LCORE-2860: parametrize model name via vllm-model-secret

Create vllm-model-secret from MODEL_NAME in pipeline.sh so both
llama-stack and lightspeed-stack pods read the model from a single
source instead of hardcoding it.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: are-ces

tests/e2e-prow/rhoai/manifests/lightspeed/llama-stack-prow.yaml

@@ -102,7 +102,10 @@ spec:
               name: vllm-api-key-secret
               key: key
         - name: INFERENCE_MODEL
-          value: "meta-llama/Llama-3.1-8B-Instruct"
+          valueFrom:
+            secretKeyRef:
+              name: vllm-model-secret
+              key: key
         - name: OPENAI_API_KEY
           valueFrom:
             secretKeyRef:

tests/e2e-prow/rhoai/pipeline.sh

@@ -64,6 +64,7 @@ create_secret() {
 create_secret hf-token-secret --from-literal=token="$HUGGING_FACE_HUB_TOKEN"
 create_secret vllm-api-key-secret --from-literal=key="$VLLM_API_KEY"
 create_secret openai-api-key-secret --from-literal=key=""
+create_secret vllm-model-secret --from-literal=key="$MODEL_NAME"
 
 # MCP token secrets for lightspeed-stack
 REPO_ROOT="$(cd "$PIPELINE_DIR/../../.." && pwd)"

tests/e2e/configuration/server-mode/lightspeed-stack-rhoai.yaml

@@ -20,7 +20,7 @@ authentication:
   module: "noop"
 inference:
   default_provider: vllm
-  default_model: ${env.INFERENCE_MODEL}
+  default_model: ${env.VLLM_MODEL}
 byok_rag:
   - rag_id: e2e-test-docs
     rag_type: inline::faiss