Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 29, 2024

• [cp] fix the device mesh access issue when CP is not used with DP (pytorch#683)

  Show description for 3e3909a
  XilunWu

  authored
  3e3909a

  Copy full SHA for 3e3909a

Commits on Nov 27, 2024

• Fix upper calls (pytorch#702)

  Show description for 2dd0008
  carmocca

  authored
  2dd0008

  Copy full SHA for 2dd0008

• fix error in pytorch#701 (pytorch#704)

  Show description for 87e2c09
  vkuzo

  authored
  87e2c09

  Copy full SHA for 87e2c09

Commits on Nov 26, 2024

• add max and abs to selective AC config (pytorch#701)

  Show description for bcebc0d
  vkuzo

  authored
  bcebc0d

  Copy full SHA for bcebc0d

Commits on Nov 25, 2024

• some cleanups

  Show description for 4d182a1
  tianyu-l

  committed
  4d182a1

  Copy full SHA for 4d182a1

• recommended practices for loss converging tests

  Show description for b39b8a4
  tianyu-l

  committed
  b39b8a4

  Copy full SHA for b39b8a4

• fix float8 delayed scaling configuration (pytorch#698)

  Show description for aff9494
  vkuzo

  authored
  aff9494

  Copy full SHA for aff9494

Commits on Nov 22, 2024

• [pp] Add support for loading schedule csv

  Show description for 5525d77
  H-Huang

  committed
  5525d77

  Copy full SHA for 5525d77

• ensure reproducible determinsitc numerics (pytorch#597)

  Show description for b00657f
  weifengpy

  authored
  b00657f

  Copy full SHA for b00657f

Commits on Nov 21, 2024

• single-host generation for integration testing (pytorch#640)

  Show description for 1c0832e
  jaysonfrancis

  authored
  1c0832e

  Copy full SHA for 1c0832e

• support 3rd-party backend (pytorch#672)

  Show description for 85f6853
  

  3 peopleauthored
  85f6853

  Copy full SHA for 85f6853

Commits on Nov 16, 2024

• Fix PP clip_grad_norm (pytorch#649)

  Show description for 046de56
  zijian-hu

  authored
  046de56

  Copy full SHA for 046de56

Commits on Nov 5, 2024

• Add script to convert pickled Llama weights to DCP (pytorch#634)

  Show description for 3247841
  rlrs

  authored
  3247841

  Copy full SHA for 3247841

Commits on Oct 31, 2024

• [BE] replace the extra DeviceMesh _flatten with mesh access (pytorch#667)

  Show description for d7cabfb
  XilunWu

  authored
  d7cabfb

  Copy full SHA for d7cabfb

• Add test for toml-based pp split points

  Show description for e1fbced
  wconstab

  committed
  e1fbced

  Copy full SHA for e1fbced

Commits on Oct 30, 2024

• [BE] remove old pytorch version warning on strided sharding since 2.5 is official released (pytorch#665)

  Show description for 2a785e9
  XilunWu

  authored
  2a785e9

  Copy full SHA for 2a785e9

• Fix parsing of pp split points from toml file

  Show description for 53d0f69
  wconstab

  committed
  53d0f69

  Copy full SHA for 53d0f69

Commits on Oct 29, 2024

• Fix data_parallel_shard_degree description (pytorch#659)

  Show description for dbb0520
  carmocca

  authored
  dbb0520

  Copy full SHA for dbb0520

Commits on Oct 28, 2024

• add paper citation

  Show description for 7310abe
  tianyu-l

  committed
  7310abe

  Copy full SHA for 7310abe

• Enable FSDP2 cpu offloading (pytorch#624)

  Show description for 193ce98
  mori360

  authored
  193ce98

  Copy full SHA for 193ce98

• Port pytorch#642's loss changes to estimation.py (pytorch#656)

  Show description for 603889a
  carmocca

  authored
  603889a

  Copy full SHA for 603889a

• Do not destroy if the world did not init (pytorch#655)

  Show description for eae7f10
  

  carmocca

  and

  awgu

  authored
  eae7f10

  Copy full SHA for eae7f10

Commits on Oct 25, 2024

• [Multimodal model] Add the decoder of the multimodal model (pytorch#626)

  Show description for ccfc02b
  fduwjj

  authored
  ccfc02b

  Copy full SHA for ccfc02b

Commits on Oct 23, 2024

• Workaround for pytorch/pytorch#138575 distributed checkpoint loading bug (pytorch#647)

  Show description for 1060fea
  

  arjvik

  and

  awgu

  authored
  1060fea

  Copy full SHA for 1060fea

• enable Context Parallel (pytorch#592)

  Show description for b19456a
  XilunWu

  authored
  b19456a

  Copy full SHA for b19456a

Commits on Oct 22, 2024

• Use expandable segments in run_llama_train.sh

  Show description for a5e57e4
  awgu

  committed
  a5e57e4

  Copy full SHA for a5e57e4

• Various cleanups around distributed setup

  Show description for e211236
  awgu

  committed
  e211236

  Copy full SHA for e211236

• Move output .float() to loss fn and compile it if compiling

  Show description for e10cb94
  awgu

  committed
  e10cb94

  Copy full SHA for e10cb94

• Remove last instances of data_parallel_degree

  Show description for c9b17cc
  awgu

  committed
  c9b17cc

  Copy full SHA for c9b17cc

Commits on Oct 18, 2024

• Add H200 bf16 peak flops (pytorch#627)

  Show description for 0edd2fb
  carmocca

  authored
  0edd2fb

  Copy full SHA for 0edd2fb

• [BE][doc]Update fsdp readme, GradScaler enabled for FSDP2 (pytorch#629)

  Show description for 511bd47
  mori360

  authored
  511bd47

  Copy full SHA for 511bd47

Commits on Oct 16, 2024

• [PP] fix CI

  Show description for 36fba84
  H-Huang

  committed
  36fba84

  Copy full SHA for 36fba84

• Add a ViT Encoder to TorchTitan (pytorch#589)

  Show description for 3858dc9
  fduwjj

  authored
  3858dc9

  Copy full SHA for 3858dc9

Commits on Oct 15, 2024

• add arxiv badge and link to torchtitan paper (pytorch#618)

  Show description for 7c3a1fa
  lessw2020

  authored
  7c3a1fa

  Copy full SHA for 7c3a1fa

Commits on Oct 14, 2024

• Delete manual pp input/output tensors after adding shape inference

  Show description for 1629fb9
  wconstab

  committed
  1629fb9

  Copy full SHA for 1629fb9

Pagination

Previous

Next