Commits

Branch selector

 master

User selector

All users

All time

Commit history

Commits on May 16, 2023

• simplify logic

  sxjscience

  committed
  6083895

  Copy full SHA for 6083895

• fix lora

  sxjscience

  committed
  fb92d99

  Copy full SHA for fb92d99

Commits on May 15, 2023

• fix typo in comments with deepspeed/ (#3537)

  Show description for c8d3f5e
  

  3 peopleauthored
  c8d3f5e

  Copy full SHA for c8d3f5e

• Syncing FusedAdam with new Apex features (#3434)

  Show description for 9685eb9
  

  jomayeri

  and

  tjruwase

  authored
  9685eb9

  Copy full SHA for 9685eb9

• share inflight registry between PartitionedParameterCoordinators (#3462)

  Show description for 4716b0f
  

  HeyangQin

  and

  tjruwase

  authored
  4716b0f

  Copy full SHA for 4716b0f

• Fix broadcast error on multi-node training with ZeroStage3 and TensorParallel=2 (#2999)

  Show description for 9f4a876
  

  YizhouZ

  and

  tjruwase

  authored
  9f4a876

  Copy full SHA for 9f4a876

Commits on May 13, 2023

• Fix passing offload_optimizer_config instead of offload_param_config to (#3420)

  Show description for 5c6da1f
  

  3 peopleauthored
  5c6da1f

  Copy full SHA for 5c6da1f

Commits on May 12, 2023

• fix if statement for self.mpu (#3497)

  mrwyattii

  authored
  1824cb5

  Copy full SHA for 1824cb5

• Doc updates (#3520)

  Show description for 82991df
  

  goodship1

  and

  loadams

  authored
  82991df

  Copy full SHA for 82991df

• fix typo with docs/ (#3523)

  Show description for d73f99d
  

  digger-yu

  and

  loadams

  authored
  d73f99d

  Copy full SHA for d73f99d

• change actions/checkout@v2 to v3 (#3526)

  digger-yu

  authored
  b3956dc

  Copy full SHA for b3956dc

Commits on May 11, 2023

• Changing monitor loss to aggregate loss over gradient accumulation steps (#3428)

  Show description for 4d269c6
  

  jomayeri

  and

  tjruwase

  authored
  4d269c6

  Copy full SHA for 4d269c6

• Skip autoTP if tp_size is 1 (#3449)

  Show description for 5979ece
  

  3 peopleauthored
  5979ece

  Copy full SHA for 5979ece

• fix spelling error with deepspeed/runtime/ (#3509)

  Show description for 254663a
  

  digger-yu

  and

  loadams

  authored
  254663a

  Copy full SHA for 254663a

• Hybrid Engine Fix Llama (#3505)

  Show description for 194053b
  

  4 peopleauthored
  194053b

  Copy full SHA for 194053b

Commits on May 10, 2023

• Update README.md (#3504)

  Show description for b303fa8
  

  goodship1

  and

  loadams

  authored
  b303fa8

  Copy full SHA for b303fa8

• Add snip_momentum structured pruning which can support higher sparse ratio with minor accuracy loss (#3300)

  Show description for 6938c44
  

  ftian1

  and

  tjruwase

  authored
  6938c44

  Copy full SHA for 6938c44

• fix regression in shard checkpoint loading in AutoTP Path caused by qkv_copy() is deleted and add UT case for shard checkpoint loading in AutoTP (#3457)

  Show description for b31b46c
  sywangyi

  authored
  b31b46c

  Copy full SHA for b31b46c

• Update Inference Engine checkpoint loading + meta tensor assertions (#2940)

  Show description for db26f8b
  

  lekurile

  and

  awan-10

  authored
  db26f8b

  Copy full SHA for db26f8b

Commits on May 9, 2023

• delete the line to keep user_zero_stages (#3473)

  Show description for 99dfac9
  

  4 peopleauthored
  99dfac9

  Copy full SHA for 99dfac9

• fix spelling error with docs/index.md (#3443)

  Show description for 58c4d23
  

  digger-yu

  and

  tjruwase

  authored
  58c4d23

  Copy full SHA for 58c4d23

• fix spelling error with deepspeed/ (#3494)

  digger-yu

  authored
  1981664

  Copy full SHA for 1981664

• fix a small type error on bf16+pp (#3441)

  Show description for 195563a
  

  3 peopleauthored
  195563a

  Copy full SHA for 195563a

Commits on May 8, 2023

• fix spelling error (#3482)

  Show description for f1fab90
  

  luliyucoordinate

  and

  loadams

  authored
  f1fab90

  Copy full SHA for f1fab90

• Fix spelling errors in comments and documents (#3486)

  Show description for 87edbc8
  

  3 peopleauthored
  87edbc8

  Copy full SHA for 87edbc8

• Doc corrections (#3435)

  Show description for 4132118
  

  goodship1

  and

  loadams

  authored
  4132118

  Copy full SHA for 4132118

• Build: Update license in setup (#3484)

  PabloEmidio

  authored
  f3f4c44

  Copy full SHA for f3f4c44

Commits on May 7, 2023

• [Docs] Update automatic-tensor-parallelism.md (#3198)

  Show description for 9715d80
  

  3 peopleauthored
  9715d80

  Copy full SHA for 9715d80

Commits on May 5, 2023

• launcher/multinode_runner.py: mapping env variables in running cmd for mpich runner (#3372)

  Show description for 4e886f0
  

  3 peopleauthored
  4e886f0

  Copy full SHA for 4e886f0

Commits on May 4, 2023

• add sharded checkpoint loading for AutoTP path to reduce the peak mem… (#3102)

  Show description for d10b8ca
  

  sywangyi

  and

  tjruwase

  authored
  d10b8ca

  Copy full SHA for d10b8ca

• Hybrid Engine Refactor and Llama Inference Support (#3425)

  Show description for 0a61d5d
  

  cmikeh2

  and

  jeffra

  authored
  0a61d5d

  Copy full SHA for 0a61d5d

Commits on May 3, 2023

• add skip for mismatched cuda versions (#3436)

  mrwyattii

  authored
  d2e6110

  Copy full SHA for d2e6110

• Enable auto TP policy for llama model (#3170)

  Show description for 2c63e34
  

  3 peopleauthored
  2c63e34

  Copy full SHA for 2c63e34

• bump to 0.9.3

  jeffra

  committed
  5314c9a

  Copy full SHA for 5314c9a

• Update 2020-09-09-sparse-attention.md (#3432)

  Show description for e0e8085
  

  goodship1

  and

  loadams

  authored
  e0e8085

  Copy full SHA for e0e8085