Commits

Branch selector

 master

User selector

adamlerer

Datepicker

All time

Commit History

Commits on Mar 18, 2020

• Half-precision just for MM

  Show description for bbf1e7a
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  bbf1e7a

  Copy full SHA for bbf1e7a

• Add more performance logging

  Show description for c7dea47
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  c7dea47

  Copy full SHA for c7dea47

• GPU eval during training (fast version)

  Show description for 98081ea
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  98081ea

  Copy full SHA for 98081ea

• Temporarily use ctypes for memory pinning because it was broken by D20249187

  Show description for baaf5df
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  baaf5df

  Copy full SHA for baaf5df

Commits on Mar 13, 2020

• Use barrier to join parameter servers when possible, which is safer

  Show description for 30588bb
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  30588bb

  Copy full SHA for 30588bb

Commits on Mar 11, 2020

• Object-oriented refactor of train_and_report_stats

  Show description for f9b8b72
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  f9b8b72

  Copy full SHA for f9b8b72

• GPU mode as separate train_gpu.py

  Show description for 529442d
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  529442d

  Copy full SHA for 529442d

• Simplify RowAdagrad (remove lr_decay)

  Show description for 4c6557c
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  4c6557c

  Copy full SHA for 4c6557c

• Make stuff accept both CPU and CUDA tensors

  Show description for 53d87ec
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  53d87ec

  Copy full SHA for 53d87ec

• Use multiple ProcessGroups for higher bandwidth

  Show description for 96f96fc
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  96f96fc

  Copy full SHA for 96f96fc

• Allocate shared memory for embeddings only once at the beginning

  Show description for f2b44a6
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  f2b44a6

  Copy full SHA for f2b44a6

• Remove background I/O

  Show description for 23349b5
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  23349b5

  Copy full SHA for 23349b5

• Make Filament2's FBLearner flow launch partition servers if requested

  Show description for e9142c9
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  e9142c9

  Copy full SHA for e9142c9

• tiny bugfix

  Show description for 3c53aa9
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  3c53aa9

  Copy full SHA for 3c53aa9

Commits on Feb 13, 2020

• Support multiple edgelist types beyond TSV (facebookresearch#120)

  Show description for fb0b3ae
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  fb0b3ae

  Copy full SHA for fb0b3ae

Commits on Sep 2, 2019

• Add new functions to TensorList

  adamlerer

  committed
  8fe90c6

  Copy full SHA for 8fe90c6

Commits on Jul 12, 2019

• Update README.md (facebookresearch#83)

  Show description for 62d3a05
  

  adamlerer

  authored and

  facebook-github-bot

  committed
  62d3a05

  Copy full SHA for 62d3a05