DeepSpeed
[inference] test suite for ds-kernels (bert, roberta, gpt2, gpt-neo, gpt-j)
#1992
Merged
Merged
[inference] test suite for ds-kernels (bert, roberta, gpt2, gpt-neo, gpt-j) #1992
jeffra
merged
48 commits into
master
from
jeffra/fill-mask-tests
add fill masks tests
bcd731c9
add a temporary fix for the Bert-Type models; TODO: Fix this for the …
19158267
add cuda graph to tests
965d109b
jeffra
changed the title
add fp16/32
05a9ca77
add q&a tests and set hf framework to pt
2acba69b
Merge branch 'master' into jeffra/fill-mask-tests
7822770a
jeffra
changed the title 
fixed assignment of DS pipeline model
5724f7e7
generalized the model/task testing and added pytest fixtures
ef350e9b
fixed syntax error
9a934fe5
keyerror with text-gen assert_fn
fbc05fef
added back gptj model
fc569e9f
Merge branch 'master' into jeffra/fill-mask-tests
fbdd6b86
added temporary fix to get gpt models passing
9c06f04c
fixing some issue for the Bert-Postln model
c10acaf4
Merge branch 'jeffra/fill-mask-tests' of github.com:microsoft/DeepSpe…
4440aedf
formatting
0519a723
added fp16 tests
31754302
fixed bad comparison
9cdd9814
avoid CUDA init when getting GPU memory
3d93c078
revert to not checking GPU memory
d864d2b1
extra import
15ed0a25
refactored validation of test config, add skip for AMD w/CUDA Graph
2c88ab1a
cuda graph requires 1.10+
0872f9f7
ensure torch > 1.10 if cuda graph enabled, also fix typo
1d08254f
prune incompatible model+task pairs
816e746a
add pretty print names for test fixtures
42841c6e
use fp16 revision of gpt-j
2677d20c
convert pipe model to half for non-DS case
3ec3f9f8
Merge branch 'master' into jeffra/fill-mask-tests
d37f9d9f
jeffra
changed the title
remove dep on hf-accelerate
8b88eeff
manual convert gpt-j fp32->fp16, seeing issues with fp16 revision
50cbfced
force inference tests to be seq
f3e40aea
added correctness tests for text-generation gpt models
d3f6dfe9
fixed device string for lm tests
71a58405
during injection reshape data on cpu to avoid extra GPU memory
f99fc0d7
increased acceptable ppl diff val
1d21343a
added perf test and fix for gpt-j-6B OOM on V100 16GB
2777c23e
remove sequential and move nv-nightly to nightly
eea4c1e2
Merge branch 'master' into jeffra/fill-mask-tests
c0d695c1
Merge branch 'master' into jeffra/fill-mask-tests
b6b6682d
skip inference tests on rocm for now
c525af27
add missing import
e4573b2a
RezaYazdaniAminabadi
approved these changes
on 2022-06-15
make performance asserts less strict
f01df0f1
fix for datasets error
368427b9
seperate inference tests from regular unit tests
3a4e12c6
added new req file for inference workflow
0ea845d1
skip test if deps aren't installed
2043d576
fix for inf reqs install
21d4bebb
mrwyattii
approved these changes
on 2022-06-15
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub