Running On <4gb Vram
#58
by Jit2024 - opened
Hi, I found out a way of running this model on <4gb Vram. And I have not used any quantization.
Repo: https://github.com/Jit-Roy/WeeLLM
WeeLLM's README reports MiniMax H3 a 3.14GB peak-VRAM figure, but leaves peak RAM and generation time blank. Could you add a complete FL2VA example with the checkpoint, frame count, resolution, generation time, and peak system RAM—including VAE decoding? That would help readers see whether the reported GPU footprint still fits their whole machine. I haven't reproduced the run.