update

configs/mcli/mitchish70.yaml

@@ -1,6 +1,6 @@
 name: olmo-70b
-image: mosaicml/pytorch:2.2.1_cu121-python3.11-ubuntu20.04
-#image: public.ecr.aws/z0f8p3z5/olmo:pytorch2.2.1_cu121-python3.11-ubuntu20.04
+# image: mosaicml/pytorch:2.2.1_cu121-python3.11-ubuntu20.04
+image: public.ecr.aws/z0f8p3z5/olmo:pytorch2.2.1_cu121-python3.11-ubuntu20.04
 scheduling:
   priority: auto
   # preemptible: true  # means it can be retried
@@ -71,7 +71,6 @@ compute:
     - inst-ht0xx-r15z3-workers
     - inst-entnk-r15z3-workers
     - inst-hvw6t-r15z3-workers
-    - inst-3to96-r15z3-workers
     - inst-4ki3x-r15z3-workers
     - inst-aixwt-r15z3-workers
     - inst-pbivr-r15z3-workers
@@ -125,13 +124,15 @@ compute:
     - inst-97xv1-r15z3-workers
     - inst-vaqst-r15z3-workers
     - inst-i6mnk-r15z3-workers
+    - inst-xtbwa-r15z3-workers
     # Bad nodes:
     # - inst-zgb86-r15z3-workers
     # - inst-hdlqg-r15z3-workers
     # - inst-6jp2q-r15z3-workers
     # - inst-bw20d-r15z3-workers
     # - inst-4zdz3-r15z3-workers
     # - inst-zlnho-r15z3-workers
+    # - inst-3to96-r15z3-workers
 integrations:
   - integration_type: git_repo
     git_repo: allenai/OLMo
@@ -204,3 +205,5 @@ command: |-
 #    --device_train_microbatch_size=2 \
 #  gpus: 896
 #    --global_train_batch_size=1792 \
+#  gpus: 600  # (75 nodes)
+#    --global_train_batch_size=1800 \