[TileRT] GLM-5.3 FP8 MI355X AgentX: tilert 0.1.6.post3 + vLLM 0.28/ATOM prefill / [TileRT] GLM-5.3 FP8 MI355X AgentX：tilert 0.1.6.post3 + vLLM 0.28/ATOM prefill

.github/AGENT_OPERATIONS.md

@@ -94,7 +94,7 @@ Multinode disaggregated results add `prefill_gpu_energy_j`, `decode_gpu_energy_j
 
 Every power result — valid or invalid, single-node or multinode — carries `power_metric_schema_version`. Version 2 defines each unprefixed `joules_per_*` field as whole-deployment GPU-board energy over the named denominator; role-scoped energy uses the explicit `prefill_*` / `decode_*` keys. Rows without the field predate the whole-deployment switch and their unprefixed joules are not comparable across topologies.
 
-For srt-slurm recipes, `telemetry.enabled: true` with `telemetry.dcgm_exporter` enables official energy collection. The Git submodule pointer at `inferencex-e2e/utils/srt-slurm` is the source of truth for the shared srt-slurm commit, used by both power and non-power NVIDIA lanes. TileRT is the single documented fork exception. CI derives `POWER_PRODUCER_SHA` from the launcher stamp. The aggregate-power and AgentX power tests validate telemetry and provenance. These local tests do not prove hardware power collection. Eligible recipe-gated `dynamo-sglang` dcgm-power lanes are validated.
+For srt-slurm recipes, `telemetry.enabled: true` with `telemetry.dcgm_exporter` enables official energy collection. The Git submodule pointer at `inferencex-e2e/utils/srt-slurm` is the source of truth for every srt-slurm job, including TileRT. CI derives `POWER_PRODUCER_SHA` from the launcher stamp. The aggregate-power and AgentX power tests validate telemetry and provenance. These local tests do not prove hardware power collection. Eligible recipe-gated `dynamo-sglang` dcgm-power lanes are validated.
 
 Power audit artifacts are named `power_audit_<result>` and contain `power_validation_<result>.json` for single-node runs or `power_validation_<result>_*.json` for multinode runs. They are uploaded even when validation fails.
 

inferencex-e2e/benchmarks/benchmark_lib.sh

@@ -2003,19 +2003,34 @@ get_native_max_context_length() {
     if [ -n "${MODEL_PATH:-}" ] && [ -d "${MODEL_PATH}" ]; then
         model_path="${MODEL_PATH}"
     fi
-    python3 -c "
+    python3 - "$model_path" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+fields = ['max_position_embeddings', 'max_sequence_length', 'seq_length', 'n_positions']
+try:
+    config = json.loads((Path(sys.argv[1]) / 'config.json').read_text())
+    for field in fields:
+        value = config.get(field)
+        if type(value) is int and value > 0:
+            print(value)
+            sys.exit(0)
+except (OSError, ValueError, AttributeError):
+    pass
+
 try:
     from transformers import AutoConfig
-    config = AutoConfig.from_pretrained('${model_path}', trust_remote_code=True)
-    for attr in ['max_position_embeddings', 'max_sequence_length', 'seq_length', 'n_positions']:
+    config = AutoConfig.from_pretrained(sys.argv[1], trust_remote_code=True)
+    for attr in fields:
         if hasattr(config, attr):
             print(getattr(config, attr))
             break
     else:
         print(0)
 except Exception:
     print(0)
-"
+PY
 }
 
 # Requested benchmark context capped at the model's native max. Sets

inferencex-e2e/benchmarks/multi_node/amd_utils/env.sh

@@ -2,21 +2,12 @@
 
 source "$(dirname "${BASH_SOURCE[0]}")/../../benchmark_lib.sh" --validation-only
 check_env_vars ENGINE
-# MoRI-IO queue-pair tuning, the UCX RoCE GID index, SGLang router logging and the
-# SGLang decode cuda-graph NCCL workaround. Only the SGLang MoRI KV path
-# below reads these. ENGINE=tilert moves KV over mooncake and starts no SGLang
-# router, so it is neither given nor reads them: validating them there would force
-# the recipe to invent MoRI tuning for a transport it never uses.
-if [[ "$ENGINE" != "tilert" ]]; then
-    check_env_vars \
-        MORI_IO_SQ_BACKOFF_TIMEOUT_US MORI_IO_QP_MAX_SEND_WR MORI_IO_QP_MAX_CQE MORI_IO_QP_MAX_SGE MORI_IO_TC_DISABLE \
-        UCX_IB_GID_INDEX MORI_APP_LOG_LEVEL SGLANG_ROUTER_STDOUT_LOGS TORCH_NCCL_BLOCKING_WAIT NCCL_BLOCKING_WAIT \
-        SGLANG_OPT_USE_AITER_INDEXER
-fi
-# Dual-engine environment setup for multi-node disaggregated serving.
-#
-# ENGINE=sglang-disagg or tilert selects the engine-specific block.
-#
+# SGLang MoRI environment.
+check_env_vars \
+    MORI_IO_SQ_BACKOFF_TIMEOUT_US MORI_IO_QP_MAX_SEND_WR MORI_IO_QP_MAX_CQE MORI_IO_QP_MAX_SGE MORI_IO_TC_DISABLE \
+    UCX_IB_GID_INDEX MORI_APP_LOG_LEVEL SGLANG_ROUTER_STDOUT_LOGS TORCH_NCCL_BLOCKING_WAIT NCCL_BLOCKING_WAIT \
+    SGLANG_OPT_USE_AITER_INDEXER
+
 # REQUIRED ENVIRONMENT VARIABLES:
 #   IBDEVICES - RDMA/InfiniBand device names (e.g., ionic_0,ionic_1,... or mlx5_0,mlx5_1,...)
 #               Set by runner or auto-detected from hostname.
@@ -119,10 +110,6 @@ else
     fi
 fi
 
-if [[ "$ENGINE" == "tilert" ]]; then
-    echo "[INFO] tilert: IBDEVICES=$IBDEVICES  NCCL_SOCKET_IFNAME=$NCCL_SOCKET_IFNAME  NCCL_IB_HCA=$NCCL_IB_HCA"
-
-else
 
     export SGLANG_USE_AITER=1
     export AITER_LOG_LEVEL=ERROR
@@ -237,5 +224,3 @@ else
             export GPU_MAX_HW_QUEUES=2
         fi
     fi
-
-fi

inferencex-e2e/benchmarks/multi_node/amd_utils/job.slurm

@@ -34,11 +34,7 @@ echo ""
 
 # Use $(pwd) not BASH_SOURCE — sbatch copies the script to /var/spool/slurmd/
 # at runtime, but the CWD remains the submit-time directory (amd_utils/).
-if [[ "$ENGINE" == "tilert" ]]; then
-    MODELS_YAML="$(pwd)/models_tilert.yaml"
-else
-    MODELS_YAML="$(pwd)/models.yaml"
-fi
+MODELS_YAML="$(pwd)/models.yaml"
 
 if [[ ! -f "$MODELS_YAML" ]]; then
     echo "Error: models YAML not found at $MODELS_YAML"
@@ -50,16 +46,6 @@ if [[ -z "${DOCKER_IMAGE_NAME:-}" ]]; then
     exit 1
 fi
 
-if [[ "$ENGINE" == "tilert" && -z "${PREFILL_IMAGE:-}" ]]; then
-    echo "Error: ENGINE=tilert requires PREFILL_IMAGE (e.g. PREFILL_IMAGE=vllm/vllm-openai-rocm:nightly-<sha> in prefill.additional-settings)."
-    exit 1
-fi
-if [[ "$ENGINE" == "tilert" ]]; then
-    # server_tilert.sh takes the container-creation barrier timeout from the
-    # recipe (no 300s default as on the SGLang path); fail here, before sbatch
-    # work is done, rather than inside the container.
-    check_env_vars CONTAINER_BARRIER_TIMEOUT
-fi
 
 # Resolve the models.yaml entry the same way server_sglang.sh does: agentic runs
 # (IS_AGENTIC) use the '<model>-AgentX' recipe, non-agentic disaggregated runs use
@@ -541,46 +527,9 @@ DOCKER_ENV_COMMON=(
 )
 
 # Engine-specific env vars
-if [[ "$ENGINE" == "tilert" ]]; then
-    DOCKER_ENV_ENGINE=(
-        -e MODEL_PATH=$DOCKER_MODEL_PATH
-        -e PREFILL_IMAGE=${PREFILL_IMAGE}
-        -e TILERT_VERSION=${TILERT_VERSION}
-        -e TILERT_PROFILE=${TILERT_PROFILE}
-        -e TILERT_MODEL_TYPE=${TILERT_MODEL_TYPE}
-        -e TILERT_MODEL_PKG=${TILERT_MODEL_PKG}
-        -e TILERT_MAX_MODEL_LEN=${TILERT_MAX_MODEL_LEN}
-        -e TILERT_TRANSPORT=${TILERT_TRANSPORT}
-        -e TILERT_PARSER=${TILERT_PARSER}
-        -e TILERT_QUEUE_TIMEOUT=${TILERT_QUEUE_TIMEOUT}
-        -e TILERT_WEIGHTS_DIR=${TILERT_WEIGHTS_DIR}
-        -e TILERT_RDMA_STRICT=${TILERT_RDMA_STRICT}
-        -e TILERT_CONVERT_LOCK_WAIT=${TILERT_CONVERT_LOCK_WAIT}
-        -e TILERT_SIMULATE_ACC_METHOD=${TILERT_SIMULATE_ACC_METHOD}
-        -e \"TILERT_EXTRA_ENV=${TILERT_EXTRA_ENV:-}\"
-        -e SERVED_MODEL_NAME=${SERVED_MODEL_NAME}
-        -e PREFILL_KV_DTYPE=${PREFILL_KV_DTYPE}
-        -e PREFILL_BLOCK_SIZE=${PREFILL_BLOCK_SIZE}
-        -e PREFILL_SPEC_TOKENS=${PREFILL_SPEC_TOKENS}
-        -e DECODE_KV_DTYPE=${DECODE_KV_DTYPE}
-        -e GPU_MEM_UTIL=${GPU_MEM_UTIL}
-        -e PREFILL_PORT=${PREFILL_PORT}
-        -e DECODE_CTRL_PORT=${DECODE_CTRL_PORT}
-        -e DECODE_HTTP_PORT=${DECODE_HTTP_PORT}
-        -e DECODE_WAIT=${DECODE_WAIT}
-        -e PREFILL_WAIT=${PREFILL_WAIT}
-        -e ROUTER_WAIT=${ROUTER_WAIT}
-        -e SKIP_CONTAINER_BARRIER=${SKIP_CONTAINER_BARRIER}
-        # Golden-acceptance selection on agentic MTP runs; unset elsewhere.
-        -e THINKING_MODE=${THINKING_MODE:-}
-        -e IBDEVICES=${IBDEVICES:-}
-        -e PYTHONPYCACHEPREFIX=/tmp/pycache
-    )
-else
-    DOCKER_ENV_ENGINE=(
-        -e SGLANG_WS_PATH=${WS_PATH}
-    )
-fi
+DOCKER_ENV_ENGINE=(
+    -e SGLANG_WS_PATH=${WS_PATH}
+)
 
 # HiCache / Mooncake settings are delivered via a bind-mounted config file rather
 # than a long list of docker -e flags. Write it once to the shared benchmark-logs
@@ -650,12 +599,6 @@ echo \"Rank \$SLURM_PROCID on \$(hostname)\"
 eval \"\$DOCKER_CMD_DETECT\"
 echo \"[docker-detect] rank \$SLURM_PROCID: DOCKER_CMD=\$DOCKER_CMD\"
 
-RANK_IMAGE=
-if [[ \"$ENGINE\" == \"tilert\" && \"\$SLURM_PROCID\" -lt \"$xP\" ]]; then
-    RANK_IMAGE=\"$PREFILL_IMAGE\"
-    echo \"[tilert] rank \$SLURM_PROCID is a prefill rank; using PREFILL_IMAGE=\$RANK_IMAGE\"
-fi
-
 exec \$DOCKER_CMD run \
     --init \
     --stop-timeout 10 \
@@ -696,7 +639,7 @@ exec \$DOCKER_CMD run \
     ${CLIENT_DOCKER_ENV} \
     --name \"$DOCKER_CONT_NAME\" \
     --entrypoint \"\" \
-    \"\${RANK_IMAGE:-$DOCKER_IMAGE_NAME}\" bash -lc '
+    \"$DOCKER_IMAGE_NAME\" bash -lc '
         set -o pipefail
         mkdir -p /run_logs/slurm_job-'\"\$SLURM_JOB_ID\"'
         '"$RUN_FILE_FULL"' 2>&1 | tee /run_logs/slurm_job-'\"\$SLURM_JOB_ID\"'/server_\$(hostname).log

inferencex-e2e/benchmarks/multi_node/amd_utils/server.sh

@@ -4,7 +4,6 @@ source "$(dirname "${BASH_SOURCE[0]}")/../../benchmark_lib.sh" --validation-only
 # Multi-Engine Disaggregated Server Dispatcher
 # Dispatches to the engine-specific server launcher based on ENGINE env var.
 #   ENGINE=sglang-disagg (default) -> server_sglang.sh (SGLang + MoRI)
-#   ENGINE=tilert                  -> server_tilert.sh (vLLM prefill + TileRT decode)
 
 check_env_vars ENGINE WS_PATH
 if [[ -f /config/hicache_mc.env ]]; then
@@ -16,8 +15,4 @@ export WS_PATH ENGINE
 
 echo "[DISPATCHER] ENGINE=$ENGINE  WS_PATH=$WS_PATH"
 
-if [[ "$ENGINE" == "tilert" ]]; then
-    source "$WS_PATH/server_tilert.sh"
-else
-    source "$WS_PATH/server_sglang.sh"
-fi
+source "$WS_PATH/server_sglang.sh"