Support for per-model OpenAI Reasoning Effort settings. #698

enricoros · Dec 25, 2024 · 557886f · 557886f
1 parent f52e350
commit 557886f
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 0 deletions.
diff --git a/src/modules/aix/server/dispatch/chatGenerate/adapters/openai.chatCompletions.ts b/src/modules/aix/server/dispatch/chatGenerate/adapters/openai.chatCompletions.ts
@@ -85,6 +85,11 @@ export function aixToOpenAIChatCompletions(openAIDialect: OpenAIDialects, model:
     payload.top_p = model.topP;
   }
 
+  // [OpenAI] Vendor-specific reasoning effort, for o1 models only as of 2024-12-24
+  if (model.vndOaiReasoningEffort) {
+    payload.reasoning_effort = model.vndOaiReasoningEffort;
+  }
+
   if (hotFixOpenAIo1Family)
     payload = _fixRequestForOpenAIO1_maxCompletionTokens(payload);
 

diff --git a/src/modules/aix/server/dispatch/wiretypes/openai.wiretypes.ts b/src/modules/aix/server/dispatch/wiretypes/openai.wiretypes.ts
@@ -233,6 +233,7 @@ export namespace OpenAIWire_API_Chat_Completions {
     stream_options: z.object({
       include_usage: z.boolean().optional(), // If set, an additional chunk will be streamed with a 'usage' field on the entire request.
     }).optional(),
+    reasoning_effort: z.enum(['low', 'medium', 'high']).optional(), // [OpenAI, 2024-12-17] reasoning effort, o1 models only for now
     response_format: z.discriminatedUnion('type', [
       z.object({
         type: z.literal('text'), // Default