add mithril h100 pool (#187)

Signed-off-by: simon-mo <[email protected]>

Author: simon-mo

buildkite/test-template-ci.j2

@@ -102,6 +102,8 @@ agents:
   queue: cpu_queue_premerge_us_east_1
   {% elif step.gpu == "a100" %}
   queue: a100_queue
+  {% elif step.gpu == "h100" %}
+  queue: mithril-h100-pool
   {% elif step.gpu == "h200" %}
   queue: skylab-h200
   {% elif step.gpu == "b200" %}
@@ -130,7 +132,7 @@ retry:
 
 {% if step.num_nodes < 2 %}
 plugins:
-  {% if step.gpu != "a100" and step.gpu != "h200" and step.gpu != "b200" %}
+  {% if step.gpu != "a100" and step.gpu != "h100" and step.gpu != "h200" and step.gpu != "b200" %}
   - docker#v5.2.0:
       image: {{ image }}
       always-pull: true
@@ -207,6 +209,38 @@ plugins:
         - /dev/shm:/dev/shm
         - /data/benchmark-hf-cache:/benchmark-hf-cache
         - /data/benchmark-vllm-cache:/root/.cache/vllm
+  {% elif step.gpu == "h100" %}
+  - kubernetes:
+      podSpec:
+        containers:
+          - image: {{ image }}
+            command:
+              - bash -c "{{ '(command nvidia-smi || true) && export VLLM_ALLOW_DEPRECATED_BEAM_SEARCH=1 && cd ' ~ ((step.working_dir or default_working_dir) | safe) ~ ' && ' ~ (step.command or (step.commands | join(" && ")) | safe) }}"
+            resources:
+              limits:
+                nvidia.com/gpu: {{ step.num_gpus or 1 }}
+            volumeMounts:
+              - name: devshm
+                mountPath: /dev/shm
+              - name: hf-cache
+                mountPath: {{ hf_home }}
+            env:
+              - name: VLLM_USAGE_SOURCE
+                value: ci-test
+              - name: NCCL_CUMEM_HOST_ENABLE
+                value: "0"
+              - name: HF_HOME
+                value: {{ hf_home }}
+        nodeSelector:
+          nvidia.com/gpu.product: NVIDIA-H100-80GB-HBM3
+        volumes:
+          - name: devshm
+            emptyDir:
+              medium: Memory
+          - name: hf-cache
+            hostPath:
+              path: /mnt/hf-cache
+              type: Directory
   {% else %}
   - kubernetes:
       podSpec: