName: ${String(delta.part.toolName)}`,
             );
           }
           return {
@@ -123,7 +123,7 @@ export class ContentDeltaAccumulator {
         case "audio": {
           if (delta.part.encoding !== "linear16") {
             throw new Error(
-              `only linear16 encoding is supported for audio concatenation. encoding: ${delta.part.encoding || "undefined"}`,
+              `only linear16 encoding is supported for audio concatenation. encoding: ${String(delta.part.encoding)}`,
             );
           }
           const concatenatedAudioData = mergeInt16Arrays(delta.part.audioData);

javascript/test/anthropic.test.ts

@@ -1,16 +1,67 @@
 /* eslint-disable @typescript-eslint/no-floating-promises */
-import { describe, it } from "node:test";
+import test, { suite } from "node:test";
 import { AnthropicModel } from "../src/anthropic/anthropic.js";
-import { getLanguageModelTests } from "./test-language-model.js";
+import { log, testLanguageModel } from "./test-language-model.js";
 
-const model = new AnthropicModel({
-  apiKey: process.env["ANTHROPIC_API_KEY"] as string,
-  modelId: "claude-3-5-sonnet-20241022",
-});
+const model = new AnthropicModel(
+  {
+    apiKey: process.env["ANTHROPIC_API_KEY"] as string,
+    modelId: "claude-3-5-sonnet-20241022",
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 3.0 / 1_000_000,
+      outputCostPerTextToken: 15.0 / 1_000_000,
+    },
+  },
+);
+
+suite("AnthropicModel", () => {
+  testLanguageModel(model);
+
+  test("convert audio part to text part if enabled", async () => {
+    const model = new AnthropicModel({
+      apiKey: process.env["ANTHROPIC_API_KEY"] as string,
+      // not an audio model
+      modelId: "claude-3-5-sonnet-20241022",
+      convertAudioPartsToTextParts: true,
+    });
+
+    const response = await model.generate({
+      messages: [
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Hello",
+            },
+          ],
+        },
+        {
+          role: "assistant",
+          content: [
+            {
+              type: "audio",
+              audioData: "",
+              transcript: "Hi there, how can I help you?",
+            },
+          ],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Goodbye",
+            },
+          ],
+        },
+      ],
+    });
+
+    log(response);
 
-describe("AnthropicModel", () => {
-  const tests = getLanguageModelTests(model);
-  tests.forEach(({ name, fn }) => {
-    it(name, fn);
+    // it should not throw a part unsupported error
   });
 });

javascript/test/cohere.test.ts

@@ -1,16 +1,67 @@
 /* eslint-disable @typescript-eslint/no-floating-promises */
-import { describe, it } from "node:test";
+import test, { suite } from "node:test";
 import { CohereModel } from "../src/cohere/cohere.js";
-import { getLanguageModelTests } from "./test-language-model.js";
+import { log, testLanguageModel } from "./test-language-model.js";
 
-const model = new CohereModel({
-  apiKey: process.env["CO_API_KEY"] as string,
-  modelId: "command-r-08-2024",
-});
+const model = new CohereModel(
+  {
+    apiKey: process.env["CO_API_KEY"] as string,
+    modelId: "command-r-08-2024",
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 0.16 / 1_000_000,
+      outputCostPerTextToken: 0.6 / 1_000_000,
+    },
+  },
+);
+
+suite("CohereModel", () => {
+  testLanguageModel(model);
+
+  test("convert audio part to text part if enabled", async () => {
+    const model = new CohereModel({
+      apiKey: process.env["CO_API_KEY"] as string,
+      // not an audio model
+      modelId: "command-r-08-2024",
+      convertAudioPartsToTextParts: true,
+    });
+
+    const response = await model.generate({
+      messages: [
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Hello",
+            },
+          ],
+        },
+        {
+          role: "assistant",
+          content: [
+            {
+              type: "audio",
+              audioData: "",
+              transcript: "Hi there, how can I help you?",
+            },
+          ],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Goodbye",
+            },
+          ],
+        },
+      ],
+    });
+
+    log(response);
 
-describe("CohereModel", () => {
-  const tests = getLanguageModelTests(model);
-  tests.forEach(({ name, fn }) => {
-    it(name, fn);
+    // it should not throw a part unsupported error
   });
 });

javascript/test/google.test.ts

@@ -0,0 +1,67 @@
+/* eslint-disable @typescript-eslint/no-floating-promises */
+import test, { suite } from "node:test";
+import { GoogleModel } from "../src/google/google.js";
+import { log, testLanguageModel } from "./test-language-model.js";
+
+const model = new GoogleModel(
+  {
+    apiKey: process.env["GOOGLE_API_KEY"] as string,
+    modelId: "gemini-1.5-pro",
+  },
+  {
+    pricing: {
+      inputCostPerTextToken: 1.25 / 1_000_000,
+      outputCostPerTextToken: 5.0 / 1_000_000,
+    },
+  },
+);
+
+suite("GoogleModel", () => {
+  testLanguageModel(model);
+
+  test("convert audio part to text part if enabled", async () => {
+    const model = new GoogleModel({
+      apiKey: process.env["GOOGLE_API_KEY"] as string,
+      // not an audio model
+      modelId: "gemini-1.5-pro",
+      convertAudioPartsToTextParts: true,
+    });
+
+    const response = await model.generate({
+      messages: [
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Hello",
+            },
+          ],
+        },
+        {
+          role: "assistant",
+          content: [
+            {
+              type: "audio",
+              audioData: "",
+              transcript: "Hi there, how can I help you?",
+            },
+          ],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "text",
+              text: "Goodbye",
+            },
+          ],
+        },
+      ],
+    });
+
+    log(response);
+
+    // it should not throw a part unsupported error
+  });
+});