[CI] Bump mosaicml version from 0.10 to 0.12 (ray-project#33080)

ray-project#31553 requires torchmetrics>=0.8, but the current mosaicml version requires torchmetrics<8. This PR upgrades the mosaicml version.

---------

Signed-off-by: Balaji Veeramani <balaji@anyscale.com>

.buildkite/pipeline.build.yml

@@ -577,7 +577,7 @@
     # Horovod needs to be installed separately (needed for API ref imports)
     - ./ci/env/install-horovod.sh
     # See https://stackoverflow.com/questions/63383400/error-cannot-uninstall-ruamel-yaml-while-creating-docker-image-for-azure-ml-a
-    - pip install mosaicml==0.10.1 --ignore-installed
+    - pip install mosaicml==0.12.1 --ignore-installed
     - ./ci/ci.sh build
 
 - label: ":octopus: Tune multinode tests"

ci/env/install-dependencies.sh

@@ -367,7 +367,6 @@ install_pip_packages() {
 
   # Additional Train test dependencies.
   if [ "${TRAIN_TESTING-}" = 1 ] || [ "${DOC_TESTING-}" = 1 ]; then
-    rm -rf "${SITE_PACKAGES}"/ruamel* # https://stackoverflow.com/questions/63383400/error-cannot-uninstall-ruamel-yaml-while-creating-docker-image-for-azure-ml-a
     pip install -U -c "${WORKSPACE_DIR}"/python/requirements.txt -r "${WORKSPACE_DIR}"/python/requirements/ml/requirements_train.txt
   fi
 

python/ray/train/tests/test_mosaic_trainer.py

@@ -258,7 +258,7 @@ def test_monitor_callbacks(ray_start_4_cpus):
     from ray.train.mosaic import MosaicTrainer
 
     # Test Callbacks involving logging (SpeedMonitor, LRMonitor)
-    from composer.callbacks import SpeedMonitor, LRMonitor, GradMonitor
+    from composer.callbacks import SpeedMonitor, LRMonitor
 
     trainer_init_config = {
         "max_duration": "1ep",
@@ -270,7 +270,6 @@ def test_monitor_callbacks(ray_start_4_cpus):
     trainer_init_config["callbacks"] = [
         SpeedMonitor(window_size=3),
         LRMonitor(),
-        GradMonitor(),
     ]
 
     trainer = MosaicTrainer(
@@ -289,7 +288,6 @@ def test_monitor_callbacks(ray_start_4_cpus):
         "wall_clock/val",
         "wall_clock/total",
         "lr-DecoupledSGDW/group0",
-        "grad_l2_norm/step",
     ]
     for column in columns_to_check:
         assert column in metrics_columns, column + " is not found"

python/requirements/ml/requirements_train.txt

@@ -2,7 +2,7 @@
 
 -r requirements_dl.txt
 
-mosaicml==0.10.1
+mosaicml==0.12.1
 mlflow==1.30.0
 tensorboardX==2.4.1