continue responses are all in JSON format

[TobiasA1]

Before submitting your bug report

• [x] I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• [x] I believe this is a bug. I'll try to join the Continue Discord for questions
• [x] 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:
- Continue version:1.0
- IDE version:vs code latest as of now
- Model:ollama
- config:
  

  
  OR link to assistant in Continue hub:

Description

for quite a long time now any time i try to use the continue extension for anything it just responds with a json code and that's the end of, no description on a nswers just a json file and it ends the chat even when i type something as simple as hi it still responds with a json file.

And also the extension has become very slow, when i use ollama elsewhere it's always blazing fast but in the continue extention it takes ages before responding and even for a very little 1.5b model

To reproduce

No response

Log output

[omsuneri]

@sestinj @Patrick-Erichsen can you please assign this issue to me !! /assign

[44cort44]

this happens to me with the qwen2.5-coder:7b and :14b models I specifically tested them both with q8_0 in agent mode - despite Continue saying they might not work well in agent mode

[RomneyDa]

@44cort44 @TobiasA1 is it specifically tool calls that show as JSON?

[Andrew1431]

This is similar to the issue I'm having and reported just now https://github.com/continuedev/continue/issues/7363 here.

I feel like at one point I had this working and now I'm just receiving these tool prompts as JSON.

[uinstinct]

hey @Andrew1431 can you please try it with system message tools as on?

let me know if you are still getting json responses back with system message tools turned on.

[zhanghm1995]

hey @Andrew1431 can you please try it with system message tools as on?

let me know if you are still getting json responses back with system message tools turned on.

@uinstinct I also encountered this JSON format response problem, and solved it by opening this setting. Does this setting affect the quality of the response?

[uinstinct]

Not necessarily as far as I have used. This setting is useful when models do not support native tool calling. Hence, parsing them using system message tools adds agentic ability to that model.

[splab]

I was seeing the exact same thing with json.

Switching the "Only use system message tools" to on and restarting all the things seems to have solved it.

[sav]

can you please try it with system message tools as on?

works for me. I was having the same issue here, using qwen2.5-coder:7b in the pre-release version (1.3.24). thank you!