How to change the language model (LLM) Swark is using?

[ozanani]

How to change the language model (LLM) Swark is using?

[ozanani]

You can configure the language model that Swark is using.
The current default is GPT 4o.

If Claude 3.5 Sonnet is available for your, it's recommended to use it:

• It allows 3x bigger input, therefore allows to include more code files in your diagram.
• It usually creates better diagrams.

To change the language model:

1. Open the Command Palette and choose Preferences: Open User Settings
2. Search for swark.languageModel
3. Choose claude-3.5-sonnet

If you encounter issues after changing the model, you can try to interact with it directly in Copilot to see if it works. You can change Copilot's model using the model picker:

[selvakanda]

I am getting the token limit warning. "Processing 5/321 files due to LLM token limit." How to configure and increase the token limit?

[ozanani]

@selvakanda the token limit is determined by the language model you use. It's a limitation of the model that cannot be changed.
Tips:

1. Try to use claude-3.5-sonnet if it's available for you. It has the largest token limit
2. You might have some big files that are included in the file search. Consider running Swark on sub-folders, or exclude specific files with the swark.excludePatterns setting:

Let me know if you need help with that!

[evlachosIQM]

I have access to Gemini2.5Pro via Github Copilot Pro license. Is there a way to add that model to the list of available models when using Swark?

[ozanani]

Yes, we can add it to the options in 'package.json'. Feel free to publish a PR on that