stas00/ml-engineering

Issues

• Adding another logbook (kinda)

  #52 opened 25 days ago by boweiliu
  4

• Empirical measurements of sustained bandwidth and flops

  #77 opened a month ago by fluidnumerics-joe
  0

• AMD MI250X MAMF efficiency is wrong

  #76 opened a month ago by rlrs
  3

• How can operations with the INT8 data type be performed using a GPU accelerator card? Is dequantization required?

  #75 opened a month ago by cpollo55
  5

• MAMF Finder usage of MxNxK differs from BLAS gemm MxNxK

  #74 opened a month ago by GKolling
  2

• PDF link in readme doesn't work

  #73 opened a month ago by sytelus
  3

• GPU utilization monitoring

  #72 opened a month ago by fortminors
  4

• Performance Profiling

  #71 opened a month ago by jeromeku
  2

• [Question] `FSDP` vs `Deepspeed ZeRO3 / ZeRO++`

  #66 opened 2 months ago by jeromeku
  2

• bb

  #67 opened 2 months ago by sbhavani
  0

• f

  #68 opened 2 months ago by sbhavani
  0

• slurm job array change nodes

  #61 opened 3 months ago by ethanhe42
  1

• MAMF - GH200

  #58 opened 3 months ago by frankschae
  10

• Question about the right hidden dim when using SwiGLU

  #45 opened 7 months ago by Thytu
  3

• Question about changing precision post training

  #41 opened 7 months ago by Thytu
  2

• TPU v4 has 1,200GB/s of mem bandwidth and not 2,400, right?

  #40 opened 7 months ago by rodrigo-f-nogueira
  1

• ML

  #37 opened 8 months ago by lelikdr
  0

• discuss the solutions to Not fully recovering spikes

  #32 opened 8 months ago by pengzhangzhi
  7

• Conflicting opinions about streaming data from cloud storage?

  #30 opened 9 months ago by hacobe
  2

• Quarto Site

  #28 opened 9 months ago by saforem2
  3

• Improve folder structure

  #15 opened 9 months ago by heyimjonas
  3

• Clarification for gradient memory in mixed precision training

  #23 opened 9 months ago by SumanthRH
  3

• convert markdown to pdf

  #6 opened 10 months ago by pengzhangzhi
  10

• pip install -r build/requirements.txt fails due to github_md_utils

  #20 opened 10 months ago by ebowman
  3

• GPU requirements and cost estimation.

  #9 opened a year ago by Anindyadeep
  4

• Daisy chain batch jobs

  #13 opened a year ago by adammoody
  1

• Minor Typo in emulate multi node

  #8 opened a year ago by anindya-saha
  4

• Missing `hparams` section

  #5 opened a year ago by jvmncs
  2

• Convert to bfloat16 failing

  #2 opened a year ago by mhillebrand
  2

• Parallel training hangs

  #1 opened 3 years ago by mhillebrand
  10