Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Jan 16, 2025

• Fix PP+CP handling freqs_cis buffer

  Show description for d989842
  wconstab

  committed
  d989842

  Copy full SHA for d989842

• Disable integration test between optimizer-in-backward and pp (pytorch#793)

  Show description for f504a14
  mori360

  authored
  f504a14

  Copy full SHA for f504a14

• Do not aggregate the losses since last log step (pytorch#779)

  Show description for 2fa6d83
  carmocca

  authored
  2fa6d83

  Copy full SHA for 2fa6d83

Commits on Jan 13, 2025

• minor fix

  Show description for 82f7387
  tianyu-l

  committed
  82f7387

  Copy full SHA for 82f7387

Commits on Jan 10, 2025

• Add validation that batch size is divisible by number of microbatches

  Show description for 95677cb
  H-Huang

  committed
  95677cb

  Copy full SHA for 95677cb

• Stream subprocess stdout

  Show description for a04662f
  H-Huang

  committed
  a04662f

  Copy full SHA for a04662f

Commits on Jan 7, 2025

• fix num_microbatches input for PP

  Show description for 90567fc
  H-Huang

  committed
  90567fc

  Copy full SHA for 90567fc

• add CP pytorch forum post link

  Show description for a85a44c
  tianyu-l

  committed
  a85a44c

  Copy full SHA for a85a44c

Commits on Jan 2, 2025

• early all-reduce total_norm in non-PP grad norm clipping

  Show description for 2a44370
  tianyu-l

  committed
  2a44370

  Copy full SHA for 2a44370

Commits on Dec 27, 2024

• mild README revamp

  Show description for 3f20451
  tianyu-l

  committed
  3f20451

  Copy full SHA for 3f20451

• fix loss with CP enabled

  Show description for 6b03ffa
  tianyu-l

  committed
  6b03ffa

  Copy full SHA for 6b03ffa

Commits on Dec 23, 2024

• [BE] improve argparser

  Show description for f6a9daa
  tianyu-l

  committed
  f6a9daa

  Copy full SHA for f6a9daa

Commits on Dec 20, 2024

• [BE] Combine OptimizerWrapper and OptimizerContainer (pytorch#738)

  Show description for ba24697
  mori360

  authored
  ba24697

  Copy full SHA for ba24697

Commits on Dec 19, 2024

• Accessible config as dict (pytorch#754)

  Show description for d67f7f9
  casper-hansen

  authored
  d67f7f9

  Copy full SHA for d67f7f9

• turning off compile on loss function

  Show description for e8977b0
  tianyu-l

  committed
  e8977b0

  Copy full SHA for e8977b0

Commits on Dec 18, 2024

• [BE] consolidate 4-GPU integration tests into 8-GPU tests and reduce frequency

  Show description for 6274377
  tianyu-l

  committed
  6274377

  Copy full SHA for 6274377

• [BE] add integration test for the generation script

  Show description for cd57683
  tianyu-l

  committed
  cd57683

  Copy full SHA for cd57683

• [BE] restructure tests and assets folders

  Show description for b0a80df
  tianyu-l

  committed
  b0a80df

  Copy full SHA for b0a80df

Commits on Dec 17, 2024

• [BE] dump compile trace to CI output for debugging, and reduce CI workload

  Show description for 00f5302
  tianyu-l

  committed
  00f5302

  Copy full SHA for 00f5302

• Run init_weights under no_grad (pytorch#747)

  Show description for 5ce8a0c
  carmocca

  authored
  5ce8a0c

  Copy full SHA for 5ce8a0c

Commits on Dec 16, 2024

• Explain TORCH_NCCL_AVOID_RECORD_STREAMS=1 for TP

  Show description for 4f7f883
  awgu

  committed
  4f7f883

  Copy full SHA for 4f7f883

• replace create_seed_checkpoint.md with a note in docs/checkpoint.md

  Show description for 3bf47d6
  tianyu-l

  committed
  3bf47d6

  Copy full SHA for 3bf47d6

Commits on Dec 12, 2024

• PYTHONHASHSEED needs to be in the range of [0,2**32 - 1]

  Show description for e846b69
  tianyu-l

  committed
  e846b69

  Copy full SHA for e846b69

• [cp] add option to choose kv shards rotation method (pytorch#728)

  Show description for c97876e
  XilunWu

  authored
  c97876e

  Copy full SHA for c97876e

• some cleanups on docs and parallel_dims

  Show description for 0186284
  tianyu-l

  committed
  0186284

  Copy full SHA for 0186284

Commits on Dec 11, 2024

• [cp] apply fsdp to model when CP is enabled without DP for correct loss and lower mem usage (pytorch#685)

  Show description for 40a0873
  XilunWu

  authored
  40a0873

  Copy full SHA for 40a0873

• [DP] change device mesh dim naming convention to make it more consistent (pytorch#720)

  Show description for cb633e3
  XilunWu

  authored
  cb633e3

  Copy full SHA for cb633e3

• test csv schedule on different runtime

  Show description for 03f17e8
  H-Huang

  committed
  03f17e8

  Copy full SHA for 03f17e8

Commits on Dec 9, 2024

• Configure RNGs appropriately for Pipeline + SPMD

  Show description for 05a8b5e
  wconstab

  committed
  05a8b5e

  Copy full SHA for 05a8b5e

• Enable optimizer in backward in TorchTitan (pytorch#705)

  Show description for 2735ced
  mori360

  authored
  2735ced

  Copy full SHA for 2735ced

• remove unused tp_enabled parameter from apply_fsdp (pytorch#722)

  Show description for 4b9a86a
  Muhtasham

  authored
  4b9a86a

  Copy full SHA for 4b9a86a

Commits on Dec 5, 2024

• Custom Dataset refactoring + docs (pytorch#715)

  Show description for 7281e0b
  msaroufim

  authored
  7281e0b

  Copy full SHA for 7281e0b

• Add checkpoint load step (pytorch#716)

  Show description for afa8229
  msaroufim

  authored
  afa8229

  Copy full SHA for afa8229

• Update readme with cu124 version

  Show description for 0ed31f5
  wconstab

  committed
  0ed31f5

  Copy full SHA for 0ed31f5

Commits on Dec 4, 2024

• Migrate CI to CUDA 12.4

  Show description for fbd07ec
  wconstab

  committed
  fbd07ec

  Copy full SHA for fbd07ec

Pagination

Previous

Next