diff --git a/benchmarks/benchmark_lib.sh b/benchmarks/benchmark_lib.sh
index a4fe70803f..9cec822cca 100644
--- a/benchmarks/benchmark_lib.sh
+++ b/benchmarks/benchmark_lib.sh
@@ -1996,7 +1996,7 @@ build_replay_cmd() {
     # rolling TTFT/ITL/throughput block and emit it every 30 seconds.
     export AIPERF_UI_REALTIME_METRICS_ENABLED=true
     REPLAY_CMD="$AIPERF_CLI profile --scenario inferencex-agentx-mvp"
-    REPLAY_CMD+=" --url http://localhost:$PORT"
+    REPLAY_CMD+=" --url ${AIPERF_SERVER_URL:-http://localhost:$PORT}"
     REPLAY_CMD+=" --endpoint /v1/chat/completions"
     REPLAY_CMD+=" --endpoint-type chat"
     REPLAY_CMD+=" --streaming"
diff --git a/benchmarks/multi_node/agentic_srt.sh b/benchmarks/multi_node/agentic_srt.sh
index 79a36da524..77c196c133 100644
--- a/benchmarks/multi_node/agentic_srt.sh
+++ b/benchmarks/multi_node/agentic_srt.sh
@@ -9,7 +9,12 @@ set -x
 INFMAX_CONTAINER_WORKSPACE="${INFMAX_CONTAINER_WORKSPACE:-/infmax-workspace}"
 source "$INFMAX_CONTAINER_WORKSPACE/benchmarks/benchmark_lib.sh"
 
-check_env_vars MODEL MODEL_PREFIX FRAMEWORK PRECISION CONC RESULT_FILENAME DURATION
+check_env_vars \
+    MODEL MODEL_PREFIX FRAMEWORK PRECISION CONC RESULT_FILENAME DURATION \
+    SRT_FRONTEND_HOST SRT_FRONTEND_PORT
+
+export AIPERF_SERVER_URL="http://${SRT_FRONTEND_HOST}:${SRT_FRONTEND_PORT}"
+echo "Using srt-slurm frontend endpoint: $AIPERF_SERVER_URL"
 
 BASE_RESULT_DIR="${RESULT_DIR:-/logs/agentic}"
 BASE_RESULT_FILENAME="$RESULT_FILENAME"
@@ -32,7 +37,7 @@ install_agentic_deps
 wait_for_agentic_servers_idle() {
     local timeout_seconds="${AIPERF_DRAIN_TIMEOUT_SECONDS:-1800}"
     local poll_seconds="${AIPERF_DRAIN_POLL_SECONDS:-10}"
-    local frontend_metrics_url="http://localhost:${PORT}/metrics"
+    local frontend_metrics_url="${AIPERF_SERVER_URL%/}/metrics"
 
     "$AIPERF_PYTHON" - \
         "$timeout_seconds" \
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p1d-tep8-c20-b5-mtp5.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p1d-tep8-c20-b5-mtp5.yaml
new file mode 100644
index 0000000000..54e5eda1e0
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p1d-tep8-c20-b5-mtp5.yaml
@@ -0,0 +1,204 @@
+name: dynamo-disagg-b300-1p1d-tep8-c20-b5-mtp5
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+  gpus_per_decode: 8
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "2.61"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: true
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 412316860416
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+        context_chunking_policy: EQUAL_PROGRESS
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 5
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1, 2, 4, 5]
+        enable_padding: true
+      enable_attention_dp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.8
+        tokens_per_block: 64
+      max_batch_size: 5
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: TRTLLM
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 8
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 5
+      stream_interval: 20
+      tensor_parallel_size: 8
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '20'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_1p1d-tep8_c20
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p3d-tep8-c1-b1-mtp7.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p3d-tep8-c1-b1-mtp7.yaml
new file mode 100644
index 0000000000..7818fd3435
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p3d-tep8-c1-b1-mtp7.yaml
@@ -0,0 +1,204 @@
+name: dynamo-disagg-b300-1p3d-tep8-c1-b1-mtp7
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 3
+  decode_workers: 3
+  gpus_per_decode: 8
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "2.91"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: true
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 412316860416
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+        context_chunking_policy: EQUAL_PROGRESS
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 7
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1]
+        enable_padding: true
+      enable_attention_dp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.8
+        tokens_per_block: 64
+      max_batch_size: 1
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: TRTLLM
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 8
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 7
+      stream_interval: 20
+      tensor_parallel_size: 8
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '1'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_1p3d-tep8_c1
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p4d-tep4-c30-b2-mtp5.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p4d-tep4-c30-b2-mtp5.yaml
new file mode 100644
index 0000000000..cffce82663
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p4d-tep4-c30-b2-mtp5.yaml
@@ -0,0 +1,204 @@
+name: dynamo-disagg-b300-1p4d-tep4-c30-b2-mtp5
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 4
+  decode_workers: 4
+  gpus_per_decode: 4
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "2.61"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: true
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 274877906944
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+        context_chunking_policy: EQUAL_PROGRESS
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 5
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1, 2]
+        enable_padding: true
+      enable_attention_dp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.8
+        tokens_per_block: 64
+      max_batch_size: 2
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: TRTLLM
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 5
+      stream_interval: 20
+      tensor_parallel_size: 4
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '30'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_1p4d-tep4_c30
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-3p4d-tep4-c60-b5-mtp5.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-3p4d-tep4-c60-b5-mtp5.yaml
new file mode 100644
index 0000000000..70ed9ff108
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-3p4d-tep4-c60-b5-mtp5.yaml
@@ -0,0 +1,204 @@
+name: dynamo-disagg-b300-3p4d-tep4-c60-b5-mtp5
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 3
+  prefill_workers: 3
+  gpus_per_prefill: 4
+  decode_nodes: 4
+  decode_workers: 4
+  gpus_per_decode: 4
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "2.61"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: true
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 197568495616
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+        context_chunking_policy: EQUAL_PROGRESS
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 5
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1, 2, 4, 5]
+        enable_padding: true
+      enable_attention_dp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.8
+        tokens_per_block: 64
+      max_batch_size: 5
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: TRTLLM
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 5
+      stream_interval: 20
+      tensor_parallel_size: 4
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '60'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_3p4d-tep4_c60
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-6p1d-dep8-c227-b16-mtp3.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-6p1d-dep8-c227-b16-mtp3.yaml
new file mode 100644
index 0000000000..2922fd5d37
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-6p1d-dep8-c227-b16-mtp3.yaml
@@ -0,0 +1,205 @@
+name: dynamo-disagg-b300-6p1d-dep8-c227-b16-mtp3
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 6
+  prefill_workers: 6
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+  gpus_per_decode: 8
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "1.99"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: true
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 197568495616
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+        context_chunking_policy: EQUAL_PROGRESS
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 3
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1, 2, 4, 8, 16]
+        enable_padding: true
+      enable_attention_dp: true
+      enable_lm_head_tp_in_adp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.9
+        tokens_per_block: 64
+      max_batch_size: 16
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 8
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 3
+      stream_interval: 20
+      tensor_parallel_size: 8
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '227'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_6p1d-dep8_c227
diff --git a/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-8p2d-dep8-c233-b16-mtp3.yaml b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-8p2d-dep8-c233-b16-mtp3.yaml
new file mode 100644
index 0000000000..1f9f2d9498
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-8p2d-dep8-c233-b16-mtp3.yaml
@@ -0,0 +1,204 @@
+name: dynamo-disagg-b300-8p2d-dep8-c233-b16-mtp3
+
+model:
+  path: nvidia/GLM-5.2-NVFP4
+  container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  precision: fp4
+
+identity:
+  model:
+    repo: "nvidia/GLM-5.2-NVFP4"
+    revision: "aec724e8c7b8ee9db3b48c01c320f63f9cdaf8aa"
+  container:
+    image: "nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1"
+  frameworks:
+    dynamo: "1.4.0.dev20260807"
+    tensorrt_llm: "1.3.0rc22.post1"
+
+dynamo:
+  install: true
+  wheel: 1.4.0.dev20260807
+  request_plane: tcp
+
+slurm:
+  time_limit: "4:00:00"
+
+health_check:
+  max_attempts: 270
+  interval_seconds: 10
+
+resources:
+  gpu_type: b300
+  gpus_per_node: 8
+  spread_workers: true
+  prefill_nodes: 8
+  prefill_workers: 8
+  gpus_per_prefill: 4
+  decode_nodes: 2
+  decode_workers: 2
+  gpus_per_decode: 8
+
+backend:
+  type: trtllm
+  prefill_environment: &server_environment
+    HF_HUB_OFFLINE: "1"
+    TRANSFORMERS_OFFLINE: "1"
+    TQDM_DISABLE: "1"
+    HF_HUB_DISABLE_PROGRESS_BARS: "1"
+    TLLM_LOG_LEVEL: INFO
+    TRTLLM_SERVER_DISABLE_GC: "1"
+    TRTLLM_WORKER_DISABLE_GC: "1"
+    TRTLLM_ENABLE_PDL: "1"
+    NCCL_GRAPH_MIXING_SUPPORT: "0"
+    MIMALLOC_PURGE_DELAY: "0"
+    PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+    TRTLLM_SERVE_ENABLE_MSGSPEC: "1"
+    TLLM_ADP_ROUTER_MATCH_RATE_THRESHOLD: "0.10"
+    TRTLLM_KV_CACHE_TRANSFER_TIMEOUT_SEC: "600"
+    UCX_TLS: rc,cuda_ipc,cuda_copy,sm,self,tcp
+    UCX_MAX_RNDV_RAILS: "2"
+    UCX_MAX_HCA_PER_GPU: inf
+    UCX_LOG_LEVEL: info
+    TRTLLM_CTX_LOCAL_HCA_PIN: "1"
+    UCX_RNDV_SCHEME: put_zcopy
+    TRTLLM_KVCACHE_SEND_BUFFER_COUNT: "1"
+    TRTLLM_KVCACHE_RECV_BUFFER_COUNT: "1"
+    TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS: "1.99"
+    DYN_TRTLLM_ENABLE_ATTENTION_DP: "1"
+    DYN_ENGINE_CONV_AFFINITY: "1"
+    DYN_TOKENIZER: fastokens
+    DYN_PUBLISH_KV_EVENTS: "0"
+  decode_environment: *server_environment
+  trtllm_config:
+    prefill:
+      attention_dp_config:
+        enable_kv_cache_aware_routing: false
+        kv_cache_routing_conversation_affinity: true
+        kv_cache_routing_max_sessions: 65536
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config: null
+      disable_overlap_scheduler: false
+      enable_attention_dp: true
+      enable_chunked_prefill: true
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: true
+        event_buffer_max_size: 0
+        free_gpu_memory_fraction: 0.75
+        host_cache_size: 197568495616
+        tokens_per_block: 64
+      max_batch_size: 256
+      max_num_tokens: 8192
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+      moe_expert_parallel_size: 4
+      num_postprocess_workers: 8
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      scheduler_config:
+        capacity_scheduler_policy: MAX_UTILIZATION
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+      speculative_config:
+        decoding_type: MTP
+        num_nextn_predict_layers: 3
+      tensor_parallel_size: 4
+    decode:
+      cache_transceiver_config:
+        backend: NIXL
+        transceiver_runtime: PYTHON
+        kv_cache_bounce_size_mb: 5120
+        max_tokens_in_buffer: 1048576
+        kv_transfer_timeout_ms: 600000
+      cuda_graph_config:
+        batch_sizes: [1, 2, 4, 8, 16]
+        enable_padding: true
+      enable_attention_dp: true
+      enable_lm_head_tp_in_adp: false
+      trust_remote_code: true
+      kv_cache_config:
+        dtype: fp8
+        enable_block_reuse: false
+        event_buffer_max_size: 0
+        host_cache_size: 0
+        free_gpu_memory_fraction: 0.9
+        tokens_per_block: 64
+      max_batch_size: 16
+      max_num_tokens: 128
+      max_seq_len: 1048576
+      moe_config:
+        backend: CUTEDSL
+        use_low_precision_moe_combine: true
+      moe_expert_parallel_size: 8
+      num_postprocess_workers: 4
+      pipeline_parallel_size: 1
+      print_iter_log: true
+      return_perf_metrics: false
+      sparse_attention_config:
+        algorithm: dsa
+        enable_heuristic_topk: true
+        use_cute_dsl_paged_mqa_logits: true
+      speculative_config:
+        decoding_type: MTP
+        max_draft_len: 3
+      stream_interval: 20
+      tensor_parallel_size: 8
+  publish_events_and_metrics: false
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+  env:
+    ETCD_LEASE_TTL: '120'
+    DYN_ROUTER_QUEUE_THRESHOLD: None
+    DYN_ROUTER_TEMPERATURE: '0'
+    DYN_TOKENIZER_CACHE: '1'
+    DYN_TOKENIZER_CACHE_BYTES: '8000000000'
+    DYN_TOKENIZER: fastokens
+    DYN_TCP_REQUEST_TIMEOUT: '30'
+    DYN_LOG: warn
+    DYN_ROUTER_SESSION_AFFINITY_TTL_SECS: '14400'
+  args:
+    router-mode: kv
+    no-kv-events: true
+    active-decode-blocks-threshold: None
+    active-prefill-tokens-threshold: None
+    active-prefill-tokens-threshold-frac: None
+  orchestrator_placement: first_decode
+
+benchmark:
+  type: custom
+  client_placement: last_decode
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: /infmax-workspace
+    RESULT_DIR: /logs/agentic
+    PORT: '8000'
+    IS_MULTINODE: 'true'
+    AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400'
+    AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+    HF_HUB_CACHE: /hf_hub_cache
+    WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126
+    AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES: "0"
+    SERVED_MODEL_NAME: GLM-5.2-NVFP4
+    MAX_MODEL_LEN: '1048576'
+    AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+    AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+    OPENAI_API_KEY: EMPTY
+    KV_OFFLOADING: none
+    MODEL: nvidia/GLM-5.2-NVFP4
+    MODEL_PREFIX: glm5.2
+    FRAMEWORK: dynamo-trt
+    PRECISION: fp4
+    CONC: '233'
+    DURATION: '3600'
+    RESULT_FILENAME: glm52_fp4_dynamo-trt_8p2d-dep8_c233
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 50ed9b4713..e541bee50b 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -8471,6 +8471,113 @@ glm5.2-fp4-b300-sglang-agentic-mtp:
       search-space:
       - { tp: 8, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: hicache }, conc-list: [1, 4, 8, 12, 16] }
 
+# GLM-5.2 NVFP4 TensorRT-LLM AgentX on B300 with Dynamo disaggregated serving.
+# Seven topology variants use NIXL KV transfer and MTP3/MTP5/MTP7 decoding.
+glm5.2-fp4-b300-dynamo-trt-agentic-mtp:
+  image: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc22.post1
+  model: nvidia/GLM-5.2-NVFP4
+  model-prefix: glm5.2
+  runner: cluster:b300-nv
+  precision: fp4
+  framework: dynamo-trt
+  router: { name: dynamo-router, version: "1.4.0.dev20260807" }
+  kv-p2p-transfer: nixl
+  multinode: true
+  disagg: true
+  scenarios:
+    agentic-coding:
+    - search-space:
+      - spec-decoding: mtp
+        conc-list: [1]
+        kv-offloading: none
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p3d-tep8-c1-b1-mtp7.yaml"
+        decode:
+          num-worker: 3
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - spec-decoding: mtp
+        conc-list: [30]
+        kv-offloading: none
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p4d-tep4-c30-b2-mtp5.yaml"
+        decode:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: false
+      - spec-decoding: mtp
+        conc-list: [20]
+        kv-offloading: none
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-1p1d-tep8-c20-b5-mtp5.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - spec-decoding: mtp
+        conc-list: [60]
+        kv-offloading: none
+        prefill:
+          num-worker: 3
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-3p4d-tep4-c60-b5-mtp5.yaml"
+        decode:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: false
+      - spec-decoding: mtp
+        conc-list: [233]
+        kv-offloading: none
+        prefill:
+          num-worker: 8
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-8p2d-dep8-c233-b16-mtp3.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: true
+      - spec-decoding: mtp
+        conc-list: [227]
+        kv-offloading: none
+        prefill:
+          num-worker: 6
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2/b300-fp4/agentic/dynamo-disagg-b300-6p1d-dep8-c227-b16-mtp3.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
+
 # First GLM-5.2 AgentX recipe on B200, shipped spec-decode-only per the AgentX
 # policy that agentic arms enable speculative decoding rather than running a
 # separate STP baseline (MODELS.md). SGLang EAGLE off GLM-5.2's built-in nextn
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 7500c62923..a2584461c7 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -6234,3 +6234,11 @@
   description:
     - "Add GB200 Kimi K3 AgentX config."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2639
+
+- config-keys:
+    - glm5.2-fp4-b300-dynamo-trt-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Add GLM-5.2 NVFP4 B300 AgentX recipes for Dynamo and TensorRT-LLM disaggregated serving with NIXL KV transfer and MTP speculative decoding."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2666
diff --git a/runners/launch_b300-nv.sh b/runners/launch_b300-nv.sh
index f7878a8ad8..a237d37495 100644
--- a/runners/launch_b300-nv.sh
+++ b/runners/launch_b300-nv.sh
@@ -44,6 +44,10 @@ elif [[ $MODEL_PREFIX == "dsv4" && $PRECISION == "fp4" && $FRAMEWORK == "dynamo-
 elif [[ $MODEL_PREFIX == "dsv4" && $PRECISION == "fp4" && $FRAMEWORK == "dynamo-sglang" ]]; then
     export MODEL_PATH="${MODEL_PATH:-/scratch/models/DeepSeek-V4-Pro}"
     export SRT_SLURM_MODEL_PREFIX="deepseek-v4-pro"
+elif [[ $MODEL_PREFIX == "glm5.2" && $PRECISION == "fp4" && $FRAMEWORK == "dynamo-trt" ]]; then
+    export SERVED_MODEL_NAME="nvidia/GLM-5.2-NVFP4"
+    export MODEL_PATH="/scratch/models/GLM-5.2-NVFP4"
+    export SRT_SLURM_MODEL_PREFIX="nvidia/GLM-5.2-NVFP4"
 elif [[ $MODEL_PREFIX == "minimaxm2.5" && $PRECISION == "fp4" && $FRAMEWORK == "dynamo-vllm" ]]; then
     export MODEL_PATH="/data/models/MiniMax-M2.5-NVFP4"
     export SRT_SLURM_MODEL_PREFIX="minimax-m2.5-nvfp4"
@@ -57,7 +61,7 @@ elif [[ $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp8" && $FRAMEWORK == "dy
     export MODEL_PATH="/data/models/MiniMax-M3-MXFP8"
     export SRT_SLURM_MODEL_PREFIX="MiniMaxAI/MiniMax-M3-MXFP8"
 else
-    echo "Unsupported model: $MODEL_PREFIX-$PRECISION. Supported models are: dsr1-fp4, dsr1-fp8, dsv4-fp4 with dynamo-vllm or dynamo-sglang, minimaxm2.5-fp4 with dynamo-vllm, minimaxm2.5-fp8 with dynamo-vllm, minimaxm3-fp4 with dynamo-vllm, minimaxm3-fp8 with dynamo-vllm"
+    echo "Unsupported model: $MODEL_PREFIX-$PRECISION. Supported models are: dsr1-fp4, dsr1-fp8, dsv4-fp4 with dynamo-vllm or dynamo-sglang, glm5.2-fp4 with dynamo-trt, minimaxm2.5-fp4 with dynamo-vllm, minimaxm2.5-fp8 with dynamo-vllm, minimaxm3-fp4 with dynamo-vllm, minimaxm3-fp8 with dynamo-vllm"
     exit 1
 fi
 
@@ -70,7 +74,20 @@ if [ -d "$SRT_REPO_DIR" ]; then
 fi
 
 # TODO(CJQ): make first class upon srt-slurm upstream refactor
-if [[ "$IS_AGENTIC" == "1" ]]; then
+if [[ "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-trt" && $MODEL_PREFIX == "glm5.2" ]]; then
+    git clone --branch v1.0.36 --single-branch https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
+    cd "$SRT_REPO_DIR" || exit 1
+    # This cluster requires root remapping to remain disabled for PMIx.
+    sed -i 's/CONTAINER_REMAP_ROOT_EXPORT = {"ENROOT_REMAP_ROOT": "yes"}/CONTAINER_REMAP_ROOT_EXPORT = {"ENROOT_REMAP_ROOT": "no"}/' src/srtctl/core/slurm.py
+    TRTLLM_RECIPES_DIR="benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2"
+    mkdir -p "$TRTLLM_RECIPES_DIR"
+    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/trtllm/glm5.2" \
+        "$TRTLLM_RECIPES_DIR"
+    if [[ "${EVAL_ONLY:-false}" == "true" ]]; then
+        find "$TRTLLM_RECIPES_DIR" -name "*.yaml" \
+            -exec sed -i '/TLLM_SPEC_DECODE_FORCE_NUM_ACCEPTED_TOKENS/d' {} +
+    fi
+elif [[ "$IS_AGENTIC" == "1" ]]; then
     git clone --branch cam/sa-submission-q2-2026 --single-branch https://github.com/cquil11/srt-slurm-nv.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR" || exit 1
 elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "dsv4" ]]; then
@@ -143,6 +160,23 @@ export ISL="$ISL"
 export OSL="$OSL"
 export EVAL_ONLY="${EVAL_ONLY:-false}"
 
+DEFAULT_MOUNTS_BLOCK='default_mounts:'
+if [[ $FRAMEWORK != "dynamo-trt" || $MODEL_PREFIX != "glm5.2" ]]; then
+    DEFAULT_MOUNTS_BLOCK+='
+  "/opt/ucx-no-ud": "/usr/local/ucx"'
+else
+    echo "Using the container UCX for glm5.2 dynamo-trt NIXL transfers"
+fi
+if [[ "$IS_AGENTIC" == "1" ]]; then
+    AIPERF_MMAP_CACHE_HOST_PATH="/data/home/sa-shared/gharunners/ai-perf-cache"
+    HF_HUB_CACHE_HOST_PATH="/data/home/sa-shared/gharunners/hf-hub-cache"
+    mkdir -p "$AIPERF_MMAP_CACHE_HOST_PATH" "$HF_HUB_CACHE_HOST_PATH"
+    chmod 777 "$AIPERF_MMAP_CACHE_HOST_PATH" "$HF_HUB_CACHE_HOST_PATH" 2>/dev/null || true
+    DEFAULT_MOUNTS_BLOCK+="
+  \"${AIPERF_MMAP_CACHE_HOST_PATH}\": \"/aiperf_mmap_cache\"
+  \"${HF_HUB_CACHE_HOST_PATH}\": \"/hf_hub_cache\""
+fi
+
 # Create srtslurm.yaml for srtctl
 SRTCTL_ROOT="${GITHUB_WORKSPACE}/${SRT_REPO_DIR}"
 echo "Creating srtslurm.yaml configuration..."
@@ -169,8 +203,7 @@ containers:
   "${IMAGE}": "${SQUASH_FILE}"
   nginx-sqsh: "${NGINX_SQUASH_FILE}"
 use_exclusive_sbatch_directive: true
-default_mounts:
-  "/opt/ucx-no-ud": "/usr/local/ucx"
+${DEFAULT_MOUNTS_BLOCK}
 EOF
 
 echo "Generated srtslurm.yaml:"
@@ -198,6 +231,16 @@ if [[ ! -f "$CONFIG_PATH" ]]; then
     exit 1
 fi
 
+if [[ "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-trt" && $MODEL_PREFIX == "glm5.2" ]]; then
+    sed -i "s|^    RESULT_FILENAME:.*|    RESULT_FILENAME: ${RESULT_FILENAME}|" "$CONFIG_PATH"
+fi
+
+# Eval runs execute lm-eval on the allocation head and connect over loopback.
+# Throughput keeps the frontend on first_decode and the client on last_decode.
+if [[ "${EVAL_ONLY:-false}" == "true" && "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-trt" && $MODEL_PREFIX == "glm5.2" ]]; then
+    sed -i 's/^  orchestrator_placement: first_decode$/  orchestrator_placement: head/' "$CONFIG_PATH"
+fi
+
 # Override the job name in the recipe with the runner name.
 sed -i "s/^name:.*/name: \"${RUNNER_NAME}\"/" "$CONFIG_PATH"
 if [[ "$MODEL_PREFIX" == "minimaxm3" && -n "$MINIMAX_M3_SLURM_EXCLUDED_NODELIST" ]]; then
@@ -218,6 +261,9 @@ fi
 if [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "dsv4" && "$MODEL_PATH" == /scratch/models/* ]]; then
     SRTCTL_APPLY_ARGS+=(--no-preflight)
 fi
+if [[ $FRAMEWORK == "dynamo-trt" && $MODEL_PREFIX == "glm5.2" && "$MODEL_PATH" == /scratch/models/* ]]; then
+    SRTCTL_APPLY_ARGS+=(--no-preflight)
+fi
 if [[ -n "$SRTCTL_SETUP_SCRIPT" ]]; then
     SRTCTL_APPLY_ARGS+=(--setup-script "$SRTCTL_SETUP_SCRIPT")
 fi
@@ -353,6 +399,19 @@ if [[ "${RUN_EVAL:-false}" == "true" || "${EVAL_ONLY:-false}" == "true" ]]; then
     else
         echo "WARNING: RUN_EVAL=true but no eval results found at $EVAL_DIR"
     fi
+
+    # srt-slurm stages eval artifacts but does not write the metadata file
+    # consumed by score validation. Reuse the canonical metadata writer so
+    # topology and recipe identity stay aligned with the workflow inputs.
+    eval_conc_value="${EVAL_CONC:-${CONC:-1}}"
+    (
+        export IS_MULTINODE=true
+        # shellcheck source=benchmarks/benchmark_lib.sh
+        source "$GITHUB_WORKSPACE/benchmarks/benchmark_lib.sh"
+        _write_lm_eval_meta_json \
+            "$GITHUB_WORKSPACE/meta_env.json" "" "$eval_conc_value"
+    )
+    echo "Wrote meta_env.json (conc=${eval_conc_value}, prefix=${MODEL_PREFIX:-unknown})"
 fi
 
 # Clean up srt-slurm outputs to prevent NFS silly-rename lock files