Hunyuan fixed fp8 seems to have some serious problems, Generation on RTX 4070 12 GB, 32 RAM
#15
by
Sikaworld1990
- opened
Input:
Output, gentime 2.267secs
Output gguf Q8 non fixed, gentime 671secs
Tencent fixed model seems to be broken, all the generation with the original, or distilled models fail to get the starting image.
We will have to wait for them to release a corrected model and then a distilled model.