Closed RandomY-2 closed 1 year ago
Merging #52 (9d44325) into main (c329d28) will decrease coverage by
23.81%
. The diff coverage is27.83%
.
@@ Coverage Diff @@
## main #52 +/- ##
===========================================
- Coverage 93.00% 69.20% -23.81%
===========================================
Files 42 43 +1
Lines 3361 3455 +94
Branches 376 364 -12
===========================================
- Hits 3126 2391 -735
- Misses 155 948 +793
- Partials 80 116 +36
Flag | Coverage Δ | |
---|---|---|
unittests | 69.20% <27.83%> (-23.75%) |
:arrow_down: |
Flags with carried forward coverage won't be shown. Click here to find out more.
Files Changed | Coverage Δ | |
---|---|---|
python/xoscar/collective/xoscar_cupy.py | 27.83% <27.83%> (ø) |
Thanks to @RandomY-2 for his contribution. Work on the current PR has been moved to #56 , so this PR is closed for now.
What do these changes do?
Add GPU collective communication using cupy NcclCommunicator (similar to alpa: https://github.com/alpa-projects/alpa/blob/main/alpa/collective/collective_group/nccl_collective_group.py but tried to implement missing apis). Will implement a process group class once #46 is finished.
test cases are similar to pygloo test cases: https://github.com/xorbitsai/xoscar/blob/main/python/xoscar/collective/tests/test_pygloo.py
Related issue number
Related #22
Check code requirements