SJTU-IPADS / PowerInfer

High-speed Large Language Model Serving on PCs with Consumer-grade GPUs
MIT License
7.98k stars 415 forks source link

the output for Q4_gguf is strange again!! #208

Open milktea888 opened 4 months ago

milktea888 commented 4 months ago

Current Behavior

the output of Q4 model is very strange.

1720086605607

And I see there was ever such a bug, in https://github.com/SJTU-IPADS/PowerInfer/issues/77 and also I see you have given a solution in commit id 79986ec58853c23e6df3d277b6724b83e996b3e2 which I switched to run is good, but it is strange again in the newest version, may I know what have you changed? And can you give me some advice on this problem? And the commit id I use is 61cac9bf25e60336bbad27ada9dbb809204473ac

And now I found that if I build the program using only cpu, the output will be ok. That is, if I build it using GPU, the output will be strange.

bingo787 commented 4 months ago

I meet this issue too!