Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[BREAKING] prevent multi-gpu usage (#4749)

Browse Source 
* prevent multi-gpu usage

* fix distributed test

* combine gpu predictor tests

* set upper bound on n_gpus
This commit is contained in:
Rong Ou
2019-08-12 14:11:35 -07:00
committed by Rory Mitchell
parent 198f3a6c4a
commit c5b229632d
14 changed files with 59 additions and 298 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
tests/distributed/distributed_gpu.py
View File

@@ -66,27 +66,6 @@ def params_basic_1x4(rank):
}), 20
def params_basic_2x2(rank):
return dict(base_params, **{
'n_gpus': 2,
'gpu_id': 2*rank,
}), 20
def params_basic_4x1(rank):
return dict(base_params, **{
'n_gpus': 4,
'gpu_id': rank,
}), 20
def params_basic_asym(rank):
return dict(base_params, **{
'n_gpus': 1 if rank == 0 else 3,
'gpu_id': rank,
}), 20
rf_update_params = {
'subsample': 0.5,
'colsample_bynode': 0.5
@@ -103,11 +82,6 @@ def wrap_rf(params_fun):
params_rf_1x4 = wrap_rf(params_basic_1x4)
params_rf_2x2 = wrap_rf(params_basic_2x2)
params_rf_4x1 = wrap_rf(params_basic_4x1)
params_rf_asym = wrap_rf(params_basic_asym)
test_name = sys.argv[1]

20
tests/distributed/runtests-gpu.sh
View File

@@ -8,23 +8,5 @@ submit="timeout 30 python ../../dmlc-core/tracker/dmlc-submit"
echo -e "\n ====== 1. Basic distributed-gpu test with Python: 4 workers; 1 GPU per worker ====== \n"
$submit --num-workers=4 python distributed_gpu.py basic_1x4 || exit 1
echo -e "\n ====== 2. Basic distributed-gpu test with Python: 2 workers; 2 GPUs per worker ====== \n"
$submit --num-workers=2 python distributed_gpu.py basic_2x2 || exit 1
echo -e "\n ====== 3. Basic distributed-gpu test with Python: 2 workers; Rank 0: 1 GPU, Rank 1: 3 GPUs ====== \n"
$submit --num-workers=2 python distributed_gpu.py basic_asym || exit 1
echo -e "\n ====== 4. Basic distributed-gpu test with Python: 1 worker; 4 GPUs per worker ====== \n"
$submit --num-workers=1 python distributed_gpu.py basic_4x1 || exit 1
echo -e "\n ====== 5. RF distributed-gpu test with Python: 4 workers; 1 GPU per worker ====== \n"
echo -e "\n ====== 2. RF distributed-gpu test with Python: 4 workers; 1 GPU per worker ====== \n"
$submit --num-workers=4 python distributed_gpu.py rf_1x4 || exit 1
echo -e "\n ====== 6. RF distributed-gpu test with Python: 2 workers; 2 GPUs per worker ====== \n"
$submit --num-workers=2 python distributed_gpu.py rf_2x2 || exit 1
echo -e "\n ====== 7. RF distributed-gpu test with Python: 2 workers; Rank 0: 1 GPU, Rank 1: 3 GPUs ====== \n"
$submit --num-workers=2 python distributed_gpu.py rf_asym || exit 1
echo -e "\n ====== 8. RF distributed-gpu test with Python: 1 worker; 4 GPUs per worker ====== \n"
$submit --num-workers=1 python distributed_gpu.py rf_4x1 || exit 1