Merge branch 'main' into packing-resumption

mosaicml · Jun 13, 2024 · 1063a5d · 1063a5d
2 parents 2098118 + c30856f
commit 1063a5d
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 1 deletion.
diff --git a/.github/workflows/docker.yaml b/.github/workflows/docker.yaml
@@ -23,6 +23,12 @@ jobs:
         - name: "2.3.0_cu121_flash2_aws"
           base_image: mosaicml/pytorch:2.3.0_cu121-python3.11-ubuntu20.04-aws
           dep_groups: "[gpu-flash2]"
+        - name: "2.3.1_cu121"
+          base_image: mosaicml/pytorch:2.3.1_cu121-python3.11-ubuntu20.04
+          dep_groups: "[gpu]"
+        - name: "2.3.1_cu121_aws"
+          base_image: mosaicml/pytorch:2.3.1_cu121-python3.11-ubuntu20.04-aws
+          dep_groups: "[gpu]"
     steps:
     - name: Maximize Build Space on Worker
       uses: easimon/maximize-build-space@v4

diff --git a/llmfoundry/utils/config_utils.py b/llmfoundry/utils/config_utils.py
@@ -115,7 +115,7 @@ class TrainConfig:
 
     # Cuda allocation configuration
     max_split_size_mb: Optional[int] = None
-    expandable_segments: bool = False
+    expandable_segments: bool = True
     cuda_load_lazy: bool = False
 
     # Distributed training parameters