sh scripts/inference/inference_hunyuan.sh 卡住

[HMZ76]

No description provided.

[HMZ76]

无法进行inference

[rlsu9]

can you show the output of the terminal to give us more clue? Thanks a lot.

[HMZ76]

[HMZ76]

会一直卡在inference

[TZYSJTU]

me too

[TZYSJTU]

me too

H800

[HMZ76]

A800

[lith0613]

A100，same problem

[rlsu9]

Hi, from the screenshot above, I think you are trying to use a single 40G GPU for inference Fast-Hunyuan which is now sufficient. However, I failed to reproduce the bug by myside as it leads to a OOM error. Please have a try with 80G VRAM in total and the bug might be fixed. We are going to release low precision version of FastHunyuan very soon.

[HMZ76]

Hi, from the screenshot above, I think you are trying to use a single 40G GPU for inference Fast-Hunyuan which is now sufficient. However, I failed to reproduce the bug by myside as it leads to a OOM error. Please have a try with 80G VRAM in total and the bug might be fixed. We are going to release low precision version of FastHunyuan very soon.

我这个80G显存呢？

[LiewFeng]

80G A800 also gets stuck.

[foreverpiano]

Can you try to use python fastvideo/utils/env_utils.py and list the env here?

[foreverpiano]

@LiewFeng @HMZ76 Can you try to use python fastvideo/utils/env_utils.py and list the env here?

[HMZ76]

@LiewFeng @HMZ76 Can you try to use python fastvideo/utils/env_utils.py and list the env here?

Thank you! I had solved this problem by changing text_encoder data type fp16 to (bf16 or fp32)