Vassilis Papadopoulos comments

Repositories
Issues
Comments

Results 4 comments of


                                            Vassilis Papadopoulos

Tokenizer training sticks for long times.

Same problem, stuck on compute merges for 10+hrs when all the other steps took ~ 3hrs. Dataset size if 300Gb, English. CPU is at 1000%

CUDA Illegal memory access on CrossEntropyLoss with large batch size, cu113, torch 1.12.1

Can confirm I encountered the same issue in pytorch 2.x !! More specifically, in my case it happens when computing CELoss between a tensor of shape (B,C,T) and a truth...

SVG exported from W&B is not displayed in Overleaf

Still a problem as of June 2024.

PowerShell Command Execution Bug in Cursor IDE (Composer - Agent)

Bump, I have the same problem. Truly a super annoying bug, and should be easily fixable. Just tell us where to configure where the calls to the terminal are made.