microsoft/DeepSpeed

Pull Requests Commits

Fix ZeRO-1 checkpointing bug (#529)

tjruwase committed 4 years ago

Verified 4f5f83e4

Make elastic checkpointing optional (#526)

tjruwase committed 4 years ago

Verified 9c577ee0

Hotfix ZeRO-1 memory/convergence fixes (#527)

jeffra committed 4 years ago

Verified fba92f19

revert model test disable

jeffra committed 4 years ago

Verified 0fda51ad

disable model tests

jeffra committed 4 years ago

Verified e09f4f89

Turn off multi-node launch if only 1 node (#322)

jeffra committed 4 years ago

Verified e69b1eec

bump DSE (#317)

jeffra committed 4 years ago

Verified de0523d4

attach empty grad to its param to ensure it's copied after reduction (#316)

jeffra committed 4 years ago

Verified e1bea67f

Update fan out flag for pdsh (#315)

jeffra committed 4 years ago

Verified 6855ba1c

Update deepspeed_lr_schedules.py (#314)

jeffra committed 4 years ago

Verified 3437342c

bumping DSE commit for pillow security fix (#312)

Shaden Smith committed 4 years ago

Verified 892ece67

Fix+tests for get_lr from lr_scheduler before training starts (#310)

jeffra committed 4 years ago

Verified cd68e6e5

updates website gems after kramdown alert (#311)

Shaden Smith committed 4 years ago

Verified 903a41a5

Add webinar link (#309)

jeffra committed 4 years ago

Verified 29c5fe26

Removing () from assertion. (#307)

Shaden Smith committed 4 years ago

Verified c35e9441

NameError: name 'mpu' is not defined (#305)

MannyKayy committed 4 years ago

Verified 9d07d756

Fix nv_peer_mem version (#304)

jeffra committed 4 years ago

Verified e50b8838

Fixing a typo (#303)

MannyKayy committed 4 years ago

Verified 97c54273

Update README.md (#302)

jeffra committed 4 years ago

Verified 67821f95

DeepSpeed webinar announcement (#301)

Shaden Smith committed 4 years ago

Verified 7ae8f8bc

bump DSExamples (#300)

jeffra committed 4 years ago

Verified 0f94f7e0

pass steps_per_print to tput timer (#299)

jeffra committed 4 years ago

Verified ec943410

updates to amp to support grad clip and grad accumulation (#290)

jeffra committed 4 years ago

Verified eb74c3f1

Avoid deadlock for unsynchronized non-zero checkpointing (#297)

tjruwase committed 4 years ago

Verified 3cc96e17

Update setup.py (#298)

jeffra committed 4 years ago

Verified 871f7e63

only global rank 0 can log tensorboard data; avoid multi gpu/node race for the log directory (#296)

arashashari committed 4 years ago

Verified 1f972427

Empty grad fix (#291)

jeffra committed 4 years ago

Verified 376818ef

Fix bug in fp32 optimizer state loading (#289)

tjruwase committed 4 years ago

Verified 607814fe

Support loading and saving ZeRO checkpoints with changing DP degree (#240)

tjruwase committed 4 years ago

Verified 7ccc9daf

Update amp docs (#287)

jeffra committed 4 years ago

Verified 366d8816

Older