/vt/ - Virtual Youtubers

>>57643320
It's a fine-tune of Llama2, meaning that the system requirements for these models would be the same as base Llama2. 13B at half-precision would likely not even fit on a 3090, but 7B at half precision takes up approximately 14 GB of VRAM: https://discuss.huggingface.co/t/llama-7b-gpu-memory-requirement/34323/8
Of course, quantization can help out with this significantly.