updated llama31 prompts

apps/remixdesktop/src/lib/InferenceServerManager.ts

@@ -243,6 +243,23 @@ export class InferenceManager implements ICompletions {
 
   }
 
+  private async _handleExistingServer() {
+    // check if the server is already running, kill it
+    try {
+      const options = { headers: { 'Content-Type': 'application/json', } }
+      const state = await axios.get(this.inferenceURL+"/state", options)
+
+      if (state.data?.status) {
+        console.log('Found existing Inference server running')
+        this.stopInferenceServer()
+        await axios.post(this.inferenceURL+"/kill", options)
+      }
+    } catch (error) {
+      // catch connection refused
+      console.log('No existing Inference server running')
+    }
+  }
+
   private async _startServer() {
     const serverAvailable = await this._downloadInferenceServer()
     if (!serverAvailable) {
@@ -250,6 +267,9 @@ export class InferenceManager implements ICompletions {
       return
     }
 
+    // kill existing server if running
+    this._handleExistingServer()
+
     return new Promise<void>((resolve, reject) => {
       let serverPath = ""
       try {
@@ -274,7 +294,7 @@ export class InferenceManager implements ICompletions {
 
       const spawnArgs = [this.port];
 
-      console.log(`Spawning process: ${serverPath} ${spawnArgs.join(' ')}`);
+      // console.log(`Spawning process: ${serverPath} ${spawnArgs.join(' ')}`);
       this.inferenceProcess = spawn(serverPath, spawnArgs);
 
       this.inferenceProcess.stdout.on('data', (data) => {
@@ -455,11 +475,11 @@ export class InferenceManager implements ICompletions {
     }
     let modelOP = undefined
     for (const model of this.selectedModels) {
-      if (model?.modelOP) {
+      if (model.modelType === ModelType.GENERAL) {
         modelOP = model.modelOP
       }
     }
-    const prompt = buildSolgptPromt(userPrompt, this.selectedModels[0]?.modelOP)
+    const prompt = buildSolgptPromt(userPrompt, modelOP)
 
     if (GenerationParams.stream_result) {
       return this._streamInferenceRequest('solidity_answer', { prompt, ...params })
@@ -468,6 +488,4 @@ export class InferenceManager implements ICompletions {
     }
   }
 
-  // kill dangling process making use of the port
-
 }

libs/remix-ai-core/src/inferencers/remote/remoteInference.ts

@@ -10,7 +10,7 @@ export class RemoteInferencer implements ICompletions {
   api_url: string
   completion_url: string
   max_history = 7
-  model_op = RemoteBackendOPModel.DEEPSEEK
+  model_op = RemoteBackendOPModel.DEEPSEEK // default model operation change this to llama if necessary
   event: EventEmitter
 
   constructor(apiUrl?:string, completionUrl?:string) {

libs/remix-ai-core/src/prompts/promptBuilder.ts

@@ -2,7 +2,7 @@ import { RemoteBackendOPModel } from "../types/types"
 import { ChatHistory } from "./chat"
 
 export const PromptBuilder = (inst, answr, modelop) => {
-  if (modelop === RemoteBackendOPModel.CODELLAMA) return ""
+  if (modelop === RemoteBackendOPModel.CODELLAMA) return `<|start_header_id|>user<|end_header_id|>${inst}<|eot_id|><|start_header_id|>assistant<|end_header_id|> ${answr}`
   if (modelop === RemoteBackendOPModel.DEEPSEEK) return "\n### INSTRUCTION:\n" + inst + "\n### RESPONSE:\n" + answr
   if (modelop === RemoteBackendOPModel.MISTRAL) return ""
 }
@@ -22,7 +22,7 @@ export const buildSolgptPromt = (userPrompt:string, modelOP:RemoteBackendOPModel
       else newPrompt += PromptBuilder(question, answer, modelOP)
     }
     // finaly
-    newPrompt = "sol-gpt " + newPrompt + PromptBuilder(userPrompt.split('sol-gpt')[1], "", modelOP)
+    newPrompt = "sol-gpt " + newPrompt + PromptBuilder(userPrompt.split('gpt')[1], "", modelOP)
     return newPrompt
   }
 }

libs/remix-ai-core/src/types/models.ts

@@ -75,7 +75,7 @@ const GenerationParams:IParams = {
   topK: 40,
   topP: 0.92,
   max_new_tokens: 2000,
-  stream_result: true,
+  stream_result: false,
 }
 
 export { DefaultModels, CompletionParams, InsertionParams, GenerationParams }