3 ms·
Waiting for Mixed Quantization with MQQ and MoE Offloading [1]. With that I was able to run Mistral 8x7B on my 10 GB VRAM rtx3080... This should work for DBRX a
by underlines 3y ago
Waiting for Mixed Quantization with MQQ and MoE Offloading [1]. With that I was able to run Mistral 8x7B on my 10 GB VRAM rtx3080... This should work for DBRX and should shave off a ton of VRAM requirement.
1. https://github.com/dvmazur/mixtral-offloading?tab=readme-ov-file https://github.com/dvmazur/mixtral-offloading?tab=readme-ov-...