Hello, I have an important question. In general, video cards degrade after long-term work with LLMs, and all because the video cards are not loaded at all at first, and when a request comes from the user, they are instantly loaded to 100%, which is why the video chip degrades. Is it possible to somehow make sure that the video card is constantly at 70-100% load? The simple fact is that the condition of constant 70-100% is more favorable for a video card than from 0% to 100% in an instant. Thanks for your reply (This comment was written with the help of a translator).
This is not true. GPU chips aren't machinery, they don't degrade from load cycling. You are looking for a solution to a problem that doesn't exist. If you're concerned about *thermal cycling* which could in theory wear out your cpu fan and cause a gpu to overheat when the fan fails, then set a power limit on the card to keep temps down.
Brother, can you make a tutorial about LM Studio to do screen sharing with character? With llava or some LLM we can run offline on my rig? I can use llava inside LMStudio, but I can't connect it to sillytavern. I didn't have any problem connect LM studio to Vpet via API.
Can we use LM Studio as backend for vision capable models then connect it to Silly Tavern? I don't understand how to set it in image captioning extension as it's not in the options. Do we need to use an OpenAI-compatible API and set it as Chat Completions in Silly Tavern? Is that possible for LM Studio?
