Latest v1.1.48 Version Analysis: Think Tag Issues with Local Model MiniMax2.1 #11709

Environment:
Running MiniMax2.1 with locally deployed model vllm

The think tag parsing differs from previous versions, causing the program to halt and wait for user input instead of running smoothly. This renders it largely unusable.
The issue occurs in the new version but not in older ones.

Additionally, Ubuntu automatically upgrades after reverting installed versions. How can I disable automatic upgrades?

Translated with DeepL.com (free version)

bad:
用户只是说"hi"，这是一个简单的打招呼。根据行为指导，我应该：

1. 响应，不需要状态更新或开场白
2. 简洁，不需要赞美或恭维
3. 直接询问需要什么帮助
   这是一个非常简单的交互，不需要使用任何工具。

Good:
Thinking: The user just said "hi" - this is a simple greeting. According to the instructions, I should be concise and direct. I should not use emojis unless explicitly requested, and I should keep my response short.
I'll just greet them back briefly.

{
"$schema": "https://opencode.ai/config.json",
"provider": {
"my-provider": {
"npm": "@ai-sdk/openai-compatible",
"name": "lvllm",
"options": {
"baseURL": "http://192.168.0.32:9070/v1"
},
"models": {
"MiniMax-M2.1-AWQ-4bit": {
"name": "MiniMax-M2.1",
"cost": {
"input": 0.10,
"output": 1.20
},
"options": {
"num_ctx": 196000,
"temperature": 0.9,
"top_p": 0.95,
"top_k": 40
}

    }
  }
}

},
"plugin": [
"oh-my-opencode"
],
"autoupdate":false
}

autoupdate false not worker?