Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

FP16 support for topK #14125

Closed
eric-haibin-lin opened this issue Feb 12, 2019 · 3 comments · Fixed by #15560
Closed

FP16 support for topK #14125

eric-haibin-lin opened this issue Feb 12, 2019 · 3 comments · Fixed by #15560
Labels
CUDA Feature request FP16 Operator

Comments

@eric-haibin-lin
Copy link
Member

TopK doesn't support fp16 inputs.
@mxnet-label-bot
Copy link
Contributor

Hey, this is the MXNet Label Bot.
Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it.
Here are my recommended labels: Feature

@eric-haibin-lin eric-haibin-lin mentioned this issue Feb 12, 2019
Closed
@stephenrawls
Copy link
Contributor

+1

This is annoying for us because we need to use topk in our CRF Viterbi decoding during inference (for k-best Viterbi).

Because topk doesn't support float16 this means we have to convert from float16 to float32 before doing k-best Viterbi decoding. Since this is the most time consuming part of inference, we would prefer to be able to benefit from float16 operations here.

@eric-haibin-lin
Copy link
Member Author

related issues: #12705 #11156 #11061

@anirudhacharya anirudhacharya mentioned this issue May 9, 2019
Closed
4 tasks
@anirudhacharya anirudhacharya mentioned this issue Jul 16, 2019
Merged
4 tasks
@fhieber fhieber mentioned this issue Apr 1, 2020
Open
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
CUDA Feature request FP16 Operator
Projects
None yet
Milestone
No milestone
3 participants
@eric-haibin-lin @stephenrawls @mxnet-label-bot