Enable BNB multi-backend support #31098
Titus-von-Koeller
merged
66 commits into
huggingface:main
from
jiqing-feng:bnb_cpu
Titus-von-Koeller
approved these changes
on 2024-07-16
SunMarc
commented
on 2024-08-27
enable cpu bnb path
846f8535
fix style
6c567037
fix code style
3f02c9be
fix 4 bit path
9ccbf109
Update src/transformers/utils/import_utils.py
89fa5ef4
add multi backend refactor tests
a52d7af1
fix style
6f678625
tweak 4bit quantizer + fix corresponding tests
ee23eb0a
tweak 8bit quantizer + *try* fixing corresponding tests
678e6732
fix dequant bnb 8bit
0858b3e7
account for Intel CPU in variability of expected outputs
c76d2430
enable cpu and xpu device map
5843f281
further tweaks to account for Intel CPU
1a864a86
fix autocast to work with both cpu + cuda
f3753fc8
fix comments
0cc1b7ea
fix comments
b6118126
switch to testing_utils.torch_device
ab4836e4
allow for xpu in multi-gpu tests
73995002
fix tests 4bit for CPU NF4
b41059c7
fix bug with is_torch_xpu_available needing to be called as func
1a7a6fe7
avoid issue where test reports attr err due to other failure
87983df8
fix formatting
7f17188b
fix typo from resolving of merge conflict
bb3ba4a2
polish based on last PR review
463c2111
fix CI
6d89ee4f
Update src/transformers/integrations/integration_utils.py
7e01cfb0
Update src/transformers/integrations/integration_utils.py
9bffc93d
ArthurZucker
approved these changes
on 2024-08-29
fix error log
01b7587c
fix error msg
171b130d
add \n in error log
5e9bf9a9
make quality
496c0467
rm bnb cuda restriction in doc
86d0016a
cpu model don't need dispatch
1c96ae90
Merge branch 'main' into bnb_cpu
495354e1
fix doc
3aec6267
fix style
daa1e275
SunMarc
commented
on 2024-09-03
check cuda avaliable in testing
d55db0e7
fix tests
a21a9168
SunMarc
approved these changes
on 2024-09-10
akx
commented
on 2024-09-11
Update docs/source/en/model_doc/chameleon.md
8ad17e82
Update docs/source/en/model_doc/llava_next.md
107e02b0
Update tests/quantization/bnb/test_4bit.py
20f6b5e2
Update tests/quantization/bnb/test_4bit.py
9ac038ef
fix doc
3bab7d76
Merge branch 'huggingface:main' into bnb_cpu
968d9c5b
fix check multibackends
08f31f87
fix import sort
9eb09705
remove check torch in bnb
b506b98e
docs: update bitsandbytes references with multi-backend info
2be41697
docs: fix small mistakes in bnb paragraph
e607b7c3
run formatting
ac108c6d
Merge remote-tracking branch 'origin/main' into bnb_cpu
82dcb0d1
reveret bnb check
c66e7e75
SunMarc
commented
on 2024-09-13
move bnb multi-backend check to import_utils
8f25ee26
akx
commented
on 2024-09-14
Update src/transformers/utils/import_utils.py
a4333cb3
fix bnb check
32cbb8d4
minor fix for bnb
4ce4b558
check lib first
937ed3bf
fix code style
e40f2847
Merge branch 'huggingface:main' into bnb_cpu
03dd03b7
Revert "run formatting"
b8093ce8
fix format
0551d239
SunMarc
commented
on 2024-09-17
give warning when bnb version is low and no cuda found]
e33e43bb
Merge branch 'huggingface:main' into bnb_cpu
ced3c288
fix device assignment check to be multi-device capable
170dd585
address akx feedback on get_avlbl_dev fn
9ba4a5e7
we don't want the function tat publicc, as docs would be too much
594f6f8a
bimehasia25815
commented
on 2024-12-28
bimehasia25815
commented
on 2024-12-28
Reviewers
Assignees
No one assigned
Labels
Quantization
Milestone
No milestone
Login to write a write a comment.
Login via GitHub