feat(configuration): improve configuration to support settings using GPU

package.json

@@ -107,6 +107,11 @@
           "default": false,
           "description": "Enable experimental chat feature."
         },
+        "firecoder.experimental.chat.useGpu": {
+          "type": "boolean",
+          "default": false,
+          "description": "Enable experimental chat feature."
+        },
         "firecoder.completion.autoMode": {
           "type": "string",
           "default": "base-small",
@@ -117,25 +122,35 @@
             "base-large"
           ],
           "enumDescriptions": [
-            "Use if you have only CPU.",
-            "Use if you have standard GPU.",
-            "Use if you have enterprise GPU."
+            "Use if you have only CPU. (RAM OR VRAM required 4 GB)",
+            "Use if you have standard GPU. (RAM OR VRAM required 10 GB)",
+            "Use if you have enterprise GPU. (RAM OR VRAM required 38 GB)"
           ]
         },
-        "firecoder.completion.manuallyMode": {
+        "firecoder.completion.autoMode.useGpu": {
+          "type": "boolean",
+          "default": false,
+          "description": "Enable experimental chat feature."
+        },
+        "firecoder.completion.manualMode": {
           "type": "string",
           "default": "base-small",
-          "markdownDescription": "Select a model for the manually mode based on your system specifications.",
+          "markdownDescription": "Select a model for the manual mode based on your system specifications.",
           "enum": [
             "base-small",
             "base-medium",
             "base-large"
           ],
           "enumDescriptions": [
-            "Use if you have only CPU.",
-            "Use if you have standard GPU.",
-            "Use if you have enterprise GPU."
+            "Use if you have only CPU. (RAM OR VRAM required 4 GB)",
+            "Use if you have standard GPU. (RAM OR VRAM required 10 GB)",
+            "Use if you have enterprise GPU. (RAM OR VRAM required 38 GB)"
           ]
+        },
+        "firecoder.completion.manualMode.useGpu": {
+          "type": "boolean",
+          "default": false,
+          "description": "Enable experimental chat feature."
         }
       }
     }

src/common/completion/index.ts

@@ -54,7 +54,7 @@ export const getInlineCompletionProvider = (
 
       const modelType = triggerAuto
         ? configuration.get("completion.autoMode")
-        : configuration.get("completion.manuallyMode");
+        : configuration.get("completion.manualMode");
       const prompt = await getPromptCompletion(
         document,
         position,

src/common/server/index.ts

@@ -105,16 +105,30 @@ class Server {
       component: "server",
       sendTelemetry: true,
     });
+    const isChatModel = this.typeModel in modelsChat;
+    const isBaseModel = this.typeModel in modelsBase;
 
-    const useGPU =
+    const gpuEnabled =
       configuration.get("experimental.useGpu.linux.nvidia") ||
       configuration.get("experimental.useGpu.osx.metal") ||
       configuration.get("experimental.useGpu.windows.nvidia");
 
-    const port = models[this.typeModel].port;
+    const useGPUChat =
+      configuration.get("experimental.chat.useGpu") && isChatModel;
 
-    const isChatModel = this.typeModel in modelsChat;
-    const isBaseModel = this.typeModel in modelsBase;
+    const useGPUCompletionAuto =
+      configuration.get("completion.autoMode.useGpu") &&
+      this.typeModel === configuration.get("completion.autoMode");
+
+    const useGPUCompletionManual =
+      configuration.get("completion.manualMode.useGpu") &&
+      this.typeModel === configuration.get("completion.manualMode");
+
+    const useGpu =
+      gpuEnabled &&
+      (useGPUCompletionAuto || useGPUCompletionManual || useGPUChat);
+
+    const port = models[this.typeModel].port;
 
     this.serverProcess = spawn(
       serverPath,
@@ -126,7 +140,7 @@ class Server {
         ...(isChatModel ? ["--ctx-size", "16384"] : ["--ctx-size", "4096"]),
         ...(isBaseModel ? ["--parallel", "4"] : []),
         ...(isMacArm64 ? ["--nobrowser"] : []),
-        ...(useGPU ? ["--n-gpu-layers", "100"] : []),
+        ...(useGpu ? ["--n-gpu-layers", "100"] : []),
         "--cont-batching",
         "--embedding",
         "--log-disable",

src/common/utils/configuration.ts

@@ -14,12 +14,21 @@ const ConfigurationProperties = {
   "experimental.chat": {
     default: false,
   },
+  "experimental.chat.useGpu": {
+    default: false,
+  },
   "completion.autoMode": {
     default: "base-small",
   },
-  "completion.manuallyMode": {
+  "completion.autoMode.useGpu": {
+    default: false,
+  },
+  "completion.manualMode": {
     default: "base-small",
   },
+  "completion.manualMode.useGpu": {
+    default: false,
+  },
 } as const;
 
 interface ConfigurationPropertiesType
@@ -36,12 +45,21 @@ interface ConfigurationPropertiesType
   "experimental.chat": {
     possibleValues: boolean;
   };
+  "experimental.chat.useGpu": {
+    possibleValues: boolean;
+  };
   "completion.autoMode": {
     possibleValues: TypeModelsBase;
   };
-  "completion.manuallyMode": {
+  "completion.autoMode.useGpu": {
+    possibleValues: boolean;
+  };
+  "completion.manualMode": {
     possibleValues: TypeModelsBase;
   };
+  "completion.manualMode.useGpu": {
+    possibleValues: boolean;
+  };
 }
 
 class Configuration {

src/extension.ts

@@ -71,7 +71,7 @@ export async function activate(context: vscode.ExtensionContext) {
         [
           ...new Set([
             configuration.get("completion.autoMode"),
-            configuration.get("completion.manuallyMode"),
+            configuration.get("completion.manualMode"),
             ...(isChatEnabled ? ["chat-medium" as const] : []),
           ]),
         ].map((serverType) => servers[serverType].startServer())