v.0.2.0 Development Notes

Hello all,

I’ve been at work fixing some issues with Bunker Love and making some new discoveries.

The first discovery, is that you don’t need to stick to a specific local LLM if you’re using LM Studio. It appears loading ANY LLM will work, so pick your poison and get to chatting with whatever model your heart desires (that’s available through LMs Studio).

I’ve also fixed some issues with Ui items, like the main menu vanishing and Claude 3’s delayed response causing the map to open before the companion was ready and the data could be saved. This would result in the game defaulting to calling out to GPT.

MAJOR FEATURE UPDATE

I’ve added Text-To-Speech for all responses from all models. I opted to buy a plugin to handle the TTS locally, offline, so that the latency was as reduced as possible and to preserve the privacy of uncensored chats.

With offline TTS, you truly get private conversations between you and you LM Studio chosen LLM. None of the conversations leave your local machine.

Additionally, if you’re using GPT or Claude 3, your TTS responses are much faster because they’re generated locally.

Response time is about 1-2 seconds.

When running an uncensored model locally, using LM Studio, I also discovered that fully loading the model onto my GPU resulted in chat responses being produced at lighting fast speeds with audio responses 1-2 seconds behind.

Pretty satisfactory, if I do say so myself.

I’m looking into making the build for v0.2.0 now, but I suspect it might be too large for Itch.io, so I’ll look into how I want to proceed with deploying the latest version.

If you’ve already purchased an earlier copy of Bunker Love, the newest v0.2.0 will be available for no additional cost- so don’t hold back if you’re thinking about checking it out now.