☆ Yσɠƚԋσʂ ☆@lemmy.ml to Technology@lemmy.mlEnglish · 21 days agoQwen3.6-27B that you can run on a laptop outperforms Qwen3.5-397B which was a flagship model requiring a commercial grade server that was released in Februaryqwen.aiexternal-linkmessage-square5linkfedilinkarrow-up144arrow-down11
arrow-up143arrow-down1external-linkQwen3.6-27B that you can run on a laptop outperforms Qwen3.5-397B which was a flagship model requiring a commercial grade server that was released in Februaryqwen.ai☆ Yσɠƚԋσʂ ☆@lemmy.ml to Technology@lemmy.mlEnglish · 21 days agomessage-square5linkfedilink
minus-square☂️-@lemmy.mllinkfedilinkarrow-up2·20 days agodid you guys run it? how much vram do you have?
minus-square☆ Yσɠƚԋσʂ ☆@lemmy.mlOPlinkfedilinkarrow-up2·20 days agoIt depends on what precision you run it at. For 4 bit, you need at least 16 GB VRAM and ideally 24 GB to be comfortable, for the full 16 bit you’d need at least 54gb. A RTX 4090 is minimum. I’ve got 64gb myself, so I can run 8 bit quant comfortably.
did you guys run it? how much vram do you have?
It depends on what precision you run it at. For 4 bit, you need at least 16 GB VRAM and ideally 24 GB to be comfortable, for the full 16 bit you’d need at least 54gb. A RTX 4090 is minimum. I’ve got 64gb myself, so I can run 8 bit quant comfortably.