Mcli config for v0-step_1T-warmup_true

configs/mcli/annealing/v0-step_1T-warmup_true.yaml

@@ -0,0 +1,36 @@
+name: v0-step_1T-warmup_true  # can't have "_" or "." here
+image: mosaicml/pytorch:2.1.2_cu121-python3.10-ubuntu20.04
+compute:
+  gpus: 32
+  cluster: r15z1p1
+  gpu_type: h100_80gb
+integrations:
+  - integration_type: git_repo
+    git_repo: allenai/OLMo
+    git_branch: dave/annealing
+    #git_commit: d765e8819f5b0be204c96b0b519de2372b0da729
+    pip_install: -e .[train]
+    ssh_clone: true
+command: |-
+  pip freeze
+  mkdir -p /root/.cache/torch/
+
+  export OMP_NUM_THREADS=8
+  export LOG_FILTER_TYPE=all_ranks
+  #export OLMO_NO_SSL=1
+
+  # warm up huggingface cache
+  pushd /root/.cache
+  curl "https://storage.googleapis.com/dirkgr-public/huggingface_cache_v3.tar.gz" | tar -xzf -
+  popd
+  export HF_DATASETS_OFFLINE=1
+
+  cd OLMo
+  
+  torchrun \
+  --master_addr $MASTER_ADDR \
+  --master_port $MASTER_PORT \
+  --nnodes $NUM_NODES \
+  --node_rank $NODE_RANK \
+  --nproc_per_node 8 \
+    scripts/train.py configs/annealing/v0-step_1T-warmup_true.yaml