Use instruct mode, with the instruction in the memory window, with the formatting discussed here: https://github.com/ggerganov/llama.cpp/issues/2262#issuecomme

Use instruct mode, with the instruction in the memory window, with the formatting discussed here: https://github.com/ggerganov/llama.cpp/issues/2262#issuecomment-1640795572 https://github.com/ggerganov/llama.cpp/issues/2262#issuecomm... If you are feeling generous, host a V2 worker using KoboldCPP. Its pretty easy to set up, and 13B/70B will run on PCs with consumer GPUs: https://github.com/db0/KoboldAI-Horde-Bridge https://github.com/db0/KoboldAI-Horde-Bridge