Errors when directly configuring existing Llama setup running outside vscode.

[tempcvrohch]

I'm running devcontainer/monorepo setup so I'm running Llama in a container outside of vscode so it can be shared by multiple editors.
I went in to settings and configured the llama-vscode endpoint but kept getting errors, looking at the developer tools:

Error fetching llama completion: TypeError: Cannot read properties of undefined (reading 'aiModel')
	at LlamaServer.getComplModelProperties (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/llama-server.js:489:32)
	at LlamaServer.getFIMCompletion (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/llama-server.js:28:59)
	at Completion.getCompletionItems (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/completion.js:66:55)
	at Object.provideInlineCompletionItems (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/architect.js:244:54)
	at ia.provideInlineCompletions (file:///vscode/vscode-server/bin/linux-x64/360a4e4fd251....7c7d25d1ad40da/out/vs/workbench/api/node/extensionHostProcess.js:140:145541)
	at file:///vscode/vscode-server/bin/linux-x64/360a4...

TypeError: Cannot read properties of undefined (reading 'aiModel')
	at LlamaServer.getComplModelProperties (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/llama-server.js:489:32)
	at LlamaServer.updateExtraContext (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/llama-server.js:93:59)
	at ExtraContext.periodicRingBufferUpdate [as _onTimeout] (/root/.vscode-server/extensions/ggml-org.llama-vscode-0.0.16/dist/extra-context.js:62:34)
	at listOnTimeout (node:internal/timers:588:17)
	at process.processTimers (node:internal/timers:523:7)

Both of these were fixed by clicking the 'Edit in settings.json' for Complition_models_list and Envs_list, adding an entry for both. Envs_list just needed an empty object.

I added this as a discussion instead of an issue because it looks like the plugin is built around selecting envs, hope it helps someone, great extension btw!

[igardev]

@tempcvrohch Thanks for reporting and great that you found a solution! Under Windows I can't reproduce it, but will try with Linux.

[igardev]

@tempcvrohch I didn't manage to reproduce the problem with Linux Mint as well. However, I have a guess what is the reason for the problem and did a small change for that in version 0.0.18. Hope this will resolve the problem. The goal is llama-vscode to work with the settings if no env is selected.

I guess now you select your newly added env after you start VS Code. If the fix works, selecting env will not be needed.

For completeness some additional information:

• If you always select the same env on starting VS Code, you could automate this with setting env_start_last_used = true.
• You could create an env only with external completion model and use it instead of the settings.