GLJeff comments

Repositories
Issues
Comments

Results 3 comments of


                                            GLJeff

In sm_simple.py, SM-G-SUM and SM-G-ABS scaling differ by sz^2

To further clarify: I believe both implementions are wrong in the sense that they are not finding a scaling vector independent of the number of states. SM-G-SUM should set: grad_output[:,...

IndexError: too many indices for array

You're using a later version of pytorch than they did. Just remove the [0]

fix 'invalid arguments' warp sync error on Volta

Is something like DeviceRadixSort even safe to run right now without your fix?