feat. add max_input_tokens config

src/lib/constants.ts

@@ -3,6 +3,7 @@ export const REDIRECT_URI =
 
 export const MODEL_CONFIG_PROMPT_IMPROVEMENT = {
   id: "deepseek-ai/DeepSeek-V3-0324",
+  max_input_tokens: 48_000,
   max_tokens: 16_000,
 }
 
@@ -11,24 +12,28 @@ const default_system_prompt = `ONLY USE HTML, CSS AND JAVASCRIPT. If you want to
 export const MODEL_CONFIG_CODE_GENERATION = [
   {
     id: "deepseek-ai/DeepSeek-V3-0324",
+    max_input_tokens: 48_000,
     max_tokens: 16_000,
     default_enable_thinking: false,
     system_prompt: default_system_prompt,
   },
   {
     id: "Qwen/Qwen3-235B-A22B",
+    max_input_tokens: 24_000,
     max_tokens: 16_000,
     default_enable_thinking: true,
     system_prompt: default_system_prompt,
   },
   {
     id: "Qwen/Qwen3-30B-A3B",
+    max_input_tokens: 24_000,
     max_tokens: 16_000,
     default_enable_thinking: true,
     system_prompt: default_system_prompt,
   },
   {
     id: "Qwen/Qwen3-32B",
+    max_input_tokens: 24_000,
     max_tokens: 16_000,
     default_enable_thinking: true,
     system_prompt: default_system_prompt,

src/lib/inference-utils.ts

@@ -7,6 +7,7 @@ const sharedEncoder = new TextEncoder();
 export interface ModelConfig {
   id: string;
   max_tokens: number;
+  max_input_tokens: number;
   system_prompt?: string;
   default_enable_thinking?: boolean;
 }
@@ -31,11 +32,11 @@ export async function getInferenceToken(request: NextRequest): Promise<{
 }
 
 export function checkTokenLimit(tokensUsed: number, modelConfig: ModelConfig) {
-  if (tokensUsed >= modelConfig.max_tokens) {
+  if (tokensUsed >= modelConfig.max_input_tokens) {
     return {
       ok: false,
       openSelectProvider: true,
-      message: `Context is too long. ${modelConfig.id} allow ${modelConfig.max_tokens} max tokens.`,
+      message: `Context is too long. ${modelConfig.id} allow ${modelConfig.max_input_tokens} max input tokens.`,
     };
   }
   return null;