Mayank Mishra

icon indicating a website link https://mayank31398.github.io/

icon company MIT-IBM Watson AI Lab icon location Boston icon location Interested in Efficient Models

Results 187 comments of Mayank Mishra

fix for fp16

@mikaylagawarecki I have fixed the blotched rebase.

fix for fp16

> nit: could we use `OpMath` rather than hardcoding the `half`, `float` case here? e.g., > > https://github.com/pytorch/pytorch/blob/bf7db4e4f9f07765e63d9ce1c98c5962bafe4608/aten/src/ATen/OpMathType.h#L16 @eqy I dont understand how to use this. Can you give an...

fix for fp16

@eqy @mikaylagawarecki I am done with the changes and all cases pass now. I am not sure how to use the OpMathType. Can one of you push changes for that?

fix for fp16

@eqy @mikaylagawarecki pinging again

fix for fp16

guys can we get this merged?

fix for fp16

@eqy @kkontny my knowledge of C++ isnt great ```cpp using opmath_t = opmath_type; upcasted_input = input.to(opmath_t()) ``` this doesn't compile, can you push a fix?

fix for fp16

@eqy figured it out I think its fixed now.

fix for fp16

@eqy pinging again

fix for fp16

failing tests seem unrelated @eqy @mikaylagawarecki

fix for fp16

@eqy any updates on this?