Autocomplete generate irrelevant content

[homjay]

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:Windows11
- Continue version:v0.8.60
- IDE version:Vscode 1.95.3
- Model:qwen2.5-coder:32b & qwen2.5-coder:1.5b
- config.json:
  
  "tabAutocompleteModel": {
    "title": "qwen2.5-coder:32b",
    // "title": "qwen2.5-coder:1.5b",
    "provider": "ollama",
    "model": "qwen2.5-coder:32b",
    "apiBase": "http://192.168.11.100:11434"
  },
  "tabAutocompleteOptions": {
    "debounceDelay": 500,
    "maxPromptTokens": 4000,
    "disableInFiles": [
      "*.md"
    ],
    "multilineCompletions": "always"
  },

Description

the generated content is text instead of code.

To reproduce

Install continues, run the Qwen model with Ollama, and configure accordingly.

Log output

No response

[arcv]

Noticed similar problem in my local setup with ollama and autocomplete. What is more interesting is that quality of the output degraded with each Continuedev update regardless of model I used. starcoder2:7b used to be fine model for my use in VSCode Unity C# project.

[JadoJodo]

I, too, am seeing this issue (with mistralai/codestral-mamba via OpenRouter).

IDE: JetBrains GoLand w/ Continue v0.0.75

[Hax7]

I have the same issue, Have you found a solution? I use qwen2.5-coder:1.5b-instruct model with ollama and vscode

[Swipe4057]

I have a similar issue with VLLM. #3372