Breaking Change: Remove the deprecated int8 calibrator related#3759
Breaking Change: Remove the deprecated int8 calibrator related#3759lanluo-nvidia merged 10 commits intomainfrom
Conversation
|
No ciflow labels are configured for this repo. |
narendasan
left a comment
There was a problem hiding this comment.
Mostly looks good, if the fx changes still work I would leave those since Meta might still be using them.
zewenli98
left a comment
There was a problem hiding this comment.
LGTM. just a minor comment
| from torch_tensorrt.fx.converters.converter_utils import ( | ||
| mark_as_int8_layer, | ||
| set_layer_name, | ||
| ) |
There was a problem hiding this comment.
Any reasons for partially replacing import lib?
narendasan
left a comment
There was a problem hiding this comment.
Its looking good, only thing is if we are switching to dynamo.types.TRTTensor we should just go to trt.ITensor and be closer to removing that types file. Dont think I need to review again after those changes
zewenli98
left a comment
There was a problem hiding this comment.
LGTM. I think we can directly use trt.ITensor instead of alias, but it doesn't matter. that's fine for now
|
An error pops up. you can import from torch_tensorrt.dynamo.utils |
Description
Remove all the int8 calibrator, int8 implicit quantization and dynamic range related.
All the quantization should be explicit.
Fixes # (issue)
Type of change
Please delete options that are not relevant and/or add your own.
Checklist: