Skip to content

[Bug] output_text used in place of input_text #154

@FyrbyAdditive

Description

@FyrbyAdditive

Describe the bug
I was unable to connect this client to my local Spark cluster running VLLM and MiniMax M2.7 without getting error messages.

To Reproduce
Steps to reproduce the behavior:

  1. Add my local server to macai as OpenAI compatible
  2. Try to send message in client
  3. See error
>> Creating new MessageManager with URL: https://magi.fyrby.internal:8000/v1 and model: cyankiwi/MiniMax-M2.7-AWQ-4bit
Streaming error: serverError("{\"detail\":\"Not Found\"}")
Error sending message: serverError("{\"detail\":\"Not Found\"}")

Expected behavior
Valid response.

Environment:

  • 2x DGX Spark running eugr VLLM docker image
  • macOS 26.5 on 48GB Macbook Pro M4

Additional context

This seems to fix it. I am not sure output_text is valid.

index 6c2805e..e1693a3 100644
--- a/macai/Utilities/APIHandlers/OpenAIResponsesHandler.swift
+++ b/macai/Utilities/APIHandlers/OpenAIResponsesHandler.swift
@@ -238,7 +238,7 @@ class OpenAIResponsesHandler: OpenAIHandlerBase, APIService {
                 let textContent = AttachmentParser.stripAttachments(from: content)
 
                 var contentArray: [[String: Any]] = []
-                let textType = role == "assistant" ? "output_text" : "input_text"
+                let textType = "input_text"
 
                 if !textContent.isEmpty {
                     contentArray.append(["type": textType, "text": textContent])```

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions