correct type errors

Author: ferenci84

core/index.d.ts

@@ -308,7 +308,7 @@ export interface CompletionOptions extends BaseCompletionOptions {
   model: string;
 }
 
-export type ChatMessageRole = "user" | "assistant" | "system" | "tool";
+export type ChatMessageRole = "user" | "assistant" | "thinking" | "system" | "tool";
 
 export type TextMessagePart = {
   type: "text";

core/llm/countTokens.ts

@@ -27,9 +27,9 @@ class LlamaEncoding implements Encoding {
 }
 
 class NonWorkerAsyncEncoder implements AsyncEncoder {
-  constructor(private readonly encoding: Encoding) {}
+  constructor(private readonly encoding: Encoding) { }
 
-  async close(): Promise<void> {}
+  async close(): Promise<void> { }
 
   async encode(text: string): Promise<number[]> {
     return this.encoding.encode(text);
@@ -366,6 +366,7 @@ function chatMessageIsEmpty(message: ChatMessage): boolean {
         message.content.trim() === "" &&
         !message.toolCalls
       );
+    case "thinking":
     case "tool":
       return false;
   }
@@ -383,8 +384,8 @@ function compileChatMessages(
 ): ChatMessage[] {
   let msgsCopy = msgs
     ? msgs
-        .map((msg) => ({ ...msg }))
-        .filter((msg) => !chatMessageIsEmpty(msg) && msg.role !== "system")
+      .map((msg) => ({ ...msg }))
+      .filter((msg) => !chatMessageIsEmpty(msg) && msg.role !== "system")
     : [];
 
   msgsCopy = addSpaceToAnyEmptyMessages(msgsCopy);
@@ -469,5 +470,6 @@ export {
   pruneLinesFromTop,
   pruneRawPromptFromTop,
   pruneStringFromBottom,
-  pruneStringFromTop,
+  pruneStringFromTop
 };
+

core/llm/index.ts

@@ -181,11 +181,11 @@ export abstract class BaseLLM implements ILLM {
         options.completionOptions?.maxTokens ??
         (llmInfo?.maxCompletionTokens
           ? Math.min(
-              llmInfo.maxCompletionTokens,
-              // Even if the model has a large maxTokens, we don't want to use that every time,
-              // because it takes away from the context length
-              this.contextLength / 4,
-            )
+            llmInfo.maxCompletionTokens,
+            // Even if the model has a large maxTokens, we don't want to use that every time,
+            // because it takes away from the context length
+            this.contextLength / 4,
+          )
           : DEFAULT_MAX_TOKENS),
     };
     this.requestOptions = options.requestOptions;
@@ -811,8 +811,8 @@ export abstract class BaseLLM implements ILLM {
             signal,
             completionOptions,
           )) {
-            completion += chunk.content;
-            yield chunk;
+
+            if (chunk.role === "assistant") {
               completion += chunk.content;
               yield chunk;
             }
@@ -912,15 +912,15 @@ export abstract class BaseLLM implements ILLM {
     );
   }
 
-  protected async *_streamComplete(
+  protected async * _streamComplete(
     prompt: string,
     signal: AbortSignal,
     options: CompletionOptions,
   ): AsyncGenerator<string> {
     throw new Error("Not implemented");
   }
 
-  protected async *_streamChat(
+  protected async * _streamChat(
     messages: ChatMessage[],
     signal: AbortSignal,
     options: CompletionOptions,

core/llm/llms/Ollama.ts

@@ -1,12 +1,12 @@
 import { JSONSchema7, JSONSchema7Object } from "json-schema";
 
-import { ChatMessage, CompletionOptions, LLMOptions } from "../../index.js";
+import { ChatMessage, ChatMessageRole, CompletionOptions, LLMOptions } from "../../index.js";
 import { renderChatMessage } from "../../util/messageContent.js";
 import { BaseLLM } from "../index.js";
 import { streamResponse } from "../stream.js";
 
 type OllamaChatMessage = {
-  role: "tool" | "user" | "assistant" | "system";
+  role: ChatMessageRole;
   content: string;
   images?: string[] | null;
   tool_calls?: {
@@ -82,10 +82,10 @@ type OllamaBaseResponse = {
   model: string;
   created_at: string;
 } & (
-  | {
+    | {
       done: false;
     }
-  | {
+    | {
       done: true;
       done_reason: string;
       total_duration: number; // Time spent generating the response in nanoseconds
@@ -96,7 +96,7 @@ type OllamaBaseResponse = {
       eval_duration: number; // Time spent generating the response in nanoseconds
       context: number[]; // An encoding of the conversation used in this response; can be sent in the next request to keep conversational memory
     }
-);
+  );
 
 type OllamaErrorResponse = {
   error: string;
@@ -105,14 +105,14 @@ type OllamaErrorResponse = {
 type OllamaRawResponse =
   | OllamaErrorResponse
   | (OllamaBaseResponse & {
-      response: string; // the generated response
-    });
+    response: string; // the generated response
+  });
 
 type OllamaChatResponse =
   | OllamaErrorResponse
   | (OllamaBaseResponse & {
-      message: OllamaChatMessage;
-    });
+    message: OllamaChatMessage;
+  });
 
 interface OllamaTool {
   type: "function";
@@ -142,7 +142,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }
@@ -331,7 +331,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }
@@ -396,7 +396,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }
@@ -482,7 +482,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }
@@ -523,7 +523,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }
@@ -549,7 +549,7 @@ class Ollama extends BaseLLM {
     const headers: Record<string, string> = {
       "Content-Type": "application/json",
     };
-  
+
     if (this.apiKey) {
       headers.Authorization = `Bearer ${this.apiKey}`;
     }

core/util/messageContent.ts

@@ -21,6 +21,7 @@ export function renderChatMessage(message: ChatMessage): string {
   switch (message.role) {
     case "user":
     case "assistant":
+    case "thinking":
     case "system":
       return stripImages(message.content);
     case "tool":
@@ -36,6 +37,7 @@ export function normalizeToMessageParts(message: ChatMessage): MessagePart[] {
   switch (message.role) {
     case "user":
     case "assistant":
+    case "thinking":
     case "system":
       return Array.isArray(message.content)
         ? message.content