Could AITemplate support mix precision inference?

[sanbuphy]

like fp32/fp16 , fp32/bf16.

Thank you very much .

[sgrigory]

@sanbuphy Unfortunately AITemplate doesn't support mixed precision/auto-cast functionality.

@chenyang78 CC'ing you for visibility, given the recent discussion on the topic

[sanbuphy]

@sanbuphy Unfortunately AITemplate doesn't support mixed precision/auto-cast functionality.

@chenyang78 CC'ing you for visibility, given the recent discussion on the topic

I see, so I cannot use AIT to compile any mixed precision models with amp.autocast in PyTorch. Thank you for your response.