From 6efd246c81a3f129b16bb6d665bca1dc0e666fcd Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 00:56:07 -0700
Subject: [PATCH 01/11] fix(changelog): accept single-node agentic eval rows in
 matrix validation

#1947 made single-node agentic recipes generate a SWE-bench eval row
(run_eval/eval_only + agentic fields) but never widened the changelog
matrix schema, so ChangelogMatrixEntry.evals (typed list[SingleNodeMatrixEntry],
fixed-seq-len only) rejects every agentic eval row -- breaking check-changelog
for any single-node agentic PR.

Widen evals to the same Union single_node already uses, and give
SingleNodeAgenticMatrixEntry optional run_eval/eval_only (None-default, so
benchmark rows are unchanged).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 utils/matrix_logic/validation.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/utils/matrix_logic/validation.py b/utils/matrix_logic/validation.py
index 6b7c671289..7403c795e4 100644
--- a/utils/matrix_logic/validation.py
+++ b/utils/matrix_logic/validation.py
@@ -291,6 +291,10 @@ class SingleNodeAgenticMatrixEntry(BaseModel):
     duration: int = Field(alias=Fields.DURATION.value)
     exp_name: str = Field(alias=Fields.EXP_NAME.value)
     scenario_type: str = Field(alias=Fields.SCENARIO_TYPE.value)
+    # Agentic eval rows (SWE-bench) carry run_eval/eval_only; benchmark rows omit
+    # them. Optional so both shapes validate and benchmark output is unchanged.
+    run_eval: Optional[bool] = Field(default=None, alias=Fields.RUN_EVAL.value)
+    eval_only: Optional[bool] = Field(default=None, alias=Fields.EVAL_ONLY.value)
 
     @model_validator(mode='after')
     def validate_kv_offload_fields(self):
@@ -889,7 +893,7 @@ class ChangelogMatrixEntry(BaseModel):
                       ] = Field(default_factory=dict)
     multi_node: dict[str, list[Union[MultiNodeMatrixEntry, MultiNodeAgenticMatrixEntry]]
                      ] = Field(default_factory=dict)
-    evals: list[SingleNodeMatrixEntry] = Field(default_factory=list)
+    evals: list[Union[SingleNodeMatrixEntry, SingleNodeAgenticMatrixEntry]] = Field(default_factory=list)
     multinode_evals: list[MultiNodeMatrixEntry] = Field(default_factory=list)
     changelog_metadata: ChangelogMetadata
 

From a53a365dbc9fde13ab1520c036ba36a09be01865 Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 01:11:10 -0700
Subject: [PATCH 02/11] [AgentX] DeepSeek-V4 B200 vLLM: add MTP
 (num_speculative_tokens=3) for every agentic arm

Add an MTP speculative-decoding twin (spec-decoding: mtp,
num_speculative_tokens=3) for every arm of dsv4-fp4-b200-vllm-agentic
(TP8 GPU-resident, TP8 SimpleCPU, DEP8 SimpleCPU, DEP8 Mooncake), each
mirroring its non-MTP conc-list, routed via the launcher's
spec-decoding=mtp suffix to dsv4_fp4_b200_vllm_mtp.sh.

New dsv4_fp4_b200_vllm_mtp.sh forks dsv4_fp4_b200_vllm.sh with only the MTP
deltas: --speculative-config {"method":"mtp","num_speculative_tokens":3} and
--max-cudagraph-capture-size scaled to MAX_NUM_SEQS*(1+N) tokens so
FULL_DECODE_ONLY still covers the largest decode batches.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../agentic/dsv4_fp4_b200_vllm_mtp.sh         | 302 ++++++++++++++++++
 configs/nvidia-master.yaml                    |   8 +
 2 files changed, 310 insertions(+)
 create mode 100755 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
new file mode 100755
index 0000000000..391a047a71
--- /dev/null
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -0,0 +1,302 @@
+#!/usr/bin/env bash
+set -eo pipefail
+set -x
+
+# Agentic trace replay benchmark for DeepSeek-V4-Pro FP4 on B200 using vLLM,
+# with MTP speculative decoding (num_speculative_tokens=3).
+#
+# Identical to dsv4_fp4_b200_vllm.sh (same image, engine args, offload, GPU
+# topologies, and agentic aiperf rig) with exactly two MTP deltas:
+#   --speculative-config '{"method": "mtp", "num_speculative_tokens": 3}'
+#   --max-cudagraph-capture-size expressed in TOKENS (see the capture block below).
+#
+# Mirrors the fixed-seq-len parallelism options (pure TP and DEP) so the
+# agentic sweep can probe both interactivity and throughput regimes:
+#   pure TP (DP_ATTENTION=false, EP_SIZE=1):  attention TP-sharded across
+#       all $TP GPUs in a single engine. Lower TPOT, lower batch.
+#   TP+EP   (DP_ATTENTION=false, EP_SIZE>1):  attention TP-sharded, MoE
+#       experts EP-sharded within the TP group.
+#   DEP     (DP_ATTENTION=true, EP_SIZE>1):   per-DP-rank attention with
+#       experts EP-sharded across DP ranks (per the vLLM blog recipe).
+#       Highest aggregate throughput at large CONC.
+#
+# Image is configured in nvidia-master.yaml. block_size=256,
+# kv-cache-dtype=fp8, FLASHINFER_MLA_SPARSE_DSV4 attention with the FP4 indexer
+# cache, FULL_DECODE_ONLY cudagraph capture, and (in EP tiers) mega-MoE backend.
+#
+# Required env vars:
+#   MODEL, TP, CONC, KV_OFFLOADING, TOTAL_CPU_DRAM_GB, RESULT_DIR
+#
+# Pure TP is GPU-resident (KV_OFFLOADING=none). DEP tiers offload KV to host
+# DRAM: KV_OFFLOADING=dram requires KV_OFFLOAD_BACKEND=vllm-simple or mooncake.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+check_env_vars MODEL TP CONC KV_OFFLOADING TOTAL_CPU_DRAM_GB RESULT_DIR DURATION EP_SIZE DP_ATTENTION
+
+if [ -z "$DCP_SIZE" ]; then
+    DCP_SIZE=1
+fi
+if [ -z "$PCP_SIZE" ]; then
+    PCP_SIZE=1
+fi
+VLLM_CP_ARGS=()
+if [ "$DCP_SIZE" -gt 1 ]; then
+    VLLM_CP_ARGS+=(--decode-context-parallel-size "$DCP_SIZE")
+fi
+if [ "$PCP_SIZE" -gt 1 ]; then
+    VLLM_CP_ARGS+=(--prefill-context-parallel-size "$PCP_SIZE")
+fi
+
+GPU_COUNT=$TP
+if [[ ! "$GPU_COUNT" =~ ^[1-9][0-9]*$ ]]; then
+    echo "Error: GPU_COUNT must be a positive integer, got '$GPU_COUNT'" >&2
+    exit 1
+fi
+export GPU_COUNT
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+    echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+# `hf download` creates the target dir if missing and is itself idempotent.
+# When MODEL_PATH is unset (stand-alone runs), fall back to the HF_HUB_CACHE
+# Either way, MODEL_PATH is what the server is launched with.
+if [[ -n "$MODEL_PATH" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+fi
+nvidia-smi
+
+# ---- Resolve traces and install deps ----------------------------------------
+resolve_trace_source
+install_agentic_deps
+
+# vllm-project/router expands the one HTTP backend into one logical worker per
+# DP rank and sends X-data-parallel-rank on forwarded requests. aiperf's
+# X-Correlation-ID is stable for every turn of a conversation; alias it to the
+# router's preferred X-Session-ID header.
+USE_VLLM_ROUTER=false
+VLLM_BACKEND_PORT="$PORT"
+if [ "$DP_ATTENTION" = "true" ]; then
+    USE_VLLM_ROUTER=true
+    VLLM_BACKEND_PORT=$((PORT + 1))
+    VLLM_ROUTER_VERSION=0.1.14
+    VLLM_ROUTER_POLICY=consistent_hash
+    VLLM_ROUTER_METRICS_PORT=$((PORT + 10000))
+    export AIPERF_HTTP_X_SESSION_ID_FROM_CORRELATION_ID=1
+    agentic_pip_install --quiet "vllm-router==$VLLM_ROUTER_VERSION"
+fi
+
+# DeepSeek-V4-Pro weights are large; engine startup can exceed default 600s.
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+
+# vllm-project/vllm#43447 keeps local SWA prefix-cache tails sparsely, while
+# vllm-project/vllm#44774 applies the same reachability policy to Mooncake's
+# store mask. 32k matches the trace-replay tuning validated for this workload.
+export VLLM_PREFIX_CACHE_RETENTION_INTERVAL=32768
+export VLLM_USE_V2_MODEL_RUNNER=1
+export VLLM_USE_RUST_FRONTEND=1
+export VLLM_DSV4_MEGA_FP8_COMBINE=1
+export VLLM_RPC_TIMEOUT=600000
+
+# ---- Server config ----------------------------------------------------------
+SERVER_LOG="$RESULT_DIR/server.log"
+ROUTER_LOG="$RESULT_DIR/router.log"
+MOONCAKE_MASTER_LOG="$RESULT_DIR/mooncake_master.log"
+mkdir -p "$RESULT_DIR"
+
+SERVER_PID=""
+ROUTER_PID=""
+MOONCAKE_MASTER_PID=""
+
+OFFLOAD_ARGS=()
+case "$KV_OFFLOAD_BACKEND" in
+    "")
+        require_agentic_kv_offload_none
+        ;;
+    vllm-simple)
+        require_agentic_kv_offload_backend vllm-simple
+        CPU_BYTES_PER_RANK=$(( TOTAL_CPU_DRAM_GB * 1000 * 1000 * 1000 / GPU_COUNT ))
+        # Identical prefixes must hash to identical block keys across DP ranks.
+        export PYTHONHASHSEED=42
+        OFFLOAD_CONFIG=$(cat <<EOF
+{
+  "kv_connector": "SimpleCPUOffloadConnector",
+  "kv_role": "kv_both",
+  "kv_connector_extra_config": {
+    "cpu_bytes_to_use_per_rank": ${CPU_BYTES_PER_RANK},
+    "lazy_offload": false,
+    "enable_cross_layers_blocks": "true"
+  }
+}
+EOF
+)
+        OFFLOAD_ARGS=(
+            --kv-transfer-config
+            "$OFFLOAD_CONFIG"
+        )
+        ;;
+    mooncake)
+        require_agentic_kv_offload_backend mooncake
+        # Embedded mode contributes one segment per GPU rank to a shared
+        # distributed store, so pre-divide the aggregate host-memory budget.
+        PER_RANK_GB=$((TOTAL_CPU_DRAM_GB / GPU_COUNT))
+
+        MOONCAKE_VERSION=0.3.11.post1
+        agentic_pip_install --quiet --no-cache-dir --no-deps \
+            --force-reinstall "mooncake-transfer-engine-cuda13==$MOONCAKE_VERSION"
+        python3 -c "from mooncake.store import MooncakeDistributedStore" >/dev/null
+
+        MOONCAKE_MASTER_PORT=$((PORT + 12000))
+        MOONCAKE_CONFIG_PATH="$RESULT_DIR/mooncake_config.json"
+        cat > "$MOONCAKE_CONFIG_PATH" <<EOF
+{
+  "mode": "embedded",
+  "metadata_server": "P2PHANDSHAKE",
+  "master_server_address": "127.0.0.1:$MOONCAKE_MASTER_PORT",
+  "global_segment_size": "${PER_RANK_GB}GB",
+  "local_buffer_size": "4GB",
+  "protocol": "rdma",
+  "device_name": "mlx5_0,mlx5_1,mlx5_2,mlx5_3,mlx5_4,mlx5_5,mlx5_10,mlx5_11",
+  "enable_offload": false
+}
+EOF
+        export MOONCAKE_CONFIG_PATH
+        export MC_ENABLE_DEST_DEVICE_AFFINITY=1
+        # Identical prefixes must hash to identical store keys across DP ranks.
+        export PYTHONHASHSEED=0
+        export WITH_NVIDIA_PEERMEM=0
+        export MC_SLICE_SIZE=1048576
+        export MC_WORKERS_PER_CTX=4
+
+        # Each rank contributes a separate segment. Evict early enough to
+        # avoid an imbalanced rank exhausting its segment.
+        MOONCAKE_EVICTION_HIGH_WATERMARK_RATIO=0.80
+        MOONCAKE_EVICTION_RATIO=0.10
+        # Mooncake's default 5s read lease is shorter than the observed
+        # transfer latency for large DSv4 hybrid-KV loads on B200 TCP.
+        MOONCAKE_KV_LEASE_TTL=60s
+
+        echo "Starting Mooncake master on port $MOONCAKE_MASTER_PORT..."
+        mooncake_master --port "$MOONCAKE_MASTER_PORT" \
+            --eviction_high_watermark_ratio="$MOONCAKE_EVICTION_HIGH_WATERMARK_RATIO" \
+            --eviction_ratio="$MOONCAKE_EVICTION_RATIO" \
+            --default_kv_lease_ttl="$MOONCAKE_KV_LEASE_TTL" \
+            > "$MOONCAKE_MASTER_LOG" 2>&1 &
+        MOONCAKE_MASTER_PID=$!
+        sleep 2
+        if ! kill -0 "$MOONCAKE_MASTER_PID" 2>/dev/null; then
+            echo "Mooncake master died during startup." >&2
+            cat "$MOONCAKE_MASTER_LOG" >&2
+            exit 1
+        fi
+        unset VLLM_USE_SIMPLE_KV_OFFLOAD
+        OFFLOAD_ARGS=(
+            --kv-transfer-config
+            '{"kv_connector":"MooncakeStoreConnector","kv_role":"kv_both","kv_connector_extra_config":{"load_async":true}}'
+        )
+        ;;
+    *)
+        echo "Error: unsupported B200 KV_OFFLOAD_BACKEND='$KV_OFFLOAD_BACKEND'" >&2
+        exit 1
+        ;;
+esac
+
+PARALLEL_ARGS=(--tensor-parallel-size "$TP" --data-parallel-size 1)
+if [ "$DP_ATTENTION" = "true" ]; then
+    PARALLEL_ARGS=(--tensor-parallel-size 1 --data-parallel-size "$TP")
+fi
+
+EP_ARGS=()
+FAST_MOE_ARGS=()
+if [ "$EP_SIZE" -gt 1 ]; then
+    EP_ARGS=(--enable-expert-parallel)
+    FAST_MOE_ARGS=(
+        --moe-backend deep_gemm_amxf4_mega_moe
+        --enable-ep-weight-filter
+        --prefill-schedule-interval 16
+    )
+fi
+
+# AgentX concurrency counts live session trees, not individual requests.
+# Subagent fan-out can push instantaneous request concurrency above CONC, so
+# leave 2x headroom rather than clipping those bursts at the scheduler.
+MAX_NUM_SEQS=$((2 * CONC))
+
+# MTP: cudagraph capture sizes are in TOKENS. A uniform decode batch of S seqs
+# verifies S*(1+num_speculative_tokens) tokens, so cap capture at
+# MAX_NUM_SEQS*(1+N) tokens -- otherwise vLLM's FULL_DECODE_ONLY ladder tops out
+# at MAX_NUM_SEQS/(1+N) seqs and the largest decode batches fall back to eager.
+NUM_SPEC_TOKENS=3
+TOKENS_PER_SEQ=$((1 + NUM_SPEC_TOKENS))
+MAX_CUDAGRAPH_CAPTURE_SIZE=$((MAX_NUM_SEQS * TOKENS_PER_SEQ))
+
+echo "Starting vllm server..."
+export TORCH_CUDA_ARCH_LIST="10.0"
+export PYTHONNOUSERSITE=1
+export VLLM_FLOAT32_MATMUL_PRECISION=high
+
+{ set +x; } 2>/dev/null
+VLLM_CMD=(
+    vllm serve "$MODEL_PATH" --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --port "$VLLM_BACKEND_PORT"
+    --trust-remote-code
+    --kv-cache-dtype fp8
+    --block-size 256
+    --max-model-len 1048576
+    --gpu-memory-utilization 0.92
+    --numa-bind
+    --enable-cumem-allocator
+    --no-enable-flashinfer-autotune
+    --tokenizer-mode deepseek_v4
+    --reasoning-parser deepseek_v4
+    --attention-config '{"backend":"FLASHINFER_MLA_SPARSE_DSV4","use_prefill_query_quantization":true,"use_fp4_indexer_cache":true}'
+    --speculative-config "{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS}"
+    --no-disable-hybrid-kv-cache-manager
+    --disable-uvicorn-access-log
+    --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY","mode":0}'
+    --max-num-seqs "$MAX_NUM_SEQS"
+    --max-cudagraph-capture-size "$MAX_CUDAGRAPH_CAPTURE_SIZE"
+    "${PARALLEL_ARGS[@]}"
+    "${VLLM_CP_ARGS[@]}"
+    "${EP_ARGS[@]}"
+    "${FAST_MOE_ARGS[@]}"
+    "${OFFLOAD_ARGS[@]}"
+)
+printf '%q ' "${VLLM_CMD[@]}" | tee "$RESULT_DIR/vllm_command.txt"
+printf '\n' | tee -a "$RESULT_DIR/vllm_command.txt"
+"${VLLM_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+echo "Server PID: $SERVER_PID"
+
+wait_for_server_ready --port "$VLLM_BACKEND_PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+if [ "$USE_VLLM_ROUTER" = "true" ]; then
+    echo "Starting native vLLM router on port $PORT for $TP DP ranks..."
+    vllm-router \
+        --worker-urls "http://localhost:$VLLM_BACKEND_PORT" \
+        --policy "$VLLM_ROUTER_POLICY" \
+        --intra-node-data-parallel-size "$TP" \
+        --host 0.0.0.0 \
+        --port "$PORT" \
+        --prometheus-host 127.0.0.1 \
+        --prometheus-port "$VLLM_ROUTER_METRICS_PORT" \
+        --request-timeout-secs 14400 \
+        --disable-retries > "$ROUTER_LOG" 2>&1 &
+    ROUTER_PID=$!
+    echo "Router PID: $ROUTER_PID"
+    wait_for_server_ready --port "$PORT" --server-log "$ROUTER_LOG" --server-pid "$ROUTER_PID"
+fi
+
+if [ "${EVAL_ONLY}" = "true" ]; then
+    run_eval --port "$PORT"
+else
+    build_replay_cmd "$RESULT_DIR"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index ff922946c5..268bb86721 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -1784,10 +1784,18 @@ dsv4-fp4-b200-vllm-agentic:
       search-space:
       # Pure TP at low concurrency.
       - { tp: 8, kv-offloading: none, conc-list: [1, 2, 4, 6, 8] }
+      # Pure TP GPU-resident + MTP (num_speculative_tokens=3)
+      - { tp: 8, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 6, 8] }
       - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 12, 16] }
+      # Pure TP SimpleCPU + MTP (num_speculative_tokens=3)
+      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 12, 16] }
       # DEP
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
+      # DEP SimpleCPU + MTP (num_speculative_tokens=3)
+      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
+      # DEP Mooncake + MTP (num_speculative_tokens=3)
+      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, spec-decoding: mtp, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
 
 dsv4-fp4-b200-trt:
   image: ghcr.io#semianalysisai/trtllm-deepseek-v4:feat-deepseek_v4-c185066

From 83ab93ae5f94c25e7eca51ac3427f72c70b26bc0 Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 01:13:16 -0700
Subject: [PATCH 03/11] docs(changelog): add B200 vLLM MTP entry (PR #2259)

Document the dsv4-fp4-b200-vllm-agentic MTP twins (num_speculative_tokens=3).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 perf-changelog.yaml | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index c620aceaa5..f0c0c93f16 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4918,3 +4918,10 @@
     - "Image: vllm/vllm-openai:nightly-dev-x86_64-cu13.0.1-904e4ec"
     - "B300: GPU-resident TP4/TP8 at conc [1,2,4,6,8,12,16,20,24,28,32] and DEP8 at conc [32,64,96,128,160,192,196,224,228] (max-num-batched-tokens 16384, long-prefill-token-threshold 4096, gpu-memory-utilization 0.92); TP4 SimpleCPU lazy-offload at conc [28,32,36,40]; DEP4 at conc [8,16,24,32,40,48,56,64,72] with both SimpleCPU and Mooncake 0.3.11.post1."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2241
+
+- config-keys:
+    - dsv4-fp4-b200-vllm-agentic
+  description:
+    - "Add an MTP speculative-decoding twin for every B200 AgentX topology (DeepSeek-V4 built-in MTP): --speculative-config {\"method\":\"mtp\",\"num_speculative_tokens\":3}, routed via spec-decoding=mtp to dsv4_fp4_b200_vllm_mtp.sh -- TP8 GPU-resident, TP8 SimpleCPU, DEP8 SimpleCPU, DEP8 Mooncake, each mirroring its non-MTP conc-list."
+    - "MTP capture sizes are in TOKENS: --max-cudagraph-capture-size scaled to MAX_NUM_SEQS*(1+num_speculative_tokens) (= MAX_NUM_SEQS*4) so FULL_DECODE_ONLY covers the largest decode batches."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2259

From 026711b3a4b39faab947986225a073d592050fc8 Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 01:28:00 -0700
Subject: [PATCH 04/11] [AgentX] B200 vLLM: split MTP into a separate
 config-key

Restructure so the sweep runs only the new MTP work, not the existing aggregate:
- Revert dsv4-fp4-b200-vllm-agentic to its main (#2224) search space -- unchanged
  vs main, so it is not re-run.
- Add dsv4-fp4-b200-vllm-agentic-mtp: MTP twins (num_speculative_tokens=3) of the
  aggregate arms (TP8 GPU-resident, TP8 SimpleCPU, DEP8 SimpleCPU, DEP8 Mooncake),
  each mirroring its non-MTP conc-list.
- Point the perf-changelog entry at only the new key, so only it sweeps.

(No separate TP8 key: B200's TP8 arms already exist in the #2224 aggregate.)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 configs/nvidia-master.yaml | 21 +++++++++++++++++----
 perf-changelog.yaml        |  5 +++--
 2 files changed, 20 insertions(+), 6 deletions(-)

diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 268bb86721..2e7a0090b2 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -1784,16 +1784,29 @@ dsv4-fp4-b200-vllm-agentic:
       search-space:
       # Pure TP at low concurrency.
       - { tp: 8, kv-offloading: none, conc-list: [1, 2, 4, 6, 8] }
+      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 12, 16] }
+      # DEP
+      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
+      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
+
+dsv4-fp4-b200-vllm-agentic-mtp:
+  image: vllm/vllm-openai:nightly-dev-x86_64-cu13.0.1-904e4ec
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: cluster:b200-dgxc
+  precision: fp4
+  framework: vllm
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - dram-utilization: 0.80
+      search-space:
       # Pure TP GPU-resident + MTP (num_speculative_tokens=3)
       - { tp: 8, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 6, 8] }
-      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 12, 16] }
       # Pure TP SimpleCPU + MTP (num_speculative_tokens=3)
       - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 12, 16] }
-      # DEP
-      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
       # DEP SimpleCPU + MTP (num_speculative_tokens=3)
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
-      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
       # DEP Mooncake + MTP (num_speculative_tokens=3)
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, spec-decoding: mtp, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
 
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index f0c0c93f16..dc58819d1e 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4920,8 +4920,9 @@
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2241
 
 - config-keys:
-    - dsv4-fp4-b200-vllm-agentic
+    - dsv4-fp4-b200-vllm-agentic-mtp
   description:
-    - "Add an MTP speculative-decoding twin for every B200 AgentX topology (DeepSeek-V4 built-in MTP): --speculative-config {\"method\":\"mtp\",\"num_speculative_tokens\":3}, routed via spec-decoding=mtp to dsv4_fp4_b200_vllm_mtp.sh -- TP8 GPU-resident, TP8 SimpleCPU, DEP8 SimpleCPU, DEP8 Mooncake, each mirroring its non-MTP conc-list."
+    - "Add MTP speculative-decoding for the B200 vLLM AgentX recipe as a separate config-key, so the existing dsv4-fp4-b200-vllm-agentic aggregate is not re-run."
+    - "dsv4-fp4-b200-vllm-agentic-mtp: MTP twins (--speculative-config {\"method\":\"mtp\",\"num_speculative_tokens\":3}) of the aggregate arms -- TP8 GPU-resident, TP8 SimpleCPU, DEP8 SimpleCPU, DEP8 Mooncake, each mirroring its non-MTP conc-list; routed via spec-decoding=mtp to dsv4_fp4_b200_vllm_mtp.sh."
     - "MTP capture sizes are in TOKENS: --max-cudagraph-capture-size scaled to MAX_NUM_SEQS*(1+num_speculative_tokens) (= MAX_NUM_SEQS*4) so FULL_DECODE_ONLY covers the largest decode batches."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2259

From bebd5e63d81bf16c9334a0d5d1843a8cb40ac8b6 Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 13:41:16 -0700
Subject: [PATCH 05/11] align synthetic acceptance length to 2.49

Use MTP synthetic rejection sampling with acceptance length 2.49, the
dsv4-pro golden AL (thinking_on, num_speculative_tokens=3) from
golden_al_distribution/dsv4_mtp.yaml.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index 391a047a71..e645672766 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -3,11 +3,11 @@ set -eo pipefail
 set -x
 
 # Agentic trace replay benchmark for DeepSeek-V4-Pro FP4 on B200 using vLLM,
-# with MTP speculative decoding (num_speculative_tokens=3).
+# with MTP speculative decoding (num_speculative_tokens=3, synthetic acceptance length 2.49).
 #
 # Identical to dsv4_fp4_b200_vllm.sh (same image, engine args, offload, GPU
 # topologies, and agentic aiperf rig) with exactly two MTP deltas:
-#   --speculative-config '{"method": "mtp", "num_speculative_tokens": 3}'
+#   --speculative-config '{"method": "mtp", "num_speculative_tokens": 3, "rejection_sample_method": "synthetic", "synthetic_acceptance_length": 2.49}'
 #   --max-cudagraph-capture-size expressed in TOKENS (see the capture block below).
 #
 # Mirrors the fixed-seq-len parallelism options (pure TP and DEP) so the
@@ -233,6 +233,9 @@ MAX_NUM_SEQS=$((2 * CONC))
 # MAX_NUM_SEQS*(1+N) tokens -- otherwise vLLM's FULL_DECODE_ONLY ladder tops out
 # at MAX_NUM_SEQS/(1+N) seqs and the largest decode batches fall back to eager.
 NUM_SPEC_TOKENS=3
+# Standardize MTP acceptance to the dsv4-pro golden AL (thinking_on,
+# num_speculative_tokens=3) from golden_al_distribution/dsv4_mtp.yaml.
+SYNTHETIC_ACCEPT_LEN=2.49
 TOKENS_PER_SEQ=$((1 + NUM_SPEC_TOKENS))
 MAX_CUDAGRAPH_CAPTURE_SIZE=$((MAX_NUM_SEQS * TOKENS_PER_SEQ))
 
@@ -257,7 +260,7 @@ VLLM_CMD=(
     --tokenizer-mode deepseek_v4
     --reasoning-parser deepseek_v4
     --attention-config '{"backend":"FLASHINFER_MLA_SPARSE_DSV4","use_prefill_query_quantization":true,"use_fp4_indexer_cache":true}'
-    --speculative-config "{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS}"
+    --speculative-config "{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"rejection_sample_method\": \"synthetic\", \"synthetic_acceptance_length\": $SYNTHETIC_ACCEPT_LEN}"
     --no-disable-hybrid-kv-cache-manager
     --disable-uvicorn-access-log
     --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY","mode":0}'

From 5187947e21cce9c8260af77b2e804fc701b15aa8 Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Fri, 17 Jul 2026 14:44:34 -0700
Subject: [PATCH 06/11] reduce GPU memory utilization from 0.92 to 0.9

Lower gpu-memory-utilization for the B200 vLLM MTP recipe to 0.9 for extra
headroom.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index e645672766..93a3a22cd5 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -253,7 +253,7 @@ VLLM_CMD=(
     --kv-cache-dtype fp8
     --block-size 256
     --max-model-len 1048576
-    --gpu-memory-utilization 0.92
+    --gpu-memory-utilization 0.9
     --numa-bind
     --enable-cumem-allocator
     --no-enable-flashinfer-autotune

From a32d07f833d673508770bab5ff26c68fef55290e Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Sat, 18 Jul 2026 02:22:20 -0700
Subject: [PATCH 07/11] Revert "fix(changelog): accept single-node agentic eval
 rows in matrix validation"

This reverts commit 6efd246c81a3f129b16bb6d665bca1dc0e666fcd.
---
 utils/matrix_logic/validation.py | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/utils/matrix_logic/validation.py b/utils/matrix_logic/validation.py
index 7403c795e4..6b7c671289 100644
--- a/utils/matrix_logic/validation.py
+++ b/utils/matrix_logic/validation.py
@@ -291,10 +291,6 @@ class SingleNodeAgenticMatrixEntry(BaseModel):
     duration: int = Field(alias=Fields.DURATION.value)
     exp_name: str = Field(alias=Fields.EXP_NAME.value)
     scenario_type: str = Field(alias=Fields.SCENARIO_TYPE.value)
-    # Agentic eval rows (SWE-bench) carry run_eval/eval_only; benchmark rows omit
-    # them. Optional so both shapes validate and benchmark output is unchanged.
-    run_eval: Optional[bool] = Field(default=None, alias=Fields.RUN_EVAL.value)
-    eval_only: Optional[bool] = Field(default=None, alias=Fields.EVAL_ONLY.value)
 
     @model_validator(mode='after')
     def validate_kv_offload_fields(self):
@@ -893,7 +889,7 @@ class ChangelogMatrixEntry(BaseModel):
                       ] = Field(default_factory=dict)
     multi_node: dict[str, list[Union[MultiNodeMatrixEntry, MultiNodeAgenticMatrixEntry]]
                      ] = Field(default_factory=dict)
-    evals: list[Union[SingleNodeMatrixEntry, SingleNodeAgenticMatrixEntry]] = Field(default_factory=list)
+    evals: list[SingleNodeMatrixEntry] = Field(default_factory=list)
     multinode_evals: list[MultiNodeMatrixEntry] = Field(default_factory=list)
     changelog_metadata: ChangelogMetadata
 

From 425fd8656656ae137a7745f966a304f8470a39aa Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Sat, 18 Jul 2026 14:16:05 -0700
Subject: [PATCH 08/11] MTP synthetic acceptance for throughput, real
 verification for eval

Make the B200 MTP --speculative-config conditional on EVAL_ONLY: throughput
points pin synthetic acceptance length 2.49 (dsv4-pro golden AL), while the
auto-selected SWE-bench accuracy run (EVAL_ONLY=true) uses real MTP with normal
target verification. Synthetic acceptance bypasses verification and corrupts the
eval output (0.0000 score).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../agentic/dsv4_fp4_b200_vllm_mtp.sh          | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index 93a3a22cd5..0ee50e029f 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -3,11 +3,12 @@ set -eo pipefail
 set -x
 
 # Agentic trace replay benchmark for DeepSeek-V4-Pro FP4 on B200 using vLLM,
-# with MTP speculative decoding (num_speculative_tokens=3, synthetic acceptance length 2.49).
+# with MTP speculative decoding (num_speculative_tokens=3): synthetic acceptance
+# length 2.49 for throughput, real target verification for the EVAL_ONLY eval.
 #
 # Identical to dsv4_fp4_b200_vllm.sh (same image, engine args, offload, GPU
 # topologies, and agentic aiperf rig) with exactly two MTP deltas:
-#   --speculative-config '{"method": "mtp", "num_speculative_tokens": 3, "rejection_sample_method": "synthetic", "synthetic_acceptance_length": 2.49}'
+#   --speculative-config: synthetic acceptance length 2.49 (throughput) vs real MTP (EVAL_ONLY); see the SPEC_CONFIG block
 #   --max-cudagraph-capture-size expressed in TOKENS (see the capture block below).
 #
 # Mirrors the fixed-seq-len parallelism options (pure TP and DEP) so the
@@ -233,9 +234,16 @@ MAX_NUM_SEQS=$((2 * CONC))
 # MAX_NUM_SEQS*(1+N) tokens -- otherwise vLLM's FULL_DECODE_ONLY ladder tops out
 # at MAX_NUM_SEQS/(1+N) seqs and the largest decode batches fall back to eager.
 NUM_SPEC_TOKENS=3
-# Standardize MTP acceptance to the dsv4-pro golden AL (thinking_on,
-# num_speculative_tokens=3) from golden_al_distribution/dsv4_mtp.yaml.
+# Throughput pins synthetic MTP acceptance to the dsv4-pro golden AL (thinking_on,
+# num_speculative_tokens=3, golden_al_distribution/dsv4_mtp.yaml). The EVAL_ONLY
+# accuracy run uses real target verification instead -- synthetic acceptance
+# bypasses verification and corrupts the SWE-bench eval (0.0000 score).
 SYNTHETIC_ACCEPT_LEN=2.49
+if [ "${EVAL_ONLY:-false}" = "true" ]; then
+    SPEC_CONFIG="{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS}"
+else
+    SPEC_CONFIG="{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"rejection_sample_method\": \"synthetic\", \"synthetic_acceptance_length\": $SYNTHETIC_ACCEPT_LEN}"
+fi
 TOKENS_PER_SEQ=$((1 + NUM_SPEC_TOKENS))
 MAX_CUDAGRAPH_CAPTURE_SIZE=$((MAX_NUM_SEQS * TOKENS_PER_SEQ))
 
@@ -260,7 +268,7 @@ VLLM_CMD=(
     --tokenizer-mode deepseek_v4
     --reasoning-parser deepseek_v4
     --attention-config '{"backend":"FLASHINFER_MLA_SPARSE_DSV4","use_prefill_query_quantization":true,"use_fp4_indexer_cache":true}'
-    --speculative-config "{\"method\": \"mtp\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"rejection_sample_method\": \"synthetic\", \"synthetic_acceptance_length\": $SYNTHETIC_ACCEPT_LEN}"
+    --speculative-config "$SPEC_CONFIG"
     --no-disable-hybrid-kv-cache-manager
     --disable-uvicorn-access-log
     --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY","mode":0}'

From 4b5cb056f8117e29586b1dd188bbb14803964eec Mon Sep 17 00:00:00 2001
From: Jeff Ma <jeffjma@umich.edu>
Date: Sun, 19 Jul 2026 10:13:59 -0700
Subject: [PATCH 09/11] add missing tool call parser for swe eval

Signed-off-by: Jeff Ma <jeffjma@umich.edu>
---
 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index 0ee50e029f..d36bc99a5a 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -266,6 +266,8 @@ VLLM_CMD=(
     --enable-cumem-allocator
     --no-enable-flashinfer-autotune
     --tokenizer-mode deepseek_v4
+    --tool-call-parser deepseek_v4
+    --enable-auto-tool-choice
     --reasoning-parser deepseek_v4
     --attention-config '{"backend":"FLASHINFER_MLA_SPARSE_DSV4","use_prefill_query_quantization":true,"use_fp4_indexer_cache":true}'
     --speculative-config "$SPEC_CONFIG"

From fc1c58487d76159f15ca8e29fdb3752f0761613a Mon Sep 17 00:00:00 2001
From: Cam Quilici <cjquilici@gmail.com>
Date: Mon, 20 Jul 2026 15:31:36 -0500
Subject: [PATCH 10/11] fix(agentic): preserve B200 MTP PCP GPU count

---
 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh | 2 +-
 configs/nvidia-master.yaml                               | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index d36bc99a5a..44a8cf7bf7 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -49,7 +49,7 @@ if [ "$PCP_SIZE" -gt 1 ]; then
     VLLM_CP_ARGS+=(--prefill-context-parallel-size "$PCP_SIZE")
 fi
 
-GPU_COUNT=$TP
+GPU_COUNT="${GPU_COUNT:-$((TP * PCP_SIZE))}"
 if [[ ! "$GPU_COUNT" =~ ^[1-9][0-9]*$ ]]; then
     echo "Error: GPU_COUNT must be a positive integer, got '$GPU_COUNT'" >&2
     exit 1
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 37e7d0e9a6..a6d9a5872d 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -979,9 +979,9 @@ dsv4-fp4-b200-vllm-agentic-mtp:
       # Pure TP GPU-resident + MTP (num_speculative_tokens=3)
       - { tp: 8, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 6, 8] }
       # Pure TP SimpleCPU + MTP (num_speculative_tokens=3)
-      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 12, 16] }
+      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, spec-decoding: mtp, conc-list: [8, 12, 16] }
       # DEP SimpleCPU + MTP (num_speculative_tokens=3)
-      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple, version: "904e4ec" }, spec-decoding: mtp, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
+      - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, spec-decoding: mtp, conc-list: [8, 16, 24, 32, 40, 48, 56, 64, 68, 72, 80], router: { name: vllm-router, version: "0.1.14" } }
       # DEP Mooncake + MTP (num_speculative_tokens=3)
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, spec-decoding: mtp, conc-list: [12, 20, 28, 36, 44, 52, 60, 68, 76], router: { name: vllm-router, version: "0.1.14" } }
 

From 579c16f00a131c11fe68473fa4e87c99c07effb7 Mon Sep 17 00:00:00 2001
From: Cam Quilici <cjquilici@gmail.com>
Date: Mon, 20 Jul 2026 15:34:48 -0500
Subject: [PATCH 11/11] fix(agentic): derive B200 MTP GPU count from topology

---
 benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
index 44a8cf7bf7..85ffd64609 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.sh
@@ -49,7 +49,7 @@ if [ "$PCP_SIZE" -gt 1 ]; then
     VLLM_CP_ARGS+=(--prefill-context-parallel-size "$PCP_SIZE")
 fi
 
-GPU_COUNT="${GPU_COUNT:-$((TP * PCP_SIZE))}"
+GPU_COUNT=$((TP * PCP_SIZE))
 if [[ ! "$GPU_COUNT" =~ ^[1-9][0-9]*$ ]]; then
     echo "Error: GPU_COUNT must be a positive integer, got '$GPU_COUNT'" >&2
     exit 1