Skip to content

LFM2 behaves differently than GPT-OSS:20b or GLM-4.7-FLASH #1010

Description

@robogeek

Describe the bug

With LFM2, Perplexica immediately starts answering, and writes its answer quickly, but it does not include several attributes that are produced when using other models like GPT-OSS:20b or GLM-4.7-FLASH.

Namely:

  • The Brainstorming section never resolved
  • Sources are not listed at the top
  • The Research Progress section does not appear
  • The answer text does not include any cited reference numbers
  • No sidebars for related videos or pictures
  • No Related section at the bottom for possible other questions

To Reproduce

I am successfully using Perplexica - generally with GPT-OSS:20b. I am using a local Ollama instance along with a separate SearXNG instance.

Switch over to the LFM2 model -- it was automatically shown in the list in the Settings.

Enter the exact same query.

Expected behavior

The answer should include all the attributes that I described are missing above.

Screenshots

Image Image

Additional context

I notice that Ollama describes LFM2 as tool using, but not thinking, whereas the other two I named are both tool using and thinking.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions