Refactor (9)

lib/llm.ts

@@ -7,6 +7,7 @@ import {
   type LLamaChatPromptOptions,
   LlamaChatSession,
   type LlamaModel,
+  TokenBias,
   createModelDownloader,
   getLlama,
   resolveChatWrapper,
@@ -27,6 +28,15 @@ export async function getModel(model: string) {
   return await llama.loadModel({ modelPath });
 }
 
+export function createBias(model: LlamaModel) {
+  const customBias = new TokenBias(model.tokenizer);
+  for (const token of model.iterateAllTokens()) {
+    const text = model.detokenize([token]);
+    if (text === "{") customBias.set(token, -0.9); // suppress JSON string
+  }
+  return customBias;
+}
+
 export const createGrammar = (assistantName: string) =>
   llama.createGrammarForJsonSchema({
     type: "object",