[Rust Frontend][gRPC] Separate liveness from inference readiness

.agents/skills/model-microbenchmark/SKILL.md

@@ -7,8 +7,9 @@ description: Time a vLLM model's forward pass on exact batch shapes (decode, pre
 
 Builds the batch without the scheduler, runs it through MRV2's real
 `execute_model`, and times the forward as a one-shot CUDA graph (prefill
-included). Sampling, the speculator, and CPU step overhead are excluded, so
-expect results somewhat below a serving step.
+included). The configured drafter is included by default; use
+`--exclude-drafter` for target-only timing. All draft tokens are accepted.
+Target sampling, rejection sampling, and CPU step overhead are excluded.
 
 ## Usage
 
@@ -44,7 +45,8 @@ Key flags (plus all `EngineArgs`):
 
 - `--kv-len-jitter 0.5` (default): prefixes drawn within ±50%, same on every
   DP rank, capped at `max_model_len`. Use `0` for exact lengths.
-- `--routing auto|real|balanced`: `auto` = balanced with dummy weights.
+- `--routing auto|real|balanced|zipf`: `auto` = Zipf with dummy weights;
+  `zipf` uses fixed alpha=1.25 skew (target only, no PCP or grouped top-k).
 - `--iters 20 --samples 5`, `--output-json out.json` (per-rank latency stats,
   `samples_us`, graph-vs-eager check).
 - Defaults: `--load-format dummy`, prefix caching off, no torch.compile.
@@ -94,7 +96,7 @@ that the JSON summary excludes.
 ## Caveats
 
 - **Dummy weights lie for data-dependent kernels.** Dummy gates route every
-  token to the same experts (hence balanced routing). Cross-check suspicious
+  token to the same experts (hence synthetic routing). Cross-check suspicious
   kernels with real weights.
 - **torch.compile models (e.g. Qwen3/3.5) look slow**: no compile here. Use
   them for correctness only; trust perf for models under `vllm/models/`.

.agents/skills/model-recipes/SKILL.md

@@ -7,9 +7,9 @@ description: Find vLLM deployment recipes and configs for LLMs and Hugging Face
 
 Model serving recipes are maintained in this repository's
 [recipes/models/ directory](../../../recipes/models). Omni (image, video,
-audio) recipes stay in vllm-project/recipes for now (moving to vllm-omni),
-and Ascend NPU content is inline in these recipes (moving to vllm-ascend
-overlays). The published catalog, rendered by
+audio) recipes live in vllm-project/vllm-omni, and Ascend NPU content lives in
+vllm-project/vllm-ascend as overlays merged into these recipes. The published
+catalog, rendered by
 [vllm-project/recipes](https://github.com/vllm-project/recipes), is at
 [recipes.vllm.ai](https://recipes.vllm.ai).
 

.buildkite/amd-disagg/cluster.sh

@@ -156,6 +156,21 @@ export MORI_GPU_ARCHS="gfx950"
 # path fails to come up.
 export MORI_ENABLE_HOST_PROXY="${MORI_ENABLE_HOST_PROXY:-1}"
 
+# NIXL UCX defaults: RC on ionic_0, GPU-direct RDMA.
+if [[ "${KV_CONNECTOR:-}" == Nixl* ]]; then
+    export UCX_TLS="${UCX_TLS:-rc_verbs,ud_verbs,tcp,self,rocm}"
+    export UCX_NET_DEVICES="${UCX_NET_DEVICES:-ionic_0:1,lo}"
+    export UCX_IB_GID_INDEX="${UCX_IB_GID_INDEX:-${_IB_GID_INDEX}}"
+    export UCX_UD_VERBS_ETH_DGID_CHECK="${UCX_UD_VERBS_ETH_DGID_CHECK:-n}"
+    export UCX_IB_GPU_DIRECT_RDMA="${UCX_IB_GPU_DIRECT_RDMA:-yes}"
+    export UCX_RMA_PPLN_ENABLE="${UCX_RMA_PPLN_ENABLE:-y}"
+    export UCX_PROTO_EMULATION_ENABLE="${UCX_PROTO_EMULATION_ENABLE:-n}"
+    export UCX_KEEPALIVE_INTERVAL="${UCX_KEEPALIVE_INTERVAL:-inf}"
+    export UCX_UD_VERBS_RX_QUEUE_LEN="${UCX_UD_VERBS_RX_QUEUE_LEN:-1024}"
+    export UCX_RCACHE_MAX_UNRELEASED="${UCX_RCACHE_MAX_UNRELEASED:-1024}"
+    export ACCURACY_NUM_CONCURRENT="${ACCURACY_NUM_CONCURRENT:-1}"
+fi
+
 # ----------------------------------------------------------------- benchmark
 export BENCHMARK_COMBINATIONS="${BENCHMARK_COMBINATIONS:-1024/128 2048/128}"
 export BENCHMARK_CON="${BENCHMARK_CON:-32 64}"

.buildkite/amd-disagg/models.yaml

@@ -105,6 +105,20 @@ models:
       tp: "--gpu-memory-utilization 0.9"
       ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"PIECEWISE","custom_ops":["+quant_fp8"]}'
 
+  - model: DeepSeek-V4-Pro
+    env:
+      VLLM_ROCM_USE_AITER: "1"
+      VLLM_ROCM_USE_AITER_MLA: "1"
+      VLLM_ROCM_USE_AITER_MOE: "1"
+      VLLM_ROCM_USE_AITER_RMSNORM: "1"
+    base_flags: "--trust-remote-code --kv-cache-dtype fp8"
+    prefill:
+      tp: "--gpu-memory-utilization 0.85"
+      ep: "--gpu-memory-utilization 0.85 --enforce-eager"
+    decode:
+      tp: "--gpu-memory-utilization 0.85"
+      ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"PIECEWISE","custom_ops":["+quant_fp8"]}'
+
   - model: GLM-5.2-FP8
     env:
       VLLM_ROCM_USE_AITER: "1"

.buildkite/amd-disagg/pipeline-disagg.yaml

@@ -102,6 +102,15 @@ steps:
   commands:
   - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
 
+- label: "DeepSeek-V4-Pro-PD-1P1D-TP8-MoRIIO-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V4-Pro NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
 - label: "GLM-5.2-FP8-PD-1P1D-TP8-MoRIIO-proxy"
   agents:
     queue: amd_mi350_ainic
@@ -310,6 +319,15 @@ steps:
   commands:
   - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
 
+- label: "DeepSeek-V4-Pro-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V4-Pro NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
 - label: "GLM-5.2-FP8-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
   agents:
     queue: amd_mi350_ainic
@@ -378,6 +396,15 @@ steps:
   commands:
   - MODEL_NAME=MiniMax-M3-MXFP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy WORKLOAD_TIMEOUT_S=10800 ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
 
+- label: "DeepSeek-V4-Pro-PD-2P2D-EP16/DP16-WideEP-MoRIIO-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 1
+  concurrency_group: "amd-disagg/mi350/4node"
+  commands:
+  - MODEL_NAME=DeepSeek-V4-Pro xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
 - label: "GLM-5.2-FP8-PD-2P2D-EP16/DP16-WideEP-MoRIIO-vllm-router"
   agents:
     queue: amd_mi350_ainic
@@ -387,3 +414,58 @@ steps:
   commands:
   - MODEL_NAME=GLM-5.2-FP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
 
+#--------------------------------------  mi350 · disagg NIXL (UCX RDMA) · SLURM  --------------------------------------#
+
+- label: "DeepSeek-V3-PD-1P1D-TP8-NIXL-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 KV_CONNECTOR=NixlConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
+- label: "DeepSeek-V3-PD-2P2D-EP16/DP16-WideEP-NIXL-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 1
+  concurrency_group: "amd-disagg/mi350/4node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=1 KV_CONNECTOR=NixlConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
+- label: "DeepSeek-V3-PD-1P1D-EP8/DP8-WideEP-NIXL-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 KV_CONNECTOR=NixlConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
+- label: "DeepSeek-V3-PD-1P1D-TP8-NIXL-PUSH-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 KV_CONNECTOR=NixlPushConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
+- label: "DeepSeek-V3-PD-1P1D-EP8/DP8-WideEP-NIXL-PUSH-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 2
+  concurrency_group: "amd-disagg/mi350/2node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 KV_CONNECTOR=NixlPushConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
+
+- label: "DeepSeek-V3-PD-2P2D-EP16/DP16-WideEP-NIXL-PUSH-vllm-router"
+  agents:
+    queue: amd_mi350_ainic
+  timeout_in_minutes: 270
+  concurrency: 1
+  concurrency_group: "amd-disagg/mi350/4node"
+  commands:
+  - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=1 KV_CONNECTOR=NixlPushConnector RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh

.buildkite/amd-disagg/run-slurm-disagg-test.sh

@@ -47,6 +47,7 @@ SHARED_MOUNT="${SHARED_MOUNT:-/data}"
 LOG_ROOT="${LOG_ROOT:-${SHARED_MOUNT}/${USER:-$(whoami)}/disagg_logs}"
 DRY_RUN="${DRY_RUN:-0}"
 MORIIO_READ_MODE="${MORIIO_READ_MODE:-0}"
+KV_CONNECTOR="${KV_CONNECTOR:-MoRIIOConnector}"
 
 WAIT="${WAIT:-0}"
 
@@ -69,7 +70,7 @@ export DISAGG_SCRIPTS_DIR="${STAGED_DIR}"
 echo "[slurm-submit] staged scripts for compute nodes: ${DISAGG_SCRIPTS_DIR}" >&2
 
 export IMAGE MODEL_NAME WIDE_EP_MODE xP yD GPUS_PER_NODE RUN_AFTER_HEALTH HEALTH_TIMEOUT_S
-export SHARED_MOUNT LOG_ROOT DRY_RUN MORIIO_READ_MODE
+export SHARED_MOUNT LOG_ROOT DRY_RUN MORIIO_READ_MODE KV_CONNECTOR
 export ROUTER_TYPE ROUTER_PORT VLLM_ROUTER_IMAGE
 # Leave the image's own ionic userspace in place instead of mounting the host's.
 export SKIP_IONIC_MOUNTS

.buildkite/amd-disagg/run_xPyD_disagg.slurm

@@ -23,6 +23,7 @@ LOG_ROOT="${LOG_ROOT:-/data/${USER:-$(id -un)}/disagg_logs}"
 RUN_AFTER_HEALTH="${RUN_AFTER_HEALTH:-accuracy}"   # bench | accuracy
 DRY_RUN="${DRY_RUN:-0}"
 MORIIO_READ_MODE="${MORIIO_READ_MODE:-0}"   # 0 -> prefill writes (default); 1 -> decode reads KV (read_mode)
+KV_CONNECTOR="${KV_CONNECTOR:-MoRIIOConnector}"   # MoRIIOConnector | NixlConnector | NixlPushConnector
 
 # ---- router / gateway (host-side; also forwarded into the container) --------
 # vllm-router -> external `vllm/vllm-router` container started on the rank-0 node (default)
@@ -232,7 +233,7 @@ echo "[disagg-pd] LOG_PATH=$LOG_PATH"
 #   rank xP .. end   -> decode
 export IPADDRS
 
-export WIDE_EP_MODE xP yD GPUS_PER_NODE RUN_AFTER_HEALTH HEALTH_TIMEOUT_S DRY_RUN MORIIO_READ_MODE
+export WIDE_EP_MODE xP yD GPUS_PER_NODE RUN_AFTER_HEALTH HEALTH_TIMEOUT_S DRY_RUN MORIIO_READ_MODE KV_CONNECTOR
 export IMAGE SHARED_MOUNT CONTAINER_SCRIPTS DISAGG_SCRIPTS_DIR
 export CLUSTER_ENV="${CLUSTER_ENV_IN}"
 export DOCKER_CONT_NAME="vllm_disagg_${MODEL_NAME:-model}_${SLURM_JOB_ID}"
@@ -247,6 +248,21 @@ export PREFILL_PORT DECODE_PORT PROXY_PORT ROUTER_METRICS_PORT
 export SERVE_PORT RPC_PORT KV_PORT LOCAL_PING_PORT
 export PREFLIGHT PREFLIGHT_MAX_VRAM_PCT PREFLIGHT_GPU_SETTLE_S
 
+# NIXL has no discovery service, so pass the worker URLs.
+ROUTER_KV_ARGS="--kv-connector moriio --vllm-discovery-address 0.0.0.0:${PROXY_PING_PORT}"
+if [[ "${KV_CONNECTOR}" == Nixl* ]]; then
+    nixl_startup_timeout=1800
+    [[ "${WIDE_EP_MODE}" == "1" && "${xP}" -gt 1 ]] && nixl_startup_timeout=7200
+    ROUTER_KV_ARGS="--kv-connector nixl --worker-startup-timeout-secs ${nixl_startup_timeout}"
+    if [[ "${WIDE_EP_MODE}" == "1" ]]; then
+        ROUTER_KV_ARGS+=" --prefill http://${IPS[0]}:${SERVE_PORT} --decode http://${IPS[$xP]}:${SERVE_PORT}"
+    else
+        for ((i = 0; i < xP; i++)); do ROUTER_KV_ARGS+=" --prefill http://${IPS[$i]}:${PREFILL_PORT}"; done
+        for ((i = xP; i < xP + yD; i++)); do ROUTER_KV_ARGS+=" --decode http://${IPS[$i]}:${DECODE_PORT}"; done
+    fi
+fi
+export ROUTER_KV_ARGS
+
 echo "[disagg-pd] IPADDRS=${IPADDRS}"
 echo "[disagg-pd] launching ${NUM_NODES} containers (1/node) -> vllm_disagg.sh node"
 
@@ -479,7 +495,7 @@ run_node_container() {
                 -e USER="$(id -un)" \
                 -e LOGNAME="$(id -un)" \
                 "$VLLM_ROUTER_IMAGE" \
-                bash -lc "exec vllm-router --vllm-pd-disaggregation --kv-connector moriio --vllm-discovery-address 0.0.0.0:$PROXY_PING_PORT --port $ROUTER_PORT --host 0.0.0.0 --intra-node-data-parallel-size $ROUTER_DP_LOCAL --policy $ROUTER_POLICY --prefill-policy $ROUTER_POLICY --decode-policy $ROUTER_POLICY --log-level info >> $LOG_PATH/vllm_router.log 2>&1" \
+                bash -lc "exec vllm-router --vllm-pd-disaggregation $ROUTER_KV_ARGS --port $ROUTER_PORT --host 0.0.0.0 --intra-node-data-parallel-size $ROUTER_DP_LOCAL --policy $ROUTER_POLICY --prefill-policy $ROUTER_POLICY --decode-policy $ROUTER_POLICY --log-level info >> $LOG_PATH/vllm_router.log 2>&1" \
               && echo "vllm-router container started ok" \
               || echo "WARN: failed to start vllm-router container on $(hostname)"
         } 2>&1 | tee -a "$LOG_PATH/vllm_router_start.log"
@@ -528,6 +544,7 @@ run_node_container() {
         -e HEALTH_TIMEOUT_S="$HEALTH_TIMEOUT_S" \
         -e DRY_RUN="$DRY_RUN" \
         -e MORIIO_READ_MODE="$MORIIO_READ_MODE" \
+        -e KV_CONNECTOR="$KV_CONNECTOR" \
         -e ROUTER_TYPE="$ROUTER_TYPE" \
         -e ROUTER_PORT="$ROUTER_PORT" \
         -e ROUTER_POLICY="$ROUTER_POLICY" \

.buildkite/amd-disagg/vllm_disagg.sh

@@ -102,6 +102,18 @@ moriio_read_mode_kv() {
     esac
 }
 
+# NIXL KV config. UCX backend.
+nixl_kv_cfg() {
+    cat <<EOF
+{
+  "kv_connector": "${KV_CONNECTOR}",
+  "kv_role": "${KV_ROLE}",
+  "kv_load_failure_policy": "recompute",
+  "kv_connector_extra_config": {"backends": ["UCX"], "kv_lease_duration": 180}
+}
+EOF
+}
+
 # ----------------------------------------------------------------- topology
 # Resolves MODEL_PATH, MODELS_YAML, the IP array, and the master/proxy addresses.
 # IP_ARRAY[0..xP-1] = prefill nodes, [xP..xP+yD-1] = decode nodes.
@@ -309,6 +321,7 @@ build_tp_cmd() {
 }
 EOF
 )
+    if [[ "${KV_CONNECTOR:-}" == Nixl* ]]; then kv_cfg="$(nixl_kv_cfg)"; fi
     CMD+=(--kv-transfer-config "${kv_cfg}")
     log "role=${ROLE} mode=tp host=${HOST_IP:-0.0.0.0}:${port} kv_role=${KV_ROLE} tp=${TP_SIZE} read_mode=${MORIIO_READ_MODE}"
 }
@@ -337,6 +350,10 @@ build_ep_cmd() {
         --distributed-timeout-seconds "${DISTRIBUTED_TIMEOUT_SECONDS}"
     )
     local _mc; read -ra _mc <<< "${MODEL_CONFIG}"; CMD+=("${_mc[@]}")
+    # 2P2D EP16 only. At 0.85 the AITER MoE sort buffer runs out of memory.
+    if [[ "${KV_CONNECTOR:-}" == Nixl* && "${ROLE}" == "prefill" && "${xP}" -gt 1 ]]; then
+        CMD+=(--gpu-memory-utilization 0.75)
+    fi
 
     if [[ "${IS_MASTER}" == "1" ]]; then
         CMD+=(--api-server-count="${GPUS_PER_NODE}")
@@ -359,6 +376,7 @@ build_ep_cmd() {
 }
 EOF
 )
+        if [[ "${KV_CONNECTOR:-}" == Nixl* ]]; then kv_cfg="$(nixl_kv_cfg)"; fi
         CMD+=(--kv-transfer-config "${kv_cfg}")
         log "role=${ROLE} mode=ep MASTER node_rank=${NODE_RANK} dp_size=${DP_GROUP_SIZE} dp_addr=${DP_MASTER_ADDR} kv_role=${KV_ROLE} all2all=${backend} read_mode=${MORIIO_READ_MODE}"
     else

.buildkite/ci_config_rocm.yaml

@@ -5,18 +5,17 @@ run_all_patterns:
   - ".dockerignore"
   - "docker/Dockerfile.rocm"
   - "docker/Dockerfile.rocm_base"
-  - "docker/ci-rocm.hcl"
   - "docker/docker-bake-rocm.hcl"
   - ".buildkite/hardware_tests/amd.yaml"
   - ".buildkite/ci_config_rocm.yaml"
-  - ".buildkite/scripts/ci-bake-rocm.sh"
-  - ".buildkite/scripts/rocm/"
+  - ".buildkite/scripts/rocm/build.sh"
+  - ".buildkite/scripts/rocm/select_aiter_nightly_wheel.py"
   - ".buildkite/scripts/hardware_ci/run-amd-test.sh"
+  - ".buildkite/scripts/hardware_ci/amd_ci_fast_exit.py"
   - "CMakeLists.txt"
   - "requirements/common.txt"
   - "requirements/rocm.txt"
   - "requirements/build/rust.txt"
-  - "requirements/build/rocm.txt"
   - "requirements/test/rocm.txt"
   - "pyproject.toml"
   - "setup.py"
@@ -25,6 +24,8 @@ run_all_patterns:
   - "rust-toolchain.toml"
   - "tools/build_rust.py"
   - "tools/install_torchcodec_rocm.sh"
+  - "tools/vllm-rocm/therock_wheels.py"
+  - "tools/ci/rocm_kernel_symbol_map.py"
   - "tests/vllm_test_utils/"
   - "vllm/envs.py"
   - "vllm/__init__.py"