Enhance multi-team workshop with dual GPU flavors and add kueue architecture diagram

Author: abhijeet-dhumal

examples/kfto-sft-llm/README.md

@@ -58,7 +58,7 @@ It uses HuggingFace SFTTrainer, with PEFT for LoRA and qLoRA, and PyTorch FSDP t
 >     * Follow the [Configure Kueue (Optional)](#configure-kueue-optional) section to set up required resources
 >     * Add the local-queue name label to your job configuration to enforce workload management
 >   * You can skip Kueue usage by:
->     > Note: Kueue Enablement via Validating Admission Policy was introduced in RHOAI-2.21. You can skip this section if using an earlier RHOAI release version.
+>     > Note: Kueue Enablement via Validating Admission Policy was introduced in RHOAI 2.21. You can skip this section if using an earlier RHOAI release version.
 >     * Disabling the existing `kueue-validating-admission-policy-binding`
 >     * Omitting the local-queue-name label in your job configuration
 

workshops/kueue/README.md

@@ -0,0 +1,23 @@
+apiVersion: kueue.x-k8s.io/v1beta1
+kind: ResourceFlavor
+metadata:
+  name: nvidia-a100-80gb
+spec:
+  nodeLabels:
+    nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
+  tolerations:
+  - key: nvidia.com/gpu
+    operator: Exists
+    effect: NoSchedule
+---
+apiVersion: kueue.x-k8s.io/v1beta1
+kind: ResourceFlavor
+metadata:
+  name: nvidia-h100-80gb
+spec:
+  nodeLabels:
+    nvidia.com/gpu.product: NVIDIA-H100-80GB-HBM3
+  tolerations:
+  - key: nvidia.com/gpu
+    operator: Exists
+    effect: NoSchedule

workshops/kueue/resources/resource_flavor.yaml

@@ -17,12 +17,20 @@ spec:
         - memory
         - nvidia.com/gpu
       flavors:
-        - name: nvidia-a100
+        - name: nvidia-h100-80gb
           resources:
             - name: cpu
-              nominalQuota: '256'
+              nominalQuota: '16'
             - name: memory
-              nominalQuota: 2000Gi
+              nominalQuota: 256Gi
             - name: nvidia.com/gpu
-              nominalQuota: '5'
+              nominalQuota: '2'
+        - name: nvidia-a100-80gb
+          resources:
+            - name: cpu
+              nominalQuota: '64'
+            - name: memory
+              nominalQuota: 1024Gi
+            - name: nvidia.com/gpu
+              nominalQuota: '8'
   stopPolicy: None

workshops/kueue/resources/resource_flavors.yaml

@@ -1,6 +1,6 @@
 apiVersion: kueue.x-k8s.io/v1beta1
 kind: LocalQueue
 metadata:
-  name: team1-alpha
+  name: team1
 spec:
   clusterQueue: team1 

workshops/kueue/resources/team1_cluster_queue.yaml

@@ -17,12 +17,20 @@ spec:
         - memory
         - nvidia.com/gpu
       flavors:
-        - name: nvidia-a100
+        - name: nvidia-h100-80gb
           resources:
             - name: cpu
-              nominalQuota: '256'
+              nominalQuota: '48'
             - name: memory
-              nominalQuota: 2000Gi
+              nominalQuota: 768Gi
             - name: nvidia.com/gpu
-              nominalQuota: '5'
+              nominalQuota: '6'
+        - name: nvidia-a100-80gb
+          resources:
+            - name: cpu
+              nominalQuota: '32'
+            - name: memory
+              nominalQuota: 512Gi
+            - name: nvidia.com/gpu
+              nominalQuota: '4'
   stopPolicy: None

workshops/kueue/resources/team1_local_queue.yaml

@@ -1,6 +1,6 @@
 apiVersion: kueue.x-k8s.io/v1beta1
 kind: LocalQueue
 metadata:
-  name: team2-alpha
+  name: team2
 spec:
   clusterQueue: team2