From 2976782e0b9958502c7b5e640dd7459e8b4d5522 Mon Sep 17 00:00:00 2001
From: igorMSoares <igorms.dev@gmail.com>
Date: Sat, 25 Jan 2025 15:25:55 -0300
Subject: [PATCH] feat: add system prompt support to ReplicateLLM and OpenAI

---
 packages/providers/openai/src/llm.ts    | 10 +++++++++-
 packages/providers/replicate/src/llm.ts | 10 +++++++++-
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/packages/providers/openai/src/llm.ts b/packages/providers/openai/src/llm.ts
index 492db09905..09c3c6a6c0 100644
--- a/packages/providers/openai/src/llm.ts
+++ b/packages/providers/openai/src/llm.ts
@@ -155,6 +155,7 @@ export class OpenAI extends ToolCallLLM<OpenAIAdditionalChatOptions> {
   topP: number;
   maxTokens?: number | undefined;
   additionalChatOptions?: OpenAIAdditionalChatOptions | undefined;
+  systemPrompt?: string | undefined;
 
   // OpenAI session params
   apiKey?: string | undefined = undefined;
@@ -185,6 +186,7 @@ export class OpenAI extends ToolCallLLM<OpenAIAdditionalChatOptions> {
     this.temperature = init?.temperature ?? 0.1;
     this.topP = init?.topP ?? 1;
     this.maxTokens = init?.maxTokens ?? undefined;
+    this.systemPrompt = init?.systemPrompt ?? undefined;
 
     this.maxRetries = init?.maxRetries ?? 10;
     this.timeout = init?.timeout ?? 60 * 1000; // Default is 60 seconds
@@ -337,12 +339,18 @@ export class OpenAI extends ToolCallLLM<OpenAIAdditionalChatOptions> {
     | AsyncIterable<ChatResponseChunk<ToolCallLLMMessageOptions>>
   > {
     const { messages, stream, tools, additionalChatOptions } = params;
+    const systemMessage: ChatMessage<ToolCallLLMMessageOptions> | undefined =
+      this.systemPrompt
+        ? { role: "system", content: this.systemPrompt }
+        : undefined;
     const baseRequestParams = <OpenAILLM.Chat.ChatCompletionCreateParams>{
       model: this.model,
       temperature: this.temperature,
       max_tokens: this.maxTokens,
       tools: tools?.map(OpenAI.toTool),
-      messages: OpenAI.toOpenAIMessage(messages),
+      messages: OpenAI.toOpenAIMessage(
+        systemMessage ? [systemMessage].concat(messages) : messages,
+      ),
       top_p: this.topP,
       ...Object.assign({}, this.additionalChatOptions, additionalChatOptions),
     };
diff --git a/packages/providers/replicate/src/llm.ts b/packages/providers/replicate/src/llm.ts
index 7012af8c02..efc2966480 100644
--- a/packages/providers/replicate/src/llm.ts
+++ b/packages/providers/replicate/src/llm.ts
@@ -111,6 +111,7 @@ export class ReplicateLLM extends BaseLLM {
   topP: number;
   maxTokens?: number;
   replicateSession: ReplicateSession;
+  systemPrompt?: string | undefined;
 
   constructor(init?: Partial<ReplicateLLM> & { noWarn?: boolean }) {
     super();
@@ -135,6 +136,7 @@ export class ReplicateLLM extends BaseLLM {
       init?.maxTokens ??
       ALL_AVAILABLE_REPLICATE_MODELS[this.model].contextWindow; // For Replicate, the default is 500 tokens which is too low.
     this.replicateSession = init?.replicateSession ?? new ReplicateSession();
+    this.systemPrompt = init?.systemPrompt ?? undefined;
   }
 
   get metadata() {
@@ -175,6 +177,12 @@ export class ReplicateLLM extends BaseLLM {
   }
 
   mapMessagesToPromptLlama3(messages: ChatMessage[]) {
+    const systemPrompt = this.systemPrompt
+      ? "<|begin_of_text|><|start_header_id|>system<|end_header_id|>" +
+        this.systemPrompt +
+        "<|eot_id|>"
+      : undefined;
+
     return {
       prompt:
         "<|begin_of_text|>" +
@@ -196,7 +204,7 @@ export class ReplicateLLM extends BaseLLM {
           );
         }, "") +
         "<|start_header_id|>assistant<|end_header_id|>\n\n",
-      systemPrompt: undefined,
+      systemPrompt,
     };
   }