Merge pull request #41 from Barqawiz/candidate-1.3.5

Candidate 1.3.7

- add support for llama-code and llama-python.
- upgrade the semantic search to support huge data memory search using `SemanticSearchPaging`.
- update the documentation.

IntelliNode/README.md

@@ -19,12 +19,14 @@
 IntelliNode is the ultimate tool to integrate with the latest language models and deep learning frameworks using **javascript**. The library provides intuitive functions for sending input to models like ChatGPT, WaveNet and Stable diffusion, and receiving generated text, speech, or images. With just a few lines of code, you can easily access the power of cutting-edge AI models to enhance your projects.
 
 # Latest Updates
-- Update the chatbot to support Llama v2. 🦙
+- Update the chatbot to support Llama v2 chat and code. 🦙
 - Add Gen function, the fastest way to generate text, speech, web pages or images. :bullettrain_side:
 - Update stable diffusion to use XL model engine.
 - Add support for hugging face inference.
-- Update chatGPT to support function calls, empowering automation.
 - Generate prompt using LLM.
+- Add support for huge data memory semantic search using `SemanticSearchPaging`.
+
+Join the [discord server](https://discord.gg/VYgCh2p3Ww) for the latest updates and community support.
 
 # Examples
 ## Functions
@@ -177,10 +179,11 @@ ProxyHelper.getInstance().setAzureOpenai(resourceName);
 Check the code to access the chatbot through a proxy: [proxy chatbot](https://github.com/Barqawiz/IntelliNode/blob/main/samples/command_sample/test_chatbot_proxy.js)
 
 # :closed_book: Documentation
-- [IntelliNode Wiki](https://github.com/Barqawiz/IntelliNode/wiki): Check the wiki page for indepeth instructions and practical use cases.
+- [IntelliNode Wiki](https://github.com/Barqawiz/IntelliNode/wiki): Check the wiki page for indepth instructions and practical use cases.
 - [Showcase](https://show.intellinode.ai/): Experience the potential of Intellinode in action, and use your keys to generate content and html pages.
 - [Samples](https://github.com/Barqawiz/IntelliNode/tree/main/samples/command_sample): Explore a code sample with detailed setup documentation to get started with Intellinode.
 - [Model Evaluation](https://github.com/Barqawiz/IntelliNode/wiki/Model-Evaluation): Demonstrate a swift approach to compare the performance of multiple models against designated target answers.
+- [Articles](https://www.intellinode.ai/articles): Tutorials and articles about intelliNode and data science topics.
 
 # Pillars
 - **The wrapper layer** provides low-level access to the latest AI models

IntelliNode/config.json

@@ -50,8 +50,14 @@
         "13b": "13b-chat",
         "70b-chat": "70b-chat",
         "13b-chat": "13b-chat",
+        "34b-code": "34b-code",
+        "34b-python": "34b-python",
+        "13b-code-instruct": "13b-code-instruct",
         "70b-chat-version": "2d19859030ff705a87c746f7e96eea03aefb71f166725aee39692f1476566d48",
-        "13b-chat-version": "d5da4236b006f967ceb7da037be9cfc3924b20d21fed88e1e94f19d56e2d3111"
+        "13b-chat-version": "d5da4236b006f967ceb7da037be9cfc3924b20d21fed88e1e94f19d56e2d3111",
+        "34b-code-version": "0666717e5ead8557dff55ee8f11924b5c0309f5f1ca52f64bb8eec405fdb38a7",
+        "34b-python-version": "9048743d22a7b19cd0abb018066809ea6af4f2b4717bef9aad3c5ae21ceac00d",
+        "13b-code-instruct-version":"da5676342de1a5a335b848383af297f592b816b950a43d251a0a9edd0113604b"
       }
     }
   }

IntelliNode/function/SemanticSearch.js

@@ -18,6 +18,7 @@ class SemanticSearch {
   }
 
   async getTopMatches(pivotItem, searchArray, numberOfMatches) {
+
       if (numberOfMatches > searchArray.length) {
         throw new Error('numberOfMatches should not be greater than the searchArray');
       }

IntelliNode/function/SemanticSearchPaging.js

@@ -0,0 +1,37 @@
+const { SemanticSearch } = require('./SemanticSearch'); // assuming path
+
+class SemanticSearchPaging extends SemanticSearch {
+  constructor(keyValue, provider, pivotItem, numberOfMatches) {
+    super(keyValue, provider);
+    this.pivotItem = pivotItem;
+    this.numberOfMatches = numberOfMatches;
+    this.textAndMatches = []; // To store { text: '...', similarity: 0.9 } results
+    this.topMatches = [];
+  }
+
+  async addNewData(newSearchItems) {
+      // get the best matches for new items
+      const newMatches = await super.getTopMatches(this.pivotItem, newSearchItems, newSearchItems.length);
+
+      // map the matches format
+      const newMatchesWithText = newMatches.map(match => ({
+        text: newSearchItems[match.index],
+        score: match.similarity,
+      }));
+
+      // combine with old top matches and sort
+      this.topMatches = [...this.topMatches, ...newMatchesWithText]
+        .sort((a, b) => b.score - a.score)
+        .slice(0, this.numberOfMatches);
+}
+
+  getCurrentTopMatches() {
+    return this.topMatches;
+  }
+
+  clean() {
+    this.topMatches = [];
+  }
+}
+
+module.exports = { SemanticSearchPaging };

IntelliNode/index.js

@@ -6,6 +6,7 @@ const { RemoteEmbedModel, SupportedEmbedModels } = require('./controller/RemoteE
 // functions
 const { Chatbot, SupportedChatModels } = require('./function/Chatbot');
 const { SemanticSearch } = require('./function/SemanticSearch');
+const { SemanticSearchPaging } = require('./function/SemanticSearchPaging');
 const { TextAnalyzer } = require('./function/TextAnalyzer');
 const { Gen } = require('./function/Gen');
 
@@ -72,5 +73,6 @@ module.exports = {
   AWSEndpointWrapper,
   Prompt,
   LLamaSageInput,
-  LLMEvaluation
+  LLMEvaluation,
+  SemanticSearchPaging
 };

IntelliNode/model/input/ChatModelInput.js

@@ -175,6 +175,11 @@ class LLamaReplicateInput extends ChatLLamaInput {
     options.model = options.model || config.getProperty('models.replicate.llama.13b');
     options.version = options.version;
     super(systemMessage, options);
+    this.top_k = options.top_k || null;
+    this.top_p = options.top_p || null;
+    this.min_new_tokens = options.min_new_tokens || null;
+    this.system_prompt = options.system_prompt || null;
+    this.repetition_penalty = options.repetition_penalty || null;
   }
 
   getChatInput() {
@@ -183,21 +188,26 @@ class LLamaReplicateInput extends ChatLLamaInput {
         this.version = config.getProperty(`models.replicate.llama.${this.model}-version`);
     }
 
-    return {
+    var myData = {
       model: this.model,
       inputData: {
           version: this.version,
           input: {
             prompt: this.prompt,
-            system_prompt: this.system_prompt,
             max_new_tokens: this.max_new_tokens,
             temperature: this.temperature,
-            top_p: this.top_p,
-            repetition_penalty: this.repetition_penalty,
             debug: this.debug
           }
       }
     };
+
+    if (this.top_k) myData.inputData.input.top_k = this.top_k;
+    if (this.top_p) myData.inputData.input.top_p = this.top_p;
+    if (this.system_prompt) myData.inputData.input.system_prompt = this.system_prompt;
+    if (this.min_new_tokens) myData.inputData.input.min_new_tokens = this.min_new_tokens;
+    if (this.repetition_penalty) myData.inputData.input.repetition_penalty = this.repetition_penalty;
+
+    return myData;
   }
 }
 

IntelliNode/package.json

@@ -1,6 +1,6 @@
 {
   "name": "intellinode",
-  "version": "1.3.4",
+  "version": "1.3.7",
   "description": "Access various AI models, such as ChatGPT, Llama, Diffusion, Cohere, WaveNet, and others",
   "main": "index.js",
   "keywords": [
@@ -17,7 +17,7 @@
     "prompt",
     "automation"
   ],
-  "author": "Intellinode",
+  "author": "IntelliNode",
   "license": "Apache",
   "repository": {
     "type": "git",

IntelliNode/test/integration/Chatbot.test.js

@@ -121,25 +121,6 @@ async function testReplicateLLamaCase1() {
   }
 }
 
-
-async function testReplicateLLamaCase2() {
-  try {
-    console.log('\nLLama test case 2: \n')
-    const input = new LLamaReplicateInput("you are helpful assistant!");
-    input.addUserMessage("Explain the plot of the Inception movie");
-    input.addAssistantMessage("The plot of the movie Inception follows a skilled thief who enters people's dreams to steal their secrets and is tasked with implanting an idea into a target's mind to alter their future actions.");
-    input.addUserMessage("Explain the plot of the dark night movie");
-
-    const responses = await replicateBot.chat(input);
-
-    responses.forEach((response) => console.log("- " + response));
-
-    assert(responses.length > 0, "testReplicateLLamaCase1 response length should be greater than 0");
-  } catch (error) {
-    console.error("Test case failed with exception:", error.message);
-  }
-}
-
 async function testSageMakerLLamaCase() {
     try {
     console.log('\nLLama sagemaker test case 1: \n')
@@ -161,6 +142,23 @@ async function testSageMakerLLamaCase() {
 
 }
 
+async function testReplicateLLamaCase3() {
+  try {
+    console.log('\nLLama test case 3: \n')
+    const input = new LLamaReplicateInput("you are helpful coding assistant!",
+                                          {model: '34b-code'});
+    input.addUserMessage("how to develop micro service using node js");
+
+    const responses = await replicateBot.chat(input);
+
+    responses.forEach((response) => console.log("- " + response));
+
+    assert(responses.length > 0, "testReplicateLLamaCase1 response length should be greater than 0");
+  } catch (error) {
+    console.error("Test case failed with exception:", error.message);
+  }
+}
+
 (async () => {
   console.log('### Openai model ###')
   await testOpenaiChatGPTCase1();
@@ -170,8 +168,9 @@ async function testSageMakerLLamaCase() {
   console.log('### Replicate llama model ###')
   await testReplicateLLamaCase1();
   await testReplicateLLamaCase2();
+  await testReplicateLLamaCase3();
 
   console.log('### SageMaker llama model ###')
-  await testSageMakerLLamaCase();
+  //await testSageMakerLLamaCase();
 
 })();

IntelliNode/test/integration/LLamav2.test.js

@@ -42,7 +42,51 @@ async function testReplicateWrapperLLama() {
   }
 }
 
+
+async function testReplicateLLamaCoder() {
+  try {
+
+    const modelName = config.getProperty('models.replicate.llama.34b-python');
+    const version = config.getProperty('models.replicate.llama.34b-python-version');
+    const inputData = { version: version,
+                input: { prompt: '# function that adds 2 number inputs.',
+                        max_new_tokens: 128,
+                        top_k: 50,
+                        top_p: 0.9,
+                        temperature: 0.1,
+                        min_new_tokens: -1,
+                        debug: false,
+                        stop_sequences: '<end>'} };
+
+    const prediction = await replicateWrapper.predict(modelName, inputData);
+
+    // check for the response every second
+    const poll = setInterval(async () => {
+      const status = await replicateWrapper.getPredictionStatus(prediction.id);
+
+      console.log('the current status: ', status.status)
+
+      if (status.status === 'succeeded' || status.status === 'failed') {
+        clearInterval(poll); // stop polling if prediction has completed or failed
+
+        if (status.status === 'succeeded') {
+          console.log('LLama Predict Result:', status.output.join(''));
+        } else {
+          console.error('LLama Prediction Failed:', status.error);
+        }
+      }
+    }, 1000);
+
+  } catch (error) {
+    console.error('LLama Error:', error);
+  }
+}
+
 (async () => {
   // test LLama v2 from Replicate host
   await testReplicateWrapperLLama();
+
+  // test LLama v2 coder
+  await testReplicateLLamaCoder();
+
 })();

IntelliNode/test/integration/SemanticSearchPaging.test.js

@@ -0,0 +1,38 @@
+require('dotenv').config();
+const assert = require('assert');
+const { SemanticSearchPaging } = require('../../function/SemanticSearchPaging');
+const { SupportedEmbedModels } = require('../../controller/RemoteEmbedModel');
+
+const openaiApiKey = process.env.OPENAI_API_KEY;
+const cohereApiKey = process.env.COHERE_API_KEY;
+
+const pivotItem = 'Hello from IntelliNode';
+
+const openaiSemanticSearch = new SemanticSearchPaging(openaiApiKey,
+                                                     SupportedEmbedModels.OPENAI,
+                                                     pivotItem, 2);
+const cohereSemanticSearch = new SemanticSearchPaging(cohereApiKey,
+                                                      SupportedEmbedModels.COHERE,
+                                                      pivotItem, 2);
+
+async function addToSessionAndTest(semanticSearch, newSearchItems) {
+
+  await semanticSearch.addNewData( newSearchItems);
+  const results = semanticSearch.getCurrentTopMatches();
+
+  console.log('Semantic Search Session Results:', results);
+  assert(results.length <= semanticSearch.numberOfMatches, 'Test passed');
+}
+
+(async () => {
+
+  // semantic search with openai embedding
+  await addToSessionAndTest(openaiSemanticSearch,  ['Greetings from IntelliNode!', 'Saluti da IntelliNode!']);
+  await addToSessionAndTest(openaiSemanticSearch, ['Hola desde IntelliNode!', 'Bonjour de IntelliNode!']);
+
+  openaiSemanticSearch.clean();
+
+  // semantic search with cohere embedding
+  await addToSessionAndTest(cohereSemanticSearch,  ['Greetings from IntelliNode!', 'Bonjour de IntelliNode!']);
+  await addToSessionAndTest(cohereSemanticSearch, ['Hola desde IntelliNode!', 'Saluti da IntelliNode!']);
+})();

IntelliNode/wrappers/ReplicateWrapper.js

@@ -27,7 +27,7 @@ class ReplicateWrapper {
     });
   }
 
-  async predict(modelVersion, inputData) {
+  async predict(modelTag, inputData) {
     const url = config.getProperty('url.replicate.predictions');
     const data = inputData;
 

samples/command_sample/README.md

@@ -25,7 +25,7 @@ The Llama model is made available by several hosting services, and Replicate is
 
 ## Samples Execution
 
-1. E-commerce sample to generate products description and images:
+1. E-commerce sample to product descriptions description and images:
 `node ecommerce_tool.js`
 
 2. Language model using openai and cohere:
@@ -46,22 +46,25 @@ The Llama model is made available by several hosting services, and Replicate is
 7. Semantic search:
 `node test_semantic_search.js`
 
-8. Text analyzer (summary & sentiment analysis):
+8. Semantic search pagination to work with huge data:
+`node test_semantic_search.js`
+
+9. Text analyzer (summary & sentiment analysis):
 `node test_text_analyzer.js`
 
-9. Huggingface simplified inference access:
+10. Huggingface simplified inference access:
 `node test_hugging_face.js`
 
-10. Azure openai sample
+11. Azure openai sample
 `node test_azure_chatbot.js <resource_name> <deployment_name>`
 
-11. Automation sample using the chatbot function call:
+12. Automation sample using the chatbot code call, it works by providing the model with your function details, and it decides to execute the code based on the user conversation:
 `node automate_s3_bucket.js`
 
-12. Llama V2 chatbot:
+13. Llama V2 chatbot:
 `node test_llama_chatbot.js`
 
-13. LLM evaluator to compare models like chatGPT, Cohere, and Llama:
+14. LLM evaluator to compare models like chatGPT, Cohere, and Llama:
 `node test_llm_evaluation.test.js`