fix: add prefill load balance method args for deepseek-r1 (#4051)

Co-authored-by: dagil-nvidia <[email protected]>

Author: biswapanda

recipes/deepseek-r1/sglang/disagg-16gpu/deploy.yaml

@@ -48,7 +48,7 @@ spec:
             timeoutSeconds: 10
             failureThreshold: 600
           image: my-registry/sglang-wideep-runtime:my-tag
-          workingDir: /workspace/examples/backends/sglang
+          workingDir: /sgl-workspace/dynamo
           command:
             - python3
             - -m
@@ -77,6 +77,7 @@ spec:
             - "0.75"
             - --host
             - 0.0.0.0
+            - --prefill-round-robin-balance
     prefill:
       dynamoNamespace: sgl-dsr1-16gpu
       componentType: worker
@@ -101,7 +102,7 @@ spec:
             timeoutSeconds: 10
             failureThreshold: 600
           image: my-registry/sglang-wideep-runtime:my-tag
-          workingDir: /workspace/examples/backends/sglang
+          workingDir: /sgl-workspace/dynamo
           command:
             - python3
             - -m
@@ -126,4 +127,6 @@ spec:
             - --mem-fraction-static
             - "0.75"
             - --host
-            - 0.0.0.0
+            - 0.0.0.0
+            - --load-balance-method
+            - round_robin

recipes/deepseek-r1/sglang/disagg-8gpu/deploy.yaml

@@ -46,7 +46,7 @@ spec:
             timeoutSeconds: 10
             failureThreshold: 600
           image: my-registry/sglang-wideep-runtime:my-tag
-          workingDir: /workspace/examples/backends/sglang
+          workingDir: /sgl-workspace/dynamo
           command:
             - python3
             - -m
@@ -73,6 +73,7 @@ spec:
             - "30001"
             - --host
             - 0.0.0.0
+            - --prefill-round-robin-balance
     prefill:
       dynamoNamespace: sgl-dsr1-8gpu
       componentType: worker
@@ -95,7 +96,7 @@ spec:
             timeoutSeconds: 10
             failureThreshold: 600
           image: my-registry/sglang-wideep-runtime:my-tag
-          workingDir: /workspace/examples/backends/sglang
+          workingDir: /sgl-workspace/dynamo
           command:
             - python3
             - -m
@@ -118,4 +119,6 @@ spec:
             - --disaggregation-bootstrap-port
             - "30001"
             - --host
-            - 0.0.0.0
+            - 0.0.0.0
+            - --load-balance-method
+            - round_robin