diff --git a/.github/workflows/ci-pr-benchmark.yaml b/.github/workflows/ci-pr-benchmark.yaml
index 0899f600..8f919c36 100644
--- a/.github/workflows/ci-pr-benchmark.yaml
+++ b/.github/workflows/ci-pr-benchmark.yaml
@@ -37,7 +37,7 @@ jobs:
 
       - name: Standup a modelservice using llm-d-inference-sim
         env:
-          LLMDBENCH_HF_TOKEN: hf-token-placeholder
+          LLMDBENCH_HF_TOKEN: ${{ secrets.LLMDBENCH_HF_TOKEN }}
         run: |
           ./setup/standup.sh -c kind_modelservice_inference-sim -t modelservice -s 0,1,2,4,7,8,9
 
diff --git a/scenarios/kind_modelservice_inference-sim.sh b/scenarios/kind_modelservice_inference-sim.sh
index 7147f560..be2663d3 100644
--- a/scenarios/kind_modelservice_inference-sim.sh
+++ b/scenarios/kind_modelservice_inference-sim.sh
@@ -14,8 +14,9 @@ export LLMDBENCH_VLLM_MODELSERVICE_DECODE_CPU_NR=0
 export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_CPU_NR=0
 export LLMDBENCH_VLLM_MODELSERVICE_DECODE_CPU_MEM=100Mi
 export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_CPU_MEM=100Mi
-export LLMDBENCH_VLLM_MODELSERVICE_URI_PROTOCOL="hf"
+# export LLMDBENCH_VLLM_MODELSERVICE_URI_PROTOCOL="hf"
 export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 export LLMDBENCH_HARNESS_PVC_SIZE=3Gi
 export LLMDBENCH_VLLM_MODELSERVICE_INFERENCE_MODEL=true
-
+export LLMDBENCH_VLLM_COMMON_PVC_ACCESSMODE=ReadWriteOnce
+export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=1Gi
diff --git a/setup/env.sh b/setup/env.sh
index 320a6b16..360889c5 100644
--- a/setup/env.sh
+++ b/setup/env.sh
@@ -65,6 +65,7 @@ export LLMDBENCH_VLLM_COMMON_BLOCK_SIZE=${LLMDBENCH_VLLM_COMMON_BLOCK_SIZE:-64}
 export LLMDBENCH_VLLM_COMMON_MAX_NUM_BATCHED_TOKENS=${LLMDBENCH_VLLM_COMMON_MAX_NUM_BATCHED_TOKENS:-4096}
 export LLMDBENCH_VLLM_COMMON_PVC_NAME=${LLMDBENCH_VLLM_COMMON_PVC_NAME:-"model-pvc"}
 export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS="${LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS:-default}"
+export LLMDBENCH_VLLM_COMMON_PVC_ACCESSMODE=${LLMDBENCH_VLLM_COMMON_PVC_ACCESSMODE:-ReadWriteMany}
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE="${LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE:-300Gi}"
 export LLMDBENCH_VLLM_COMMON_PVC_DOWNLOAD_TIMEOUT=${LLMDBENCH_VLLM_COMMON_PVC_DOWNLOAD_TIMEOUT:-"2400"}
 export LLMDBENCH_VLLM_COMMON_HF_TOKEN_KEY="${LLMDBENCH_VLLM_COMMON_HF_TOKEN_KEY:-"HF_TOKEN"}"
diff --git a/setup/functions.py b/setup/functions.py
index 44f3f967..a5d50f58 100644
--- a/setup/functions.py
+++ b/setup/functions.py
@@ -251,6 +251,7 @@ def validate_and_create_pvc(
     pvc_name: str,
     pvc_size: str,
     pvc_class: str,
+    pvc_access_mode: str = 'ReadWriteMany',
     dry_run: bool = False
 ):
     announce("Provisioning model storage…")
@@ -294,7 +295,7 @@ def validate_and_create_pvc(
             "namespace": namespace,
         },
         "spec": {
-            "accessModes": ["ReadWriteMany"],
+            "accessModes": [pvc_access_mode],
             "resources": {
                 "requests": {"storage": pvc_size}
             },
diff --git a/setup/steps/04_ensure_model_namespace_prepared.py b/setup/steps/04_ensure_model_namespace_prepared.py
index ee1055f5..b33dbf3e 100644
--- a/setup/steps/04_ensure_model_namespace_prepared.py
+++ b/setup/steps/04_ensure_model_namespace_prepared.py
@@ -73,12 +73,6 @@ def main():
     ev = {}
     environment_variable_to_dict(ev)
 
-    env_cmd=f'source "{ev["control_dir"]}/env.sh"'
-    result = llmdbench_execute_cmd(actual_cmd=env_cmd, dry_run=ev["control_dry_run"], verbose=ev["control_verbose"])
-    if result != 0:
-        announce(f"❌ Failed while running \"{env_cmd}\" (exit code: {result})")
-        exit(result)
-
     api = kube_connect(f'{ev["control_work_dir"]}/environment/context.ctx')
     if ev["control_dry_run"] :
         announce("DRY RUN enabled. No actual changes will be made.")
@@ -110,6 +104,10 @@ def main():
             protocol, pvc_and_model_path = model_artifact_uri.split("://") # protocol var unused but exists in prev script
             pvc_name, model_path = pvc_and_model_path.split('/', 1) # split from first occurence
 
+            announce(f'pvc_name = {ev["vllm_common_pvc_name"]}')
+            announce(f'pvc_size = {ev["vllm_common_pvc_model_cache_size"]}')
+            announce(f'pvc_class = {ev["vllm_common_pvc_storage_class"]}')
+            announce(f'pvc_access_mode = {ev["vllm_common_pvc_accessmode"]}')
             validate_and_create_pvc(
                 api=api,
                 namespace=ev["vllm_common_namespace"],
@@ -117,6 +115,7 @@ def main():
                 pvc_name=ev["vllm_common_pvc_name"],
                 pvc_size=ev["vllm_common_pvc_model_cache_size"],
                 pvc_class=ev["vllm_common_pvc_storage_class"],
+                pvc_access_mode=ev["vllm_common_pvc_accessmode"],
                 dry_run=ev["control_dry_run"]
             )