torch.ao.quantization deprecation tracker (original) (raw)

We are deprecating torch.ao.quantization and will migrate pt2e quantization to torchao.

API changes:

# before
from torch.ao.quantization.quantize_pt2e import prepare_pt2e, prepare_qat_pt2e, convert_pt2e
# after
from torchao.quantization.pt2e.quantize_pt2e import prepare_pt2e, prepare_qat_pt2e, convert_pt2e

# TODO: more

• pt2e quantization code duplicated in torchao, new changes to pt2e flow should now happen in torchao
• Migrate existing use cases of pt2e quantization to use torchao
• Added deprecation warning in pytorch
• update pytorch/pytorch docs and tutorials, deprecate old eager/fx tutorials and point the pytorch pt2e docs to torchao
• Delete torch.ao.quantization code in pytorch in 2.10 (if there are no blockers)