Examples, langchain and release (#95)

Signed-off-by: Tomas Pilar <tomas.pilar@ibm.com>

.github/workflows/node.js.yml

@@ -50,6 +50,27 @@ jobs:
       - run: yarn --frozen-lockfile
       - run: yarn test
 
+  examples:
+    runs-on: ubuntu-latest
+
+    env:
+      GENAI_API_KEY: ${{ secrets.TEST_API_KEY }}
+
+    strategy:
+      matrix:
+        node-version: [18.18.x]
+        # See supported Node.js release schedule at https://nodejs.org/en/about/releases/
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Use Node.js ${{ matrix.node-version }}
+        uses: actions/setup-node@v4
+        with:
+          node-version: ${{ matrix.node-version }}
+          cache: 'yarn'
+      - run: yarn --frozen-lockfile
+      - run: yarn examples
+
   build:
     runs-on: ubuntu-latest
 

README.md

@@ -28,7 +28,7 @@ The SDK supports both TypeScript and JavaScript as well as ESM and CommonJS.
 
 ## Key features
 
-- ⚡️ Performant - processes 1k of short inputs in about 4 minutes
+- ⚡️ Performant - processes 1k of short inputs in under a minute
 - ☀️ Fault-tolerant - retry strategies and overflood protection
 - 🚦 Handles concurrency limiting - even if you have multiple parallel jobs running
 - 📌 Aligned with the REST API - clear structure that mirrors service endpoints and data
@@ -126,14 +126,15 @@ Standalone API reference is NOT available at the moment, please refer to the [RE
 The following example showcases how you can integrate GenAI into your project.
 
 ```typescript
+import { Client } from '@ibm-generative-ai/node-sdk';
 import { GenAIModel } from '@ibm-generative-ai/node-sdk/langchain';
 
 const model = new GenAIModel({
   modelId: 'google/flan-ul2',
   parameters: {},
-  configuration: {
+  client: new Client({
     apiKey: 'pak-.....',
-  },
+  }),
 });
 ```
 
@@ -171,15 +172,16 @@ console.log(text); // ArcticAegis
 ### Streaming
 
 ```typescript
+import { Client } from '@ibm-generative-ai/node-sdk';
 import { GenAIModel } from '@ibm-generative-ai/node-sdk/langchain';
 
 const model = new GenAIModel({
   modelId: 'google/flan-ul2',
   stream: true,
   parameters: {},
-  configuration: {
+  client: new Client({
     apiKey: 'pak-.....',
-  },
+  }),
 });
 
 await model.invoke('Tell me a joke.', {
@@ -196,15 +198,16 @@ await model.invoke('Tell me a joke.', {
 ### Chat support
 
 ```typescript
+import { Client } from '@ibm-generative-ai/node-sdk';
 import { GenAIChatModel } from '@ibm-generative-ai/node-sdk/langchain';
 import { SystemMessage, HumanMessage } from '@langchain/core/messages';
 
 const client = new GenAIChatModel({
   model_id: 'meta-llama/llama-2-70b-chat',
-  configuration: {
+  client: new Client({
     endpoint: process.env.ENDPOINT,
     apiKey: process.env.API_KEY,
-  },
+  }),
   parameters: {
     decoding_method: 'greedy',
     min_new_tokens: 10,

examples/chat.ts

@@ -1,6 +1,6 @@
 import { Client } from '../src/index.js';
 
-import { CHAT_MODEL } from './constants.js';
+import { CHAT_MODEL } from './shared/constants.js';
 
 const client = new Client({
   apiKey: process.env.GENAI_API_KEY,

examples/generate.ts

@@ -1,6 +1,6 @@
 import { Client } from '../src/index.js';
 
-import { MODEL } from './constants.js';
+import { MODEL } from './shared/constants.js';
 
 const client = new Client({
   apiKey: process.env.GENAI_API_KEY,

examples/history.ts

@@ -1,6 +1,6 @@
 import { Client } from '../src/index.js';
 
-import { CHAT_MODEL } from './constants.js';
+import { CHAT_MODEL } from './shared/constants.js';
 
 const client = new Client({
   apiKey: process.env.GENAI_API_KEY,
@@ -24,7 +24,7 @@ const client = new Client({
     messages: [{ role: 'user', content: 'How are you?' }],
   });
   const { results } = await client.request.chat({
-    conversationId: conversation_id,
+    conversation_id,
   });
   for (const request of results) {
     console.log(request);

examples/langchain/llm-chat.ts

@@ -1,14 +1,15 @@
 import { HumanMessage } from '@langchain/core/messages';
 
 import { GenAIChatModel } from '../../src/langchain/llm-chat.js';
+import { Client } from '../../src/index.js';
 
 const makeClient = () =>
   new GenAIChatModel({
     model_id: 'meta-llama/llama-2-70b-chat',
-    configuration: {
+    client: new Client({
       endpoint: process.env.ENDPOINT,
       apiKey: process.env.API_KEY,
-    },
+    }),
     parameters: {
       decoding_method: 'greedy',
       min_new_tokens: 1,

examples/langchain/llm.ts

@@ -1,13 +1,14 @@
+import { Client } from '../../src/index.js';
 import { GenAIModel } from '../../src/langchain/index.js';
 
 const makeClient = (stream?: boolean) =>
   new GenAIModel({
     modelId: 'google/flan-t5-xl',
     stream,
-    configuration: {
+    client: new Client({
       endpoint: process.env.ENDPOINT,
       apiKey: process.env.API_KEY,
-    },
+    }),
     parameters: {
       decoding_method: 'greedy',
       min_new_tokens: 5,

examples/models.ts

@@ -1,6 +1,6 @@
 import { Client } from '../src/index.js';
 
-import { MODEL } from './constants.js';
+import { MODEL } from './shared/constants.js';
 
 const client = new Client({
   apiKey: process.env.GENAI_API_KEY,

examples/tune.ts

@@ -1,3 +1,6 @@
+import { blob } from 'node:stream/consumers';
+import { createReadStream } from 'node:fs';
+
 import { Client } from '../src/index.js';
 
 const client = new Client({
@@ -26,13 +29,24 @@ const client = new Client({
   const { results: tuneTypes } = await client.tune.types({});
   console.log(tuneTypes);
 
+  // Upload file for tuning
+  const { result: file } = await client.file.create({
+    purpose: 'tune',
+    file: {
+      name: 'tune_input.jsonl',
+      content: (await blob(
+        createReadStream('examples/assets/tune_input.jsonl'),
+      )) as any,
+    },
+  });
+
   // Create a tune
   const { result: createdTune } = await client.tune.create({
     name: 'Awesome Tune',
     tuning_type: 'prompt_tuning',
     model_id: 'google/flan-t5-xl',
     task_id: 'generation',
-    training_file_ids: ['fileId'],
+    training_file_ids: [file.id],
   });
   console.log(createdTune);
 
@@ -50,4 +64,7 @@ const client = new Client({
 
   // Delete the tune
   await client.tune.delete({ id: createdTune.id });
+
+  // Detele the file
+  await client.file.delete({ id: file.id });
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ibm-generative-ai/node-sdk",
-  "version": "2.0.2",
+  "version": "2.0.3",
   "description": "IBM Generative AI Node.js SDK (Tech Preview)",
   "keywords": [
     "ai",
@@ -58,13 +58,7 @@
     "postpack": "pinst --enable",
     "generate": "./scripts/generate.sh",
     "generate:new": "node ./scripts/generate.js",
-    "example:run": "ts-node -r dotenv-flow/config",
-    "example:generate": "yarn run example:run examples/generate.ts",
-    "example:tune": "yarn run example:run examples/tune.ts",
-    "example:history": "yarn run example:run examples/history.ts",
-    "example:file": "yarn run example:run examples/file.ts",
-    "example:chat": "yarn run example:run examples/chat.ts",
-    "example:models": "yarn run example:run examples/models.ts"
+    "examples": "./scripts/examples.sh"
   },
   "peerDependencies": {
     "@langchain/core": ">=0.1.0"

scripts/examples.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+set -e
+
+# Run all modules in examples/ directory
+for file in examples/* examples/langchain/*; do 
+    if [ -f "$file" ]; then
+        echo "Running example $file"
+        npx ts-node -r dotenv-flow/config "$file" > /dev/null
+    fi 
+done

src/langchain/llm-chat.ts

@@ -21,10 +21,12 @@ type TextChatInput = TextChatCreateInput & TextChatCreateStreamInput;
 export type GenAIChatModelParams = BaseChatModelParams &
   Omit<TextChatInput, 'messages' | 'prompt_template_id'> & {
     model_id: NonNullable<TextChatInput['model_id']>;
-    configuration?: Configuration;
-  };
+  } & (
+    | { client: Client; configuration?: never }
+    | { client?: never; configuration: Configuration }
+  );
 export type GenAIChatModelOptions = BaseChatModelCallOptions &
-  Partial<Omit<GenAIChatModelParams, 'configuration'>>;
+  Partial<Omit<GenAIChatModelParams, 'client' | 'configuration'>>;
 
 export class GenAIChatModel extends BaseChatModel<GenAIChatModelOptions> {
   protected readonly client: Client;
@@ -47,6 +49,7 @@ export class GenAIChatModel extends BaseChatModel<GenAIChatModelOptions> {
     parent_id,
     use_conversation_parameters,
     trim_method,
+    client,
     configuration,
     ...options
   }: GenAIChatModelParams) {
@@ -60,7 +63,7 @@ export class GenAIChatModel extends BaseChatModel<GenAIChatModelOptions> {
     this.parentId = parent_id;
     this.useConversationParameters = use_conversation_parameters;
     this.trimMethod = trim_method;
-    this.client = new Client(configuration);
+    this.client = client ?? new Client(configuration);
   }
 
   async _generate(

src/langchain/llm.ts

@@ -15,12 +15,14 @@ import {
   TextGenerationCreateOutput,
 } from '../schema.js';
 
-interface BaseGenAIModelOptions {
+type BaseGenAIModelOptions = {
   stream?: boolean;
   parameters?: Record<string, any>;
   timeout?: number;
-  configuration?: Configuration;
-}
+} & (
+  | { client: Client; configuration?: never }
+  | { client?: never; configuration: Configuration }
+);
 
 export type GenAIModelOptions =
   | (BaseGenAIModelOptions & { modelId?: string; promptId?: never })
@@ -41,6 +43,7 @@ export class GenAIModel extends BaseLLM {
     stream = false,
     parameters,
     timeout,
+    client,
     configuration,
     ...baseParams
   }: GenAIModelOptions & BaseLLMParams) {
@@ -51,7 +54,7 @@ export class GenAIModel extends BaseLLM {
     this.timeout = timeout;
     this.isStreaming = Boolean(stream);
     this.parameters = parameters || {};
-    this.#client = new Client(configuration);
+    this.#client = client ?? new Client(configuration);
   }
 
   #createPayload(

tests/e2e/client.test.ts

@@ -61,7 +61,8 @@ describe('client', () => {
         ]);
       };
 
-      test('should correctly process moderation chunks during streaming', async () => {
+      // TODO remove skip after server bug is fixed or when schema is updated
+      test.skip('should correctly process moderation chunks during streaming', async () => {
         const stream = await makeValidStream({
           min_new_tokens: 1,
           max_new_tokens: 5,