DeepSpeed
encoded ds config into command line argument when launching child processes in autotuning
#2524
Merged

encoded ds config into command line argument when launching child processes in autotuning #2524

cli99 merged 20 commits into master from encoded-ds-config
cli99
cli992 years ago (edited 2 years ago)

This PR encodes the autotuning experiments ds configurations passed from the master tuning process to the child process. This is a required fix for AML environment, where child process cannot read a path wrote by the master process on a shared storage.

Huggingface assumes deepspeed_config is either a dictionary or a file path, to make it work with the PR for autotuning, the check needs to be updated as well.

cli99 rollback ds config changes
d5e9a3cb
cli99 fix format
8f6517ad
cli99 Fix error when output_file is a relative path without a prefix (#2397)
b685aeb1
cli99 fix restuls and exprs path to use absolute path
f8c95138
cli99 use base64 encoded ds config as cmd arg
cd0f6eae
cli99 fix format
154890b1
cli99 remove assert
fb86d551
tjruwase Merge branch 'master' into encoded-ds-config
a9198073
cli99 write out optimial config after tuning
20c1b1a1
cli99 fix format
d3f34920
cli99 no need to update ds config path when encoding ds config
0154d346
cli99 udpate
14080499
cli99 merge with master
27e1db45
cli99 cli99 requested a review from jeffra jeffra 2 years ago
cli99 cli99 requested a review from tjruwase tjruwase 2 years ago
cli99 do not use abs path for result and expr dir
1a9623e6
cli99 fix conflicts
c098e687
cli99 fix run mode
7df118e3
cli99 Merge branch 'master' into encoded-ds-config
73a29574
cli99 cli99 marked this pull request as ready for review 2 years ago
cli99 fix format
d6104115
cli99 fix format
b833a798
mrwyattii
mrwyattii approved these changes on 2022-11-22
tjruwase Merge branch 'master' into encoded-ds-config
35b65085
cli99 cli99 merged abe4fc6b into master 2 years ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone