microsoft/DeepSpeed

Pull Requests Commits

fix schedule issue

jomayeri committed 364 days ago

ffeab32f

Cuda op to run sub op

jomayeri committed 1 year ago

b014d3c2

completions return lists

jomayeri committed 1 year ago

8532a922

first done wait policy, load balance schedule policy

jomayeri committed 1 year ago

3828954f

out of order completion queue for thread pools

jomayeri committed 1 year ago

77afaf64

developing our own cuda op

jomayeri committed 1 year ago

0c57fafd

a better approach to pools

jomayeri committed 1 year ago

1f23650f

correct iterator use

jomayeri committed 1 year ago

1e1d9302

adding thread pools

jomayeri committed 1 year ago

8c436210

adding hoooks for op parallel

jomayeri committed 1 year ago

92fa9a4f

Sanitize inputs to eval() (#6745)

loadams committed 1 year ago

Verified 9a2c209c

Update path for BingBertSquad from DeepSpeedExamples (#6746)

loadams committed 1 year ago

Verified 877aa0db

AIO File Offsets (#6641)

jomayeri committed 1 year ago

Verified b692cdea

add zero3 ```module_granularity_threshold ``` to zero optimization. (#6649)

inkcherry committed 1 year ago

Verified 7af3a4be

Add data type check for bf16 (#6742)

hwchen2017 committed 1 year ago

Verified 73d974ee

Add Domino code (#6733)

zhangsmallshark committed 1 year ago

Verified fabab197

Fix Type Name Inconsistency & Typo in cpu_adam (#6732)

xylian86 committed 1 year ago

Verified 99e9cbed

Update AMD apex version (#6739)

loadams committed 1 year ago

Verified b45ca263

Add COMMITTER file (#6741)

tjruwase committed 1 year ago

Verified b7e2ff50

Update GH hosted workflows to 24.04 (#6717)

loadams committed 1 year ago

Verified 08555662

Update version.txt after 0.15.4 release (#6731)

loadams committed 1 year ago

Verified 057d25be

Switch what versions of python are supported (#5676)

loadams committed 1 year ago

Verified a1b0c35a

Update flake8 version (#6722)

loadams committed 1 year ago

Verified 3beda32e

Update yapf version (#6721)

loadams committed 1 year ago

Verified d2a47189

Use one param coordinator for both train/inference scenarios (#6662)

tohtana committed 1 year ago

Verified 351569dd

[Bug Fix] Support threads_per_head < 64 for wavefront size of 64 (#6622)

jagadish-amd committed 1 year ago

Verified 2b41d621

Pin transformers to 4.45.2 in nv-ds-chat workflow (#6710)

loadams committed 1 year ago

Verified 6c08b7f9

Update URL in README Pipeline Status for Huawei Ascend NPU (#6706)

xuedinge233 committed 1 year ago

Verified 9068acb6

Explictly set device when reusing dist env (#6696)

tohtana committed 1 year ago

Verified b24dfa9d

Free memory in universal checkpointing tests (#6693)

tohtana committed 1 year ago

Verified 95ea95fc

Older