Add GPU support to NVFlare demo (#9552)

dmlc · Sep 6, 2023 · 0f35493 · 0f35493
1 parent 3b9e590
commit 0f35493
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 7 deletions.
diff --git a/demo/nvflare/horizontal/README.md b/demo/nvflare/horizontal/README.md
@@ -85,8 +85,8 @@ shutdown server
 ## Training with GPUs
 
 To demo with Federated Learning using GPUs, make sure your machine has at least 2 GPUs.
-Build XGBoost with the federated learning plugin enabled along with CUDA, but with NCCL
-turned off (see the [README](../../plugin/federated/README.md)).
+Build XGBoost with the federated learning plugin enabled along with CUDA
+(see the [README](../../plugin/federated/README.md)).
 
-Modify `config/config_fed_client.json` and set `use_gpus` to `true`, then repeat the steps
+Modify `../config/config_fed_client.json` and set `use_gpus` to `true`, then repeat the steps
 above.
diff --git a/demo/nvflare/horizontal/custom/trainer.py b/demo/nvflare/horizontal/custom/trainer.py
@@ -67,7 +67,7 @@ def _do_training(self, fl_ctx: FLContext):
  dtest = xgb.DMatrix('agaricus.txt.test?format=libsvm')
 
  # Specify parameters via map, definition are same as c++ version
- param = {'max_depth': 2, 'eta': 1, 'objective': 'binary:logistic'}
+ param = {'tree_method': 'hist', 'max_depth': 2, 'eta': 1, 'objective': 'binary:logistic'}
  if self._use_gpus:
  self.log_info(fl_ctx, f'Training with GPU {rank}')
  param['device'] = f"cuda:{rank}"

diff --git a/demo/nvflare/vertical/README.md b/demo/nvflare/vertical/README.md
@@ -56,4 +56,9 @@ shutdown server
 
 ## Training with GPUs
 
-Currently GPUs are not yet supported by vertical federated XGBoost.
+To demo with Vertical Federated Learning using GPUs, make sure your machine has at least 2 GPUs.
+Build XGBoost with the federated learning plugin enabled along with CUDA
+(see the [README](../../plugin/federated/README.md)).
+
+Modify `../config/config_fed_client.json` and set `use_gpus` to `true`, then repeat the steps
+above.
diff --git a/demo/nvflare/vertical/custom/trainer.py b/demo/nvflare/vertical/custom/trainer.py
@@ -77,13 +77,15 @@ def _do_training(self, fl_ctx: FLContext):
  'gamma': 1.0,
  'max_depth': 8,
  'min_child_weight': 100,
- 'tree_method': 'approx',
+ 'tree_method': 'hist',
  'grow_policy': 'depthwise',
  'objective': 'binary:logistic',
  'eval_metric': 'auc',
  }
  if self._use_gpus:
- self.log_info(fl_ctx, 'GPUs are not currently supported by vertical federated XGBoost')
+ if self._use_gpus:
+ self.log_info(fl_ctx, f'Training with GPU {rank}')
+ param['device'] = f"cuda:{rank}"
 
  # specify validations set to watch performance
  watchlist = [(dtest, "eval"), (dtrain, "train")]