Constantly loses connection to LLM (both local and cloud) on long responses

[d00mus]

Before submitting your bug report

• I believe this is a bug. I'll try to join the Continue Discord for questions
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows 10
- Continue version: 0.8.68 - for VS Code and 0.0.92 for IDEA
- IDE version: VS Code 1.96.4 and JetBrains IDEA 2023.1.7
- Model: Qwen 2.5 coder 32b on local lmstudio but the same issue with cloud DeepSeek v3
- config:
  
{
  "models": [
    {
      "apiBase": "http://127.0.0.1:1234/v1/",
      "title": "Lmstudio",
      "model": "qwen2.5-coder-32b",
      "contextLength": 65536,
      "provider": "lmstudio",
      "timeout": 6000000
    },
    {
      "title": "DeepSeek Coder",
      "model": "deepseek-coder",
      "contextLength": 128000,
      "apiKey": "ххх",
      "provider": "deepseek",
      "timeout": 6000000
    },
    {
      "title": "DeepSeek",
      "model": "deepseek-chat",
      "contextLength": 128000,
      "apiKey": "ххх",
      "provider": "deepseek",
      "timeout": 6000000
    },
    {
      "title": "DeepSeek R1",
      "model": "deepseek-reasoner",
      "contextLength": 128000,
      "apiKey": "ххх",
      "provider": "deepseek",
      "timeout": 60000
    }
  ],

Description

It just stopping receiving streamed response from LLM after about 400 lines of generated code and no any related errors in the plugin logs.

But In the logs of lmstudio server i'm seening this:
025-02-26 13:48:11 [INFO]
Finished streaming response
2025-02-26 13:48:11 [INFO]
[LM STUDIO SERVER] Client disconnected. Stopping generation... (If the model is busy processing the prompt, it will finish first.)

It looks like the plugin just closes connection for some reason.

Tried to increase timeout - it didn't help.

As the result it just returns a half of the file that being generated in that moment

To reproduce

1. Configure it to use cloud DeepSeek V3 or local LLM via lmstudio with Qwen 2.5 coder 32b (didn't try others),
2. Ask it to fix some problem in the relatively big file. like 800 lines of the java code and provide all the context, about 5 related files.
3. Wait it to start getting streamed response from the LLM
4. It will stop getting generated code somewhere in the middle leaving you with the half of the fixed file.

The problem is reproduceable with both, with the local LLM and with the cloud DeepSeek so I don't think it related to my local LLM or to my internet connection. And I don't have any similar issues with any other tool I use.

Log output

This is the tail of the log but I don't think these lines relate to this particular problem(different timestamp) but it does not contains any others. 
[2025-02-26T10:02:37] Error: Connection error. 
[2025-02-26T10:02:37] Error running handler for "llm/streamChat":  Error: Connection error.
[2025-02-26T10:05:10] Error: Connection error. 
[2025-02-26T10:05:10] Error running handler for "llm/streamChat":  Error: Connection error.
[2025-02-26T10:06:18] Error: Connection error. 
[2025-02-26T10:06:18] Error running handler for "llm/streamChat":  Error: Connection error.
[2025-02-26T10:06:42] Error: Connection error. 
[2025-02-26T10:06:42] Error running handler for "llm/streamChat":  Error: Connection error.