2 Bit Embedding Conversion Operator support. (#43077)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/43077
2 Bit Embedding weight conversion operation is quite similar to
4 bit embedding weight conversion.
The diff contains both the
1. 2bit packing op `embedding_bag_2bit_prepack`.
2. 2bit unpacking op `embedding_bag_2bit_unpack`.
Comments about the op are inline with the op definition.
Test Plan: buck test caffe2/test:quantization -- test_embedding_bag_2bit_unpack
Reviewed By: supriyar
Differential Revision: D23143262
fbshipit-source-id: fd8877f049ac1f7eb4bc580e588dc95f8b1edef0
Author
Radhakrishnan Venkataramani