pytorch
8efaab1b - Add long tensor type to AddFakeFp16 Op (#58504)

Commit
3 years ago
Add long tensor type to AddFakeFp16 Op (#58504) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/58504 .. to support QRT inline_CVR models to avoid failure ``` [DataPreproc] User preprocessing error: c10::Error: [enforce fail at operator.h:1307] . Unsupported type of tensor: long (Error from operator: input: "sparse_nn_2/HistogramBinningCalibrationByFeature_2/cast_22/cast_22_5" input: "sparse_nn_2/HistogramBinningCalibrationByFeature_2/mul_5/Mul" output: "sparse_nn_2/HistogramBinningCalibrationByFeature_2/add_7/Add_2" name: "" type: "AddFakeFp16" arg { name: "broadcast" i: 1 } device_option { extra_info: "inference_split:force_merge" extra_info: "inference_split:force_merge" }) ``` f273407515 Test Plan: f273692411 Reviewed By: hx89 Differential Revision: D28513550 fbshipit-source-id: 86892e1a98b5219cd187731018ce2692b231fb58
Author
Parents
Loading