DeepSpeed
[RFC] add device abstraction to allow other device than CUDA be used
#2221
Merged
Merged
[RFC] add device abstraction to allow other device than CUDA be used #2221
jeffra
merged
86 commits into
deepspeedai:master
from
delock:gma/device-abstraction
[device abstraction] add device abstraction to allow other device tha…
0a849d56
Merge branch '202208-base' into 202208
e4f40f07
[rebase-202208] additional changes needed when rebase to 202208
4a216ea3
Merge branch '20220824-base' into 20220824
2137642f
[rebase] cleanup direct cuda usage after merge
089657eb
[precommit] fix pre-commit issues
d5a8424d
Merge branch 'master' into gma/device-abstraction
96d07659
[pin_memory] make pin_memory select device type
ac64c7ac
Merge branch 'master' into gma/device-abstraction
02c3a57c
[downstream] merge from xpu support downstream
522b24b4
Merge branch 'master' into gma/device-abstraction
a3b1e028
Merge branch 'master' into gma/device-abstraction
4557c33f
Merge branch 'up-master' into gma/merge-upstream-20220921
2ef7d6c3
[device] port cuda device to literal_device() in new tests
9656321b
[accel_runtime] add pin_memory to accelerator runtime interface.
65729e3c
[accelerator abstraction] merge from #2320
f94d53e7
Merge branch 'up-master' into gma/device-abstraction
6005abe6
change call site of literal_device, on_accel_device and accel_runtime…
31c0997f
add new interface definition from olruwase/accelerator_abstraction
1785c263
[accelerator abstraction] remove name() from interface, device_name()…
17203a4f
merge with master (ec13da6ba7cabc44bb4745a64a208b8580792954)
e8daea6d
Merge branch 'up-master' into gma/device-abstraction
cfd23ede
[OpBuilder] Add op builder abstraction
13bbbdf5
Merge branch 'up-master' into gma/device-abstraction
06e39a56
convert op builder usage in merged code
257490f2
[OpBuilder] add create_op_builder interface in abstract_accelerator.py
c93b9998
[OpBuilder] fix op builder usage in tests
9858d423
[OpBuilder] fix <op builder>.NAME usage in tests to follow op builder…
68ce0064
import get_accelerator from deepspeed.accelerator directly
4b62dab1
[OpBuilder] remove unused function and sync with main
c5b20702
add missing get_accelerator import
9532843d
fix obsolete name in CPU Adam which should be create_op_builder
07296951
fix create_op_builder calls
be517d8a
fix misuse of new accelerator abstraction interface in tests
3af870fa
Merge from downstream for bug fixing
8fa64b92
merge from downstream
48735388
delock
commented
on 2022-11-03
delock
commented
on 2022-11-03
remove SYCL_KERNEL specific code
61b10b06
Merge branch 'up-master(9cfcf7431a02a)' into gma/device-abstraction
457d2817
Merge branch 'up-master(6f77da1bae506)' into gma/device-abstraction
fea46040
Merge branch 'up-master(3ca9878d8e92a)' into gma/device-abstraction
f80a9075
merge from downstream for bugs fixes
3b0b14c1
Merge branch 'up-master(be5ec506bd5219a)' into gma/device-abstraction
b375e463
fix torch.cuda in new files
18b3c953
tjruwase
commented
on 2022-11-11
tjruwase
commented
on 2022-11-11
tjruwase
commented
on 2022-11-11
tjruwase
commented
on 2022-11-11
tjruwase
commented
on 2022-11-11
use OpBuilder name symbol, improve env_report, fix typo, fix get_acce…
97695f58
Merge branch 'master' into gma/device-abstraction
93e157bb
fix missing () in get_accelerator for ds_attention.py
b1c5384c
import deepspeed.accelerator.get_accelerator only when torch_availabl…
91fb9483
Merge branch 'up-master' into gma/device-abstraction
8f89c2b0
Change reference of InferenceSpecializedBuilder to name string, Infer…
26e628d8
convert new code with CUDA references
91f5cb22
remove unneeded get_accelerator import in op_builder/__init__.py
5a1ae0e3
[setup] fix build error when pytorch is not installed in environment
05842b64
Handle the case when torch is not installed during deepspeed installa…
24d2b38d
Merge branch 'master' into gma/device-abstraction
c26e5d43
Merge branch 'up-master' into gma/device-abstraction
4116ba51
port new cuda specific code
bea648fa
revert changes in __init__.py since new mechanism no longer requires …
94253d45
Merge branch 'up-master' into gma/device-abstraction
2acad489
use old op builder interface
77af66a9
Merge branch 'up-master' into gma/device-abstraction
8ec0905a
remove bypass code in set_accelerator_visible
bd9d275a
delock
commented
on 2023-01-27
delock
commented
on 2023-01-27
revert changes in quantizer according to latest op builder interface
f1e75ff9
Merge branch 'master' into gma/device-abstraction
98602825
port additional torch.cuda code in deepspeed
c26da463
Merge branch 'master' into gma/device-abstraction
cb46cf46
Merge branch 'master' into gma/device-abstraction
b74a47c5
Merge branch 'master' into gma/device-abstraction
6e557291
Merge branch 'master' into gma/device-abstraction
3c186d23
tjruwase
commented
on 2023-02-08
tjruwase
commented
on 2023-02-08
tjruwase
commented
on 2023-02-08
tjruwase
approved these changes
on 2023-02-08
follow comments
667c878a
Merge branch 'up-master' into gma/device-abstraction
d693dad7
fix format
7a9e7eab
fix new code with cuda specific code
538148b6
Merge branch 'master' into gma/device-abstraction
af8cee24
Merge branch 'master' into gma/device-abstraction
3dd816c6
Merge branch 'master' into gma/device-abstraction
abf31b65
Merge branch 'master' into gma/device-abstraction
9539defc
Merge branch 'master' into gma/device-abstraction
6ac4de4e
Merge branch 'master' into gma/device-abstraction
b5513045
port cuda specific code in module injection
238dc1e5
Merge branch 'master' into gma/device-abstraction
da254d76
Merge branch 'master' into gma/device-abstraction
33ace54f
Merge branch 'up-master' into gma/device-abstraction
3d572bb9
add licensing message
4f9f6c27
Merge branch 'master' into gma/device-abstraction
e92fd929
Merge branch 'master' into gma/device-abstraction
136ba270
Merge branch 'master' into gma/device-abstraction
9569b467
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub