microsoft/DeepSpeed

Pull Requests Commits

return tuple fixes.

awan-10 committed 3 years ago

da1adf3c

Modify inference engine (#1520)

chunyang-wen committed 3 years ago

Verified f0122007

[unit tests] allow unique port for tests

jeffra committed 3 years ago

0af15b98

fstr for multnode_runner (#1532)

chunyang-wen committed 3 years ago

Verified 93c71831

[docs] fix 404 (#1531)

stas00 committed 3 years ago

Verified 76f2b5e5

typo in profiler.py (#1527)

ncfrey committed 3 years ago

Verified 2c62d657

Adding Tutel to MoE layer (#1528)

alexandremuzio committed 3 years ago

Verified 2887349c

Use fstr in launcher (#1521)

chunyang-wen committed 3 years ago

Verified cf1f1601

make conv flops counting general for 1,2,3d (#1518)

cli99 committed 3 years ago

Verified f9b37801

jeffra committed 3 years ago

426dd2b5

Prevent creation of local temp directory (#1494)

aphedges committed 3 years ago

Verified 91defd7c

Unify use f str (#1511)

chunyang-wen committed 3 years ago

Verified df5b0884

[code readability] pipe (#1510)

stas00 committed 3 years ago

Verified bf1725bb

Remove redundant pass (#1509)

chunyang-wen committed 3 years ago

Verified 85ce85dd

Bfloat16 zero2 (#1398)

raamjad committed 3 years ago

Verified 648f7bfa

Transformer kernel - fix unit test (#1503)

RezaYazdaniAminabadi committed 3 years ago

Verified 2c5bba6d

ZeRO3, improved parameter all-gather operation (#1188)

zarzen committed 3 years ago

Verified c0eeb69d

update CL doc (#1506)

conglongli committed 3 years ago

Verified 7f5a3add

allow passing hostfile to ds_ssh (#1504)

jeffra committed 3 years ago

Verified 163f568f

Fixing the transformer APIs to return tuple as the output (if needed) (#1491)

RezaYazdaniAminabadi committed 3 years ago

Verified ee6a92c0

[docs] Update DSE Megatron links (#1500)

tjruwase committed 3 years ago

Verified a4fff53d

Fix OneCycleLR zero division error (#1498)

AnesBenmerzoug committed 3 years ago

Verified db5d8ba2

Fix typo (#1501)

manuelciosici committed 3 years ago

Verified e976accb

enable/disable moe token dropping. (#1492)

awan-10 committed 3 years ago

Verified 56635d5b

Synchronize folder creation; Single latest file writer (#1486)

tjruwase committed 3 years ago

Verified 99bd592d

fix read-the-docs based on this issue: https://github.com/sphinx-doc/sphinx/issues/9727 (#1489)

jeffra committed 3 years ago

Verified 24dd285f

Proposal of how we might use sparse tensors for gradients (#1484)

Dipet committed 3 years ago

Verified fcb3ca5e

Add cpu adagrad (#1358)

fumihwh committed 3 years ago

Verified 8abdaee2

add moe+zero ckpt unit test. (#1429)

awan-10 committed 3 years ago

Verified 0b77d1d9

conglongli committed 3 years ago

Verified 29bee73f

Older