ref: enable deepseek

package.json

@@ -2,7 +2,7 @@
   "name": "llama-coder",
   "displayName": "Llama Coder",
   "description": "Better and self-hosted Github Copilot replacement",
-  "version": "0.0.6",
+  "version": "0.0.7",
   "icon": "icon.png",
   "publisher": "ex3ndr",
   "repository": {
@@ -70,14 +70,15 @@
               "codellama:34b-code-q4_K_S",
               "codellama:34b-code-q4_K_M",
               "codellama:34b-code-q6_K",
-              "deepseek-coder:6.7b-instruct-q4_K_S",
-              "deepseek-coder:6.7b-instruct-q4_K_M",
-              "deepseek-coder:6.7b-instruct-q8_0",
-              "deepseek-coder:6.7b-instruct-fp16",
-              "deepseek-coder:33b-instruct-q4_K_S",
-              "deepseek-coder:33b-instruct-q4_K_M",
-              "deepseek-coder:33b-instruct-q8_0",
-              "deepseek-coder:33b-instruct-fp16"
+              "deepseek-coder:1.3b-base-q4_0",
+              "deepseek-coder:1.3b-base-q4_1",
+              "deepseek-coder:1.3b-base-q8_0",
+              "deepseek-coder:6.7b-base-q4_K_S",
+              "deepseek-coder:6.7b-base-q8_0",
+              "deepseek-coder:6.7b-base-fp16",
+              "deepseek-coder:33b-base-q4_K_S",
+              "deepseek-coder:33b-base-q4_K_M",
+              "deepseek-coder:33b-base-fp16"
             ],
             "default": "codellama:7b-code-q4_K_M",
             "description": "Inference model to use"

src/prompts/adaptors/adaptPrompt.ts

@@ -1,10 +1,24 @@
-export function adaptPrompt(args: { model: string, prefix: string, suffix: string }): string {
+export function adaptPrompt(args: { model: string, prefix: string, suffix: string }): { prompt: string, stop: string[] } {
 
     // Starcoder format
     if (args.model.startsWith('deepseek-coder')) {
-        return `<｜fim▁begin｜>${args.prefix}<｜fim▁hole｜>${args.suffix}<｜fim▁end｜>`;
+
+        if (args.suffix.length < 1000) {
+            return {
+                prompt: args.prefix,
+                stop: [`<END>`]
+            };
+        }
+
+        return {
+            prompt: `<｜fim▁begin｜>${args.prefix}<｜fim▁hole｜>${args.suffix}<｜fim▁end｜>`,
+            stop: [`<｜fim▁begin｜>`, `<｜fim▁hole｜>`, `<｜fim▁end｜>`, `<END>`]
+        };
     }
 
     // Codellama format
-    return `<PRE> ${args.prefix} <SUF>${args.suffix} <MID>`;
+    return {
+        prompt: `<PRE> ${args.prefix} <SUF>${args.suffix} <MID>`,
+        stop: [`<PRE>`, `<SUF>`, `<MID>`, `<END>`]
+    };
 }

src/prompts/autocomplete.ts

@@ -14,12 +14,15 @@ export async function autocomplete(args: {
     canceled?: () => boolean,
 }): Promise<string> {
 
+    let prompt = adaptPrompt({ prefix: args.prefix, suffix: args.suffix, model: args.model });
+
     // Calculate arguments
     let data = {
         model: args.model,
-        prompt: adaptPrompt({ prefix: args.prefix, suffix: args.suffix, model: args.model }),
+        prompt: prompt.prompt,
         raw: true,
         options: {
+            stop: prompt.stop,
             num_predict: args.maxTokens,
             temperature: args.temperature
         }

src/prompts/filter.ts

@@ -13,9 +13,9 @@ export function isNotNeeded(doc: vscode.TextDocument, position: vscode.Position,
     // }
 
     // Avoid autocomplete when system menu is shown (ghost text is hidden anyway)
-    if (context.selectedCompletionInfo) {
-        return true;
-    }
+    // if (context.selectedCompletionInfo) {
+    //     return true;
+    // }
 
     return false;
 }

src/prompts/processors/detectLanguage.ts

@@ -0,0 +1,10 @@
+// import path from 'path';
+
+// let languages: { [key: string]: {} } = {
+
+// };
+
+// export function fileHeaderProcessor(uri: string, languageId: string): string | null {
+//     let basename = path.basename(uri);
+//     let extname = 
+// }