From b4fd077b70279324617f6c180dc11393f529ba06 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 14:34:29 -0400
Subject: [PATCH 1/9] feat: add Kimi-K3 MXFP4 B200 aggregated TP8xPP2
 Dynamo-vLLM agentic recipe
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (no expert
parallelism) for the agentic-coding trace replay. Dedicated bring-up image
vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1,
fastsafetensors load format, kimi_k3 tool-call/reasoning parsers. Model
pre-staged at /lustre/fsw/models/Kimi-K3; launch_b200-dgxc.sh gains the
kimik3/fp4 model-path mapping, the agentic recipe overlay, and the agentic
cache default_mounts used by the GB200/GB300 agentic paths.

中文：新增 Kimi-K3 MXFP4 B200 聚合式 TP8xPP2 Dynamo-vLLM 智能体编码基准测试配方
（2 节点 / 16 GPU，纯 TP，不启用专家并行(EP)）。使用专用 bring-up 镜像
vllm/vllm-openai:kimi-k3（VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1、
fastsafetensors 加载格式、kimi_k3 工具调用/推理解析器）。模型已预置于
/lustre/fsw/models/Kimi-K3；启动器 launch_b200-dgxc.sh 增加 kimik3/fp4
模型路径映射、智能体配方覆盖及智能体缓存挂载。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../agentic/agg-b200-tp8pp2-agentic.yaml      | 109 ++++++++++++++++++
 configs/nvidia-master.yaml                    |  43 +++++++
 perf-changelog.yaml                           |  11 ++
 runners/launch_b200-dgxc.sh                   |  28 +++++
 4 files changed, 191 insertions(+)
 create mode 100644 benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
new file mode 100644
index 0000000000..2ded17c2fd
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -0,0 +1,109 @@
+name: "kimik3-vllm-agg-b200-tp8pp2-agentic"
+
+# Kimi-K3 MXFP4 B200 AGGREGATED TP8 x PP2 agentic recipe (2 nodes / 16 GPUs).
+# The native MXFP4 checkpoint (2.8T total params, ~1.4TB of weights) does not
+# fit one 8xB200 node, so TP8 shards attention/dense (/8) and PP2 splits the
+# 93 layers (/2) across 16 GPUs. Plain TP (NOT TEP): expert parallelism is
+# deliberately off, so the 896 routed experts are TP-sharded inside each
+# pipeline stage. Node allocation = tp*pp/gpus_per_node = 8*2/8 = 2 nodes.
+# Aggregated (single worker, decode num-worker 0) — no P/D split, no NIXL.
+# VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION fuses the K3 LatentMoE tail path in
+# the kimi-k3 bring-up image.
+model:
+  path: "kimik3"
+  container: "vllm/vllm-openai:kimi-k3"
+  precision: "fp4"
+
+identity:
+  model:
+    repo: "moonshotai/Kimi-K3"
+  container:
+    image: "vllm/vllm-openai:kimi-k3"
+  frameworks:
+    dynamo: "1.3.0.dev20260618"
+
+dynamo:
+  install: true
+  wheel: "1.3.0.dev20260618"
+
+slurm:
+  time_limit: "8:00:00"
+
+health_check:
+  interval_seconds: 10
+  max_attempts: 1440
+
+resources:
+  gpu_type: "b200"
+  gpus_per_node: 8
+  agg_nodes: 2
+  agg_workers: 1
+  gpus_per_agg: 16
+
+infra:
+  etcd_nats_dedicated_node: false
+  nats_max_payload_mb: 32
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+  aggregated_environment:
+    VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION: "1"
+    VLLM_SERVER_DEV_MODE: "1"
+    # ~1.4TB of MXFP4 weights off shared Lustre: keep the engine-ready window
+    # generous, and let one long AgentX request hold a PP stage beyond vLLM's
+    # 300-second model-execution default.
+    VLLM_ENGINE_READY_TIMEOUT_S: "3600"
+    VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: "1800"
+    NCCL_CUMEM_ENABLE: "1"
+    TILELANG_CLEANUP_TEMP_FILES: "1"
+    UCX_MEMTYPE_CACHE: "n"
+    UCX_MEMTYPE_REG_WHOLE: "n"
+    UCX_NET_DEVICES: "mlx5_0:1,mlx5_1:1,mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_10:1,mlx5_11:1"
+    HF_HUB_CACHE: "/hf_hub_cache"
+    HUGGINGFACE_HUB_CACHE: "/hf_hub_cache"
+  vllm_config:
+    aggregated:
+      served-model-name: "moonshotai/Kimi-K3"
+      tensor-parallel-size: 8
+      pipeline-parallel-size: 2
+      trust-remote-code: true
+      load-format: fastsafetensors
+      moe-backend: auto
+      gpu-memory-utilization: 0.95
+      no-enable-flashinfer-autotune: true
+      enable-auto-tool-choice: true
+      tool-call-parser: kimi_k3
+      reasoning-parser: kimi_k3
+      # Full native 1M context: the AgentX trace carries multi-hundred-k-token
+      # requests, and K3's KDA layers keep per-token KV small (only the 24
+      # gated-MLA layers hold cache). Cap prefill chunks so a single long
+      # request cannot OOM a pipeline stage; let vLLM pick max-num-seqs.
+      max-model-len: 1048576
+      max-num-batched-tokens: 8192
+
+sbatch_directives:
+  segment: "1"
+
+srun_options:
+  container-remap-root: ""
+
+benchmark:
+  type: custom
+  aiperf_server_metrics: true
+  command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
+  env:
+    INFMAX_CONTAINER_WORKSPACE: "/infmax-workspace"
+    RESULT_DIR: "/logs/agentic"
+    PORT: "8000"
+    # Keep the aggregate worker in the multinode result schema so ingestion
+    # uses the zero decode-worker count instead of duplicating TP into P and D.
+    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"
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 36ae816809..bf1cb716d5 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -8325,3 +8325,46 @@ qwen3.5-fp8-gb200-dynamo-sglang-mtp:
           tp: 16
           ep: 16
           dp-attn: true
+
+# Kimi-K3 MXFP4 B200 aggregated vLLM via Dynamo (TP8 x PP2, 2 nodes / 16
+# GPUs), agentic bring-up. The native MXFP4 checkpoint (2.8T total params,
+# ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense
+# and PP2 splits layers. Plain TP (NOT TEP): ep 1, no expert parallelism —
+# the 896 routed experts are TP-sharded within each pipeline stage. Node
+# count = tp*pp/gpus_per_node = 8*2/8 = 2. Aggregated (prefill num-worker 1 +
+# decode num-worker 0, RECIPES.md section 5) — the single worker serves both
+# phases, so no P/D KV transfer. Dedicated kimi-k3 vLLM bring-up image with
+# VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 and the kimi_k3 tool-call/reasoning
+# parsers.
+# Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+kimik3-fp4-b200-dynamo-vllm-agentic:
+  image: vllm/vllm-openai:kimi-k3
+  model: moonshotai/Kimi-K3
+  model-prefix: kimik3
+  runner: cluster:b200-dgxc
+  precision: fp4
+  framework: dynamo-vllm
+  router: { name: dynamo-router, version: "1.3.0.dev20260618" }
+  multinode: true
+  disagg: false
+  scenarios:
+    agentic-coding:
+    - search-space:
+      - spec-decoding: none
+        conc-list: [8, 16, 32]
+        prefill:
+          num-worker: 1
+          tp: 8
+          pp: 2
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml"
+        # The aggregate worker also performs decode; keep the decode worker
+        # count at zero so result aggregation counts the 16 GPUs only once.
+        decode:
+          num-worker: 0
+          tp: 8
+          pp: 2
+          ep: 1
+          dp-attn: false
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 0524d0c0c9..0a4e40426b 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5116,3 +5116,14 @@
   description:
     - "Bump image from lmsysorg/sglang:v0.5.14-rocm720-mi35x to lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260726"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2349
+
+- config-keys:
+    - kimik3-fp4-b200-dynamo-vllm-agentic
+  description:
+    - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
+    - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
+    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --enable-auto-tool-choice, --tool-call-parser kimi_k3, --reasoning-parser kimi_k3, --trust-remote-code"
+    - "Full native 1M max-model-len for the AgentX trace (K3's KDA layers keep per-token KV small; only the 24 gated-MLA layers hold cache), max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; conc 8/16/32"
+    - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, the agentic recipe overlay into the srt-slurm-nv clone, and the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
+    - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX
diff --git a/runners/launch_b200-dgxc.sh b/runners/launch_b200-dgxc.sh
index a276644575..e166dbb595 100644
--- a/runners/launch_b200-dgxc.sh
+++ b/runners/launch_b200-dgxc.sh
@@ -72,6 +72,10 @@ elif [[ $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp4" ]]; then
     # NVFP4 checkpoint, pre-staged on the b200-dgxc scratch tree.
     export MODEL_PATH="/scratch/fsw/models/MiniMax-M3-NVFP4"
     export SRT_SLURM_MODEL_PREFIX="minimax-m3-nvfp4"
+elif [[ $MODEL_PREFIX == "kimik3" && $PRECISION == "fp4" ]]; then
+    # Native MXFP4 checkpoint, pre-staged on the SRE-managed Lustre tree.
+    export MODEL_PATH="/lustre/fsw/models/Kimi-K3"
+    export SRT_SLURM_MODEL_PREFIX="kimik3"
 else
     echo "Unsupported model prefix/precision: $MODEL_PREFIX/$PRECISION"
     echo "Available models under /lustre/fsw/models:"
@@ -107,6 +111,13 @@ if [[ "$IS_MULTINODE" == "true" ]]; then
     if [[ "$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
+        # Overlay InferenceX-staged agentic recipes onto the clone (cp -rT so
+        # an upstream stub directory is merged rather than nested).
+        if [[ $MODEL_PREFIX == "kimik3" ]]; then
+            mkdir -p recipes/vllm/kimi-k3/agentic
+            cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic" \
+                recipes/vllm/kimi-k3/agentic
+        fi
     elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "dsv4" ]]; then
         git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
         cd "$SRT_REPO_DIR" || exit 1
@@ -207,6 +218,22 @@ if [[ "$IS_MULTINODE" == "true" ]]; then
     export OSL="$OSL"
     export EVAL_ONLY="${EVAL_ONLY:-false}"
 
+    # Agentic runs bind-mount two persistent caches into every worker
+    # container (Lustre, shared across nodes): aiperf's content-addressed
+    # dataset mmap cache and the HF hub cache holding the trace dataset
+    # download. The container-side paths are referenced by the agentic
+    # recipes' benchmark.env (AIPERF_DATASET_MMAP_CACHE_DIR=/aiperf_mmap_cache,
+    # HF_HUB_CACHE=/hf_hub_cache).
+    DEFAULT_MOUNTS_BLOCK=""
+    if [[ "$IS_AGENTIC" == "1" ]]; then
+        HF_HUB_CACHE_HOST_PATH="/lustre/fsw/gharunners/hf-hub-cache"
+        mkdir -p "$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="default_mounts:
+  ${AIPERF_MMAP_CACHE_HOST_PATH}: /aiperf_mmap_cache
+  ${HF_HUB_CACHE_HOST_PATH}: /hf_hub_cache"
+    fi
+
     # Create srtslurm.yaml for srtctl (used by both frameworks)
     SRTCTL_ROOT="${GITHUB_WORKSPACE}/${SRT_REPO_DIR}"
     echo "Creating srtslurm.yaml configuration..."
@@ -234,6 +261,7 @@ containers:
   "${IMAGE}": "${SQUASH_FILE}"
   nginx-sqsh: "${NGINX_SQUASH_FILE}"
 use_exclusive_sbatch_directive: true
+${DEFAULT_MOUNTS_BLOCK}
 EOF
 
     echo "Generated srtslurm.yaml:"

From 4dbbdc8a5908a6d6177674be46a226b721be80f3 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 14:35:35 -0400
Subject: [PATCH 2/9] docs: link PR #2355 in changelog entry and MODELS rows
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

中文：在更新日志条目与 MODELS 表格行中补充 PR #2355 链接。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 MODELS.md           | 2 +-
 MODELS_zh.md        | 2 +-
 perf-changelog.yaml | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/MODELS.md b/MODELS.md
index 3fc822420e..31cb9df252 100644
--- a/MODELS.md
+++ b/MODELS.md
@@ -18,7 +18,7 @@ This document tracks every model benchmarked by InferenceX-e2e: when it was adde
 | Model architecture class | Prefix | Date added | Active scenarios | Deprecated scenarios |
 |---|---|---|---|---|
 | Qwen3.8 2.4T | `qwen3.8` | TBD | Agentic coding | |
-| Kimi-K3 | `kimik3` | 2026-07-27 | Agentic coding | |
+| Kimi-K3 | `kimik3` | 2026-07-27 ([#2355](https://github.com/SemiAnalysisAI/InferenceX/pull/2355)) | Agentic coding | |
 | GLM-5.2 | `glm5.2` | 2026-07-18 ([#2268](https://github.com/SemiAnalysisAI/InferenceX/pull/2268)) | Agentic coding | |
 | MiniMax-M3 | `minimaxm3` | 2026-06-12 ([#1724](https://github.com/SemiAnalysisAI/InferenceX/pull/1724)) | Single-turn 8k1k, Agentic coding | Single-turn 1k1k |
 | DeepSeek-V4-Pro | `dsv4` | 2026-04-24 ([#1130](https://github.com/SemiAnalysisAI/InferenceX/pull/1130)) | Single-turn 8k1k, Agentic coding | Single-turn 1k1k |
diff --git a/MODELS_zh.md b/MODELS_zh.md
index 177a4b4ba3..e71067c13b 100644
--- a/MODELS_zh.md
+++ b/MODELS_zh.md
@@ -18,7 +18,7 @@
 | 模型架构类别 | 前缀 | 加入日期 | 启用场景 | 已弃用场景 |
 |---|---|---|---|---|
 | Qwen3.8 2.4T | `qwen3.8` | 待定 | 智能体编码 | |
-| Kimi-K3 | `kimik3` | 2026-07-27 | 智能体编码 | |
+| Kimi-K3 | `kimik3` | 2026-07-27 ([#2355](https://github.com/SemiAnalysisAI/InferenceX/pull/2355)) | 智能体编码 | |
 | GLM-5.2 | `glm5.2` | 2026-07-18（[#2268](https://github.com/SemiAnalysisAI/InferenceX/pull/2268)） | 智能体编码 | |
 | MiniMax-M3 | `minimaxm3` | 2026-06-12（[#1724](https://github.com/SemiAnalysisAI/InferenceX/pull/1724)） | 单轮 8k1k、智能体编码 | 单轮 1k1k |
 | DeepSeek-V4-Pro | `dsv4` | 2026-04-24（[#1130](https://github.com/SemiAnalysisAI/InferenceX/pull/1130)） | 单轮 8k1k、智能体编码 | 单轮 1k1k |
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 0a4e40426b..5a7800bd48 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5126,4 +5126,4 @@
     - "Full native 1M max-model-len for the AgentX trace (K3's KDA layers keep per-token KV small; only the 24 gated-MLA layers hold cache), max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; conc 8/16/32"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, the agentic recipe overlay into the srt-slurm-nv clone, and the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
-  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2355

From e8d42a73cdc7126fae0a3c1a30b8e54132dff1f1 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 14:48:01 -0400
Subject: [PATCH 3/9] fix: pin agentic srt-slurm to NVIDIA v1.0.36, dynamo
 1.2.1, conc-8 smoke test
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The cquil11/srt-slurm-nv cam/sa-submission-q2-2026 fork rejected the recipe
(benchmark.aiperf_server_metrics: Unknown field). Switch the b200-dgxc
agentic clone to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341),
drop the aiperf_server_metrics field, pin dynamo wheel/router to 1.2.1 (the
combination validated with v1.0.36), and reduce the bring-up to a single
conc-8 smoke test.

中文：cquil11/srt-slurm-nv 分支的 srtctl 校验拒绝了配方字段
benchmark.aiperf_server_metrics（Unknown field）。将 b200-dgxc 智能体路径改用
上游 NVIDIA/srt-slurm v1.0.36（已在 #2302/#2341 验证），移除该字段，dynamo
wheel/router 固定为 1.2.1，并将 bring-up 缩减为单并发（conc 8）冒烟测试。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml    |  5 ++---
 configs/nvidia-master.yaml                          |  6 ++++--
 perf-changelog.yaml                                 |  5 +++--
 runners/launch_b200-dgxc.sh                         | 13 ++++++++++---
 4 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index 2ded17c2fd..cbee00a2b5 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -20,11 +20,11 @@ identity:
   container:
     image: "vllm/vllm-openai:kimi-k3"
   frameworks:
-    dynamo: "1.3.0.dev20260618"
+    dynamo: "1.2.1"
 
 dynamo:
+  wheel: "1.2.1"
   install: true
-  wheel: "1.3.0.dev20260618"
 
 slurm:
   time_limit: "8:00:00"
@@ -94,7 +94,6 @@ srun_options:
 
 benchmark:
   type: custom
-  aiperf_server_metrics: true
   command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh
   env:
     INFMAX_CONTAINER_WORKSPACE: "/infmax-workspace"
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index bf1cb716d5..82c69638c8 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -8344,14 +8344,16 @@ kimik3-fp4-b200-dynamo-vllm-agentic:
   runner: cluster:b200-dgxc
   precision: fp4
   framework: dynamo-vllm
-  router: { name: dynamo-router, version: "1.3.0.dev20260618" }
+  router: { name: dynamo-router, version: "1.2.1" }
   multinode: true
   disagg: false
   scenarios:
     agentic-coding:
     - search-space:
+      # Single-concurrency smoke test for the bring-up; widen the conc curve
+      # once the topology is proven green.
       - spec-decoding: none
-        conc-list: [8, 16, 32]
+        conc-list: [8]
         prefill:
           num-worker: 1
           tp: 8
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 5a7800bd48..b8b11fa508 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5123,7 +5123,8 @@
     - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
     - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --enable-auto-tool-choice, --tool-call-parser kimi_k3, --reasoning-parser kimi_k3, --trust-remote-code"
-    - "Full native 1M max-model-len for the AgentX trace (K3's KDA layers keep per-token KV small; only the 24 gated-MLA layers hold cache), max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; conc 8/16/32"
-    - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, the agentic recipe overlay into the srt-slurm-nv clone, and the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
+    - "Full native 1M max-model-len for the AgentX trace (K3's KDA layers keep per-token KV small; only the 24 gated-MLA layers hold cache), max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
+    - "Dynamo 1.2.1 (ai-dynamo wheel + dynamo-router), the version validated with srt-slurm v1.0.36 in #2341's GB300 aggregate AgentX recipes"
+    - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2355
diff --git a/runners/launch_b200-dgxc.sh b/runners/launch_b200-dgxc.sh
index e166dbb595..1dd4e19d14 100644
--- a/runners/launch_b200-dgxc.sh
+++ b/runners/launch_b200-dgxc.sh
@@ -109,14 +109,21 @@ if [[ "$IS_MULTINODE" == "true" ]]; then
 
     # TODO(CJQ): make first class upon srt-slurm upstream refactor
     if [[ "$IS_AGENTIC" == "1" ]]; then
-        git clone --branch cam/sa-submission-q2-2026 --single-branch https://github.com/cquil11/srt-slurm-nv.git "$SRT_REPO_DIR"
+        # Agentic recipes use NVIDIA/srt-slurm v1.0.36, the upstream version
+        # validated in InferenceX PR #2302/#2341 for the vLLM agentic path
+        # (BenchmarkType.CUSTOM + benchmark.command/env, DynamoConfig.wheel,
+        # srun_options propagation, per-node DP, matching Dynamo health
+        # counts). Keep it pinned so sweeps are reproducible. Note the older
+        # cquil11/srt-slurm-nv cam/sa-submission-q2-2026 fork previously
+        # cloned here rejects newer recipe schema fields.
+        git clone --branch v1.0.36 --single-branch https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR" || exit 1
         cd "$SRT_REPO_DIR" || exit 1
         # Overlay InferenceX-staged agentic recipes onto the clone (cp -rT so
         # an upstream stub directory is merged rather than nested).
         if [[ $MODEL_PREFIX == "kimik3" ]]; then
-            mkdir -p recipes/vllm/kimi-k3/agentic
+            mkdir -p recipes/vllm/kimi-k3/agentic || exit 1
             cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic" \
-                recipes/vllm/kimi-k3/agentic
+                recipes/vllm/kimi-k3/agentic || exit 1
         fi
     elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "dsv4" ]]; then
         git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"

From c1e2a568101d37556da96f9fd80c262f30d2dabb Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 14:54:05 -0400
Subject: [PATCH 4/9] fix: drop OpenAI-frontend tool-choice flags from
 dynamo-vllm worker args
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The dynamo-vllm worker entrypoint rejected --enable-auto-tool-choice
--tool-call-parser kimi_k3 (unrecognized arguments; different arg parser
than vllm serve). Chat parsing happens at the dynamo frontend — same
convention as the DSv4 GB300 agentic recipes. Keep --reasoning-parser
kimi_k3 (accepted by the worker). Also drop the explicit max-model-len and
let vLLM derive the native 1M window from the model config, mirroring the
agentic recipe convention.

中文：dynamo-vllm worker 入口不接受 --enable-auto-tool-choice 与
--tool-call-parser kimi_k3（unrecognized arguments，与 vllm serve 的参数解析器
不同），聊天解析由 dynamo 前端处理，与 DSv4 GB300 智能体配方约定一致；保留
worker 可接受的 --reasoning-parser kimi_k3。同时移除显式 max-model-len，
由 vLLM 从模型配置推导原生 1M 上下文窗口。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../agentic/agg-b200-tp8pp2-agentic.yaml      | 19 ++++++++++++-------
 perf-changelog.yaml                           |  4 ++--
 2 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index cbee00a2b5..dcb1ec7afa 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -76,14 +76,19 @@ backend:
       moe-backend: auto
       gpu-memory-utilization: 0.95
       no-enable-flashinfer-autotune: true
-      enable-auto-tool-choice: true
-      tool-call-parser: kimi_k3
+      # Note: --enable-auto-tool-choice / --tool-call-parser are NOT set on
+      # the worker. The dynamo-vllm worker entrypoint doesn't accept them
+      # (different arg parser than `vllm serve`; first sweep attempt died with
+      # "unrecognized arguments"). Chat parsing happens at the dynamo
+      # frontend, not at the worker — same convention as the DSv4 GB300
+      # agentic recipes. --reasoning-parser IS accepted by the worker.
       reasoning-parser: kimi_k3
-      # Full native 1M context: the AgentX trace carries multi-hundred-k-token
-      # requests, and K3's KDA layers keep per-token KV small (only the 24
-      # gated-MLA layers hold cache). Cap prefill chunks so a single long
-      # request cannot OOM a pipeline stage; let vLLM pick max-num-seqs.
-      max-model-len: 1048576
+      # No explicit max-model-len: let vLLM derive the native 1M window from
+      # the model config (agentic trajectories blow past any small cap, and
+      # K3's KDA layers keep per-token KV small — only the 24 gated-MLA
+      # layers hold cache). Prefix caching stays on (default) for trajectory
+      # reuse. Cap prefill chunks so a single long request cannot OOM a
+      # pipeline stage; let vLLM pick max-num-seqs.
       max-num-batched-tokens: 8192
 
 sbatch_directives:
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index b8b11fa508..61b86f3db3 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5122,8 +5122,8 @@
   description:
     - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
-    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --enable-auto-tool-choice, --tool-call-parser kimi_k3, --reasoning-parser kimi_k3, --trust-remote-code"
-    - "Full native 1M max-model-len for the AgentX trace (K3's KDA layers keep per-token KV small; only the 24 gated-MLA layers hold cache), max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
+    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --reasoning-parser kimi_k3, --trust-remote-code. --enable-auto-tool-choice/--tool-call-parser are NOT set: the dynamo-vllm worker entrypoint rejects them (unrecognized arguments; different arg parser than vllm serve) — chat parsing happens at the dynamo frontend, same convention as the DSv4 GB300 agentic recipes"
+    - "No explicit max-model-len (vLLM derives the native 1M window from the model config; K3's KDA layers keep per-token KV small — only the 24 gated-MLA layers hold cache), prefix caching on for trajectory reuse, max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
     - "Dynamo 1.2.1 (ai-dynamo wheel + dynamo-router), the version validated with srt-slurm v1.0.36 in #2341's GB300 aggregate AgentX recipes"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"

From ef35fd1f030fd821fa31436d9a26f425e77aaeec Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 15:20:42 -0400
Subject: [PATCH 5/9] fix: try dynamo wheel 1.2.0.dev20260426 for Kimi-K3
 frontend tokenizer
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Third sweep attempt: the engine loaded and served (TP8xPP2 healthy in ~14
min), but dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken
model_type 'kimi_k3' (supported: kimi, kimi_k2, kimi_k25, deepseek_v3), so
the model never registered and all chat completions returned 404, aborting
the AgentX warmup. Switch to the 1.2.0.dev20260426 wheel used by the DSv4
GB300/B200 Dynamo-vLLM recipes. Upstream published v1.4.0-kimi-k3-dev.1
(2026-07-27) as the day-zero K3 build if this wheel also lacks support.

中文：第三次扫描中引擎已成功加载并提供服务（TP8xPP2 约 14 分钟就绪），但
dynamo 1.2.1 的 rust 前端分词器不支持 Kimi-K3 的 tiktoken model_type
'kimi_k3'，模型未能注册，所有请求返回 404，AgentX 预热中止。改用 DSv4
GB300/B200 Dynamo-vLLM 配方所用的 1.2.0.dev20260426 wheel；如仍不支持，
上游已于 2026-07-27 发布 day-zero 构建 v1.4.0-kimi-k3-dev.1。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml         | 4 ++--
 configs/nvidia-master.yaml                                    | 2 +-
 perf-changelog.yaml                                           | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index dcb1ec7afa..d36158073f 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -20,11 +20,11 @@ identity:
   container:
     image: "vllm/vllm-openai:kimi-k3"
   frameworks:
-    dynamo: "1.2.1"
+    dynamo: "1.2.0.dev20260426"
 
 dynamo:
-  wheel: "1.2.1"
   install: true
+  wheel: "1.2.0.dev20260426"
 
 slurm:
   time_limit: "8:00:00"
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 82c69638c8..3dd5261b27 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -8344,7 +8344,7 @@ kimik3-fp4-b200-dynamo-vllm-agentic:
   runner: cluster:b200-dgxc
   precision: fp4
   framework: dynamo-vllm
-  router: { name: dynamo-router, version: "1.2.1" }
+  router: { name: dynamo-router, version: "1.2.0.dev20260426" }
   multinode: true
   disagg: false
   scenarios:
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 61b86f3db3..2001751894 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5124,7 +5124,7 @@
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
     - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --reasoning-parser kimi_k3, --trust-remote-code. --enable-auto-tool-choice/--tool-call-parser are NOT set: the dynamo-vllm worker entrypoint rejects them (unrecognized arguments; different arg parser than vllm serve) — chat parsing happens at the dynamo frontend, same convention as the DSv4 GB300 agentic recipes"
     - "No explicit max-model-len (vLLM derives the native 1M window from the model config; K3's KDA layers keep per-token KV small — only the 24 gated-MLA layers hold cache), prefix caching on for trajectory reuse, max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
-    - "Dynamo 1.2.1 (ai-dynamo wheel + dynamo-router), the version validated with srt-slurm v1.0.36 in #2341's GB300 aggregate AgentX recipes"
+    - "Dynamo 1.2.0.dev20260426 (ai-dynamo wheel + dynamo-router), the wheel used by the DSv4 GB300/B200 Dynamo-vLLM recipes. Note: dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registers and all chat completions 404 (third sweep attempt); upstream published v1.4.0-kimi-k3-dev.1 on 2026-07-27 as the day-zero K3 build if this wheel also lacks kimi_k3 tiktoken support"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2355

From be6c56e6a1a6717d3fc018570e21471f496c7138 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 15:26:38 -0400
Subject: [PATCH 6/9] fix: pin dynamo to day-zero Kimi-K3 commit, restore
 kimi_k3 parser flags
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Pin dynamo to ba83080ecd31c1ce918559e576d3c5bc9e092ff1 ("feat: Added
support for Kimi-K3", tag v1.4.0-kimi-k3-dev.1) via srt-slurm's
hash-cached source install: it adds the kimi_k3 tiktoken tokenizer to the
rust frontend (dynamo <=1.2.1 404s every request because the model never
registers) and accepts the kimi_k3 tool-call/reasoning parser worker args,
so restore --enable-auto-tool-choice --tool-call-parser kimi_k3
--reasoning-parser kimi_k3.

中文：将 dynamo 固定到 day-zero Kimi-K3 提交 ba83080（"feat: Added support
for Kimi-K3"，标签 v1.4.0-kimi-k3-dev.1），通过 srt-slurm 的哈希缓存源码
安装：该提交为 rust 前端新增 kimi_k3 tiktoken 分词器（dynamo <=1.2.1 因模型
无法注册而全部返回 404），worker 亦支持 kimi_k3 解析器参数，故恢复
--enable-auto-tool-choice --tool-call-parser kimi_k3 --reasoning-parser kimi_k3。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../agentic/agg-b200-tp8pp2-agentic.yaml      | 21 ++++++++++++-------
 configs/nvidia-master.yaml                    |  2 +-
 perf-changelog.yaml                           |  4 ++--
 3 files changed, 16 insertions(+), 11 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index d36158073f..22362d730c 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -20,11 +20,16 @@ identity:
   container:
     image: "vllm/vllm-openai:kimi-k3"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
+    dynamo: "1.4.0-kimi-k3-dev.1"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
+  # Day-zero Kimi-K3 dynamo ("feat: Added support for Kimi-K3", tag
+  # v1.4.0-kimi-k3-dev.1 == this commit): adds the kimi_k3 tiktoken tokenizer
+  # to the rust frontend (dynamo <=1.2.1 only knows kimi/kimi_k2/kimi_k25/
+  # deepseek_v3, so the model never registers and every request 404s) and the
+  # kimi_k3 tool-call/reasoning parser worker args.
+  hash: "ba83080ecd31c1ce918559e576d3c5bc9e092ff1"
 
 slurm:
   time_limit: "8:00:00"
@@ -76,12 +81,12 @@ backend:
       moe-backend: auto
       gpu-memory-utilization: 0.95
       no-enable-flashinfer-autotune: true
-      # Note: --enable-auto-tool-choice / --tool-call-parser are NOT set on
-      # the worker. The dynamo-vllm worker entrypoint doesn't accept them
-      # (different arg parser than `vllm serve`; first sweep attempt died with
-      # "unrecognized arguments"). Chat parsing happens at the dynamo
-      # frontend, not at the worker — same convention as the DSv4 GB300
-      # agentic recipes. --reasoning-parser IS accepted by the worker.
+      # kimi_k3 parsers: the day-zero K3 dynamo build pinned above accepts
+      # these on the worker. Dynamo <=1.2.1 rejected --enable-auto-tool-choice
+      # / --tool-call-parser as unrecognized arguments (different arg parser
+      # than `vllm serve`).
+      enable-auto-tool-choice: true
+      tool-call-parser: kimi_k3
       reasoning-parser: kimi_k3
       # No explicit max-model-len: let vLLM derive the native 1M window from
       # the model config (agentic trajectories blow past any small cap, and
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 3dd5261b27..89bf2fad2b 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -8344,7 +8344,7 @@ kimik3-fp4-b200-dynamo-vllm-agentic:
   runner: cluster:b200-dgxc
   precision: fp4
   framework: dynamo-vllm
-  router: { name: dynamo-router, version: "1.2.0.dev20260426" }
+  router: { name: dynamo-router, version: "1.4.0-kimi-k3-dev.1" }
   multinode: true
   disagg: false
   scenarios:
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 2001751894..c687f6da00 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5122,9 +5122,9 @@
   description:
     - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
-    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --reasoning-parser kimi_k3, --trust-remote-code. --enable-auto-tool-choice/--tool-call-parser are NOT set: the dynamo-vllm worker entrypoint rejects them (unrecognized arguments; different arg parser than vllm serve) — chat parsing happens at the dynamo frontend, same convention as the DSv4 GB300 agentic recipes"
+    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --enable-auto-tool-choice, --tool-call-parser kimi_k3, --reasoning-parser kimi_k3, --trust-remote-code (the kimi_k3 parser worker args require the day-zero K3 dynamo below; dynamo <=1.2.1 rejects the tool-choice flags as unrecognized arguments)"
     - "No explicit max-model-len (vLLM derives the native 1M window from the model config; K3's KDA layers keep per-token KV small — only the 24 gated-MLA layers hold cache), prefix caching on for trajectory reuse, max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
-    - "Dynamo 1.2.0.dev20260426 (ai-dynamo wheel + dynamo-router), the wheel used by the DSv4 GB300/B200 Dynamo-vLLM recipes. Note: dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registers and all chat completions 404 (third sweep attempt); upstream published v1.4.0-kimi-k3-dev.1 on 2026-07-27 as the day-zero K3 build if this wheel also lacks kimi_k3 tiktoken support"
+    - "Dynamo hash-pinned to ba83080ecd31c1ce918559e576d3c5bc9e092ff1 ('feat: Added support for Kimi-K3', tag v1.4.0-kimi-k3-dev.1, published 2026-07-27) via srt-slurm's hash-cached source install. Required because dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registered with the frontend and every chat completion 404'd, aborting the AgentX warmup (third sweep attempt); the TP8xPP2 engine itself loaded and served (health-ready in ~14 min via fastsafetensors)"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2355

From c6917e640fb34327cd684189f5741ae475ada231 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 15:32:17 -0400
Subject: [PATCH 7/9] fix: use dynamo namespaced --dyn-* kimi_k3 parser args on
 the worker
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Replace the vLLM OpenAI-frontend spellings (--enable-auto-tool-choice /
--tool-call-parser) with dynamo's namespaced worker args:
--dyn-tool-call-parser kimi_k3 --reasoning-parser kimi_k3
--dyn-reasoning-parser kimi_k3.

中文：将 vLLM OpenAI 前端风格参数（--enable-auto-tool-choice /
--tool-call-parser）替换为 dynamo 命名空间的 worker 参数：
--dyn-tool-call-parser kimi_k3 --reasoning-parser kimi_k3
--dyn-reasoning-parser kimi_k3。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml    | 13 +++++++------
 perf-changelog.yaml                                 |  2 +-
 2 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index 22362d730c..ab4c5314b0 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -81,13 +81,14 @@ backend:
       moe-backend: auto
       gpu-memory-utilization: 0.95
       no-enable-flashinfer-autotune: true
-      # kimi_k3 parsers: the day-zero K3 dynamo build pinned above accepts
-      # these on the worker. Dynamo <=1.2.1 rejected --enable-auto-tool-choice
-      # / --tool-call-parser as unrecognized arguments (different arg parser
-      # than `vllm serve`).
-      enable-auto-tool-choice: true
-      tool-call-parser: kimi_k3
+      # kimi_k3 parsers via dynamo's namespaced worker args (--dyn-*), which
+      # the day-zero K3 dynamo build pinned above provides. The vLLM
+      # OpenAI-frontend spellings (--enable-auto-tool-choice /
+      # --tool-call-parser) are rejected by the dynamo worker entrypoint as
+      # unrecognized arguments (different arg parser than `vllm serve`).
+      dyn-tool-call-parser: kimi_k3
       reasoning-parser: kimi_k3
+      dyn-reasoning-parser: kimi_k3
       # No explicit max-model-len: let vLLM derive the native 1M window from
       # the model config (agentic trajectories blow past any small cap, and
       # K3's KDA layers keep per-token KV small — only the 24 gated-MLA
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index c687f6da00..43fd4df391 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5122,7 +5122,7 @@
   description:
     - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
-    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --enable-auto-tool-choice, --tool-call-parser kimi_k3, --reasoning-parser kimi_k3, --trust-remote-code (the kimi_k3 parser worker args require the day-zero K3 dynamo below; dynamo <=1.2.1 rejects the tool-choice flags as unrecognized arguments)"
+    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --trust-remote-code, and kimi_k3 parsers via dynamo's namespaced worker args --dyn-tool-call-parser kimi_k3 --reasoning-parser kimi_k3 --dyn-reasoning-parser kimi_k3 (the vLLM OpenAI-frontend spellings --enable-auto-tool-choice/--tool-call-parser are rejected by the dynamo worker entrypoint as unrecognized arguments)"
     - "No explicit max-model-len (vLLM derives the native 1M window from the model config; K3's KDA layers keep per-token KV small — only the 24 gated-MLA layers hold cache), prefix caching on for trajectory reuse, max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
     - "Dynamo hash-pinned to ba83080ecd31c1ce918559e576d3c5bc9e092ff1 ('feat: Added support for Kimi-K3', tag v1.4.0-kimi-k3-dev.1, published 2026-07-27) via srt-slurm's hash-cached source install. Required because dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registered with the frontend and every chat completion 404'd, aborting the AgentX warmup (third sweep attempt); the TP8xPP2 engine itself loaded and served (health-ready in ~14 min via fastsafetensors)"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"

From 288349f0b49fbf72a7c2abc904c86c345a240cc8 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 15:44:16 -0400
Subject: [PATCH 8/9] =?UTF-8?q?feat:=20parser-flag=20experiment=20B=20?=
 =?UTF-8?q?=E2=80=94=20plain=20vLLM=20tool-call-parser=20spellings?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Variant of the Kimi-K3 B200 agg TP8xPP2 agentic bring-up (#2355) using
--tool-call-parser kimi_k3 --reasoning-parser kimi_k3 on the worker.
Sibling experiments try dynamo's --dyn-* namespaced args and no parser
flags; whichever goes green first merges.

中文：Kimi-K3 B200 聚合式 TP8xPP2 智能体 bring-up（#2355）的解析器参数实验
变体 B：worker 使用 --tool-call-parser kimi_k3 --reasoning-parser kimi_k3。
兄弟实验分别尝试 dynamo --dyn-* 命名空间参数与不带解析器参数；先通过者合并。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml     | 12 +++++-------
 perf-changelog.yaml                                  |  2 +-
 2 files changed, 6 insertions(+), 8 deletions(-)

diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
index ab4c5314b0..7109b14097 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml
@@ -81,14 +81,12 @@ backend:
       moe-backend: auto
       gpu-memory-utilization: 0.95
       no-enable-flashinfer-autotune: true
-      # kimi_k3 parsers via dynamo's namespaced worker args (--dyn-*), which
-      # the day-zero K3 dynamo build pinned above provides. The vLLM
-      # OpenAI-frontend spellings (--enable-auto-tool-choice /
-      # --tool-call-parser) are rejected by the dynamo worker entrypoint as
-      # unrecognized arguments (different arg parser than `vllm serve`).
-      dyn-tool-call-parser: kimi_k3
+      # kimi_k3 parsers via the plain vLLM spellings (parser-flag experiment
+      # variant B — sibling PRs try the --dyn-* namespaced args and no parser
+      # flags). Note dynamo <=1.2.1's worker rejected --tool-call-parser as
+      # unrecognized; the day-zero K3 dynamo pinned above may accept it.
+      tool-call-parser: kimi_k3
       reasoning-parser: kimi_k3
-      dyn-reasoning-parser: kimi_k3
       # No explicit max-model-len: let vLLM derive the native 1M window from
       # the model config (agentic trajectories blow past any small cap, and
       # K3's KDA layers keep per-token KV small — only the 24 gated-MLA
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 43fd4df391..06341f4bf1 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5122,7 +5122,7 @@
   description:
     - "Add Kimi-K3 MXFP4 B200 aggregated multinode Dynamo-vLLM agentic-coding bring-up (new model on B200; first kimik3 benchmark config)"
     - "Aggregated TP8 x PP2 across 2 B200 nodes (16 GPUs), plain TP (NOT TEP: ep 1, no enable-expert-parallel) — the native MXFP4 checkpoint (2.8T total params, ~1.4TB weights) does not fit one 8xB200 node, so TP8 shards attention/dense and PP2 splits the 93 layers. Aggregated mode (prefill num-worker 1 + decode num-worker 0, RECIPES.md section 5): one worker serves prefill and decode, no P/D KV transfer"
-    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --trust-remote-code, and kimi_k3 parsers via dynamo's namespaced worker args --dyn-tool-call-parser kimi_k3 --reasoning-parser kimi_k3 --dyn-reasoning-parser kimi_k3 (the vLLM OpenAI-frontend spellings --enable-auto-tool-choice/--tool-call-parser are rejected by the dynamo worker entrypoint as unrecognized arguments)"
+    - "Dedicated bring-up image vllm/vllm-openai:kimi-k3 with VLLM_ENABLE_K3_LATENT_MOE_TAIL_FUSION=1 (fuses the K3 LatentMoE tail path), --load-format fastsafetensors, --moe-backend auto, --gpu-memory-utilization 0.95, --no-enable-flashinfer-autotune, --trust-remote-code, and kimi_k3 parsers via the plain vLLM spellings --tool-call-parser kimi_k3 --reasoning-parser kimi_k3 (parser-flag experiment variant B; sibling agentic-experiment PRs try dynamo's --dyn-* namespaced args and no parser flags)"
     - "No explicit max-model-len (vLLM derives the native 1M window from the model config; K3's KDA layers keep per-token KV small — only the 24 gated-MLA layers hold cache), prefix caching on for trajectory reuse, max-num-batched-tokens 8192 so a single long prefill cannot OOM a pipeline stage; single-concurrency smoke test at conc 8 (widen the curve once the topology is proven green)"
     - "Dynamo hash-pinned to ba83080ecd31c1ce918559e576d3c5bc9e092ff1 ('feat: Added support for Kimi-K3', tag v1.4.0-kimi-k3-dev.1, published 2026-07-27) via srt-slurm's hash-cached source install. Required because dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registered with the frontend and every chat completion 404'd, aborting the AgentX warmup (third sweep attempt); the TP8xPP2 engine itself loaded and served (health-ready in ~14 min via fastsafetensors)"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"

From 2fe84a8d149d09e5497978d316b5851980846d9e Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Mon, 27 Jul 2026 15:46:15 -0400
Subject: [PATCH 9/9] docs: point changelog and MODELS rows at experiment PR
 #2357
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

中文：将更新日志条目与 MODELS 表格行链接指向实验 PR #2357。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 MODELS.md           | 2 +-
 MODELS_zh.md        | 2 +-
 perf-changelog.yaml | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/MODELS.md b/MODELS.md
index 31cb9df252..00936552c9 100644
--- a/MODELS.md
+++ b/MODELS.md
@@ -18,7 +18,7 @@ This document tracks every model benchmarked by InferenceX-e2e: when it was adde
 | Model architecture class | Prefix | Date added | Active scenarios | Deprecated scenarios |
 |---|---|---|---|---|
 | Qwen3.8 2.4T | `qwen3.8` | TBD | Agentic coding | |
-| Kimi-K3 | `kimik3` | 2026-07-27 ([#2355](https://github.com/SemiAnalysisAI/InferenceX/pull/2355)) | Agentic coding | |
+| Kimi-K3 | `kimik3` | 2026-07-27 ([#2357](https://github.com/SemiAnalysisAI/InferenceX/pull/2357)) | Agentic coding | |
 | GLM-5.2 | `glm5.2` | 2026-07-18 ([#2268](https://github.com/SemiAnalysisAI/InferenceX/pull/2268)) | Agentic coding | |
 | MiniMax-M3 | `minimaxm3` | 2026-06-12 ([#1724](https://github.com/SemiAnalysisAI/InferenceX/pull/1724)) | Single-turn 8k1k, Agentic coding | Single-turn 1k1k |
 | DeepSeek-V4-Pro | `dsv4` | 2026-04-24 ([#1130](https://github.com/SemiAnalysisAI/InferenceX/pull/1130)) | Single-turn 8k1k, Agentic coding | Single-turn 1k1k |
diff --git a/MODELS_zh.md b/MODELS_zh.md
index e71067c13b..7be80603bd 100644
--- a/MODELS_zh.md
+++ b/MODELS_zh.md
@@ -18,7 +18,7 @@
 | 模型架构类别 | 前缀 | 加入日期 | 启用场景 | 已弃用场景 |
 |---|---|---|---|---|
 | Qwen3.8 2.4T | `qwen3.8` | 待定 | 智能体编码 | |
-| Kimi-K3 | `kimik3` | 2026-07-27 ([#2355](https://github.com/SemiAnalysisAI/InferenceX/pull/2355)) | 智能体编码 | |
+| Kimi-K3 | `kimik3` | 2026-07-27 ([#2357](https://github.com/SemiAnalysisAI/InferenceX/pull/2357)) | 智能体编码 | |
 | GLM-5.2 | `glm5.2` | 2026-07-18（[#2268](https://github.com/SemiAnalysisAI/InferenceX/pull/2268)） | 智能体编码 | |
 | MiniMax-M3 | `minimaxm3` | 2026-06-12（[#1724](https://github.com/SemiAnalysisAI/InferenceX/pull/1724)） | 单轮 8k1k、智能体编码 | 单轮 1k1k |
 | DeepSeek-V4-Pro | `dsv4` | 2026-04-24（[#1130](https://github.com/SemiAnalysisAI/InferenceX/pull/1130)） | 单轮 8k1k、智能体编码 | 单轮 1k1k |
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 06341f4bf1..4980393654 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5127,4 +5127,4 @@
     - "Dynamo hash-pinned to ba83080ecd31c1ce918559e576d3c5bc9e092ff1 ('feat: Added support for Kimi-K3', tag v1.4.0-kimi-k3-dev.1, published 2026-07-27) via srt-slurm's hash-cached source install. Required because dynamo 1.2.1's rust frontend tokenizer rejects Kimi-K3's tiktoken model_type kimi_k3 (only kimi/kimi_k2/kimi_k25/deepseek_v3 supported), so the model never registered with the frontend and every chat completion 404'd, aborting the AgentX warmup (third sweep attempt); the TP8xPP2 engine itself loaded and served (health-ready in ~14 min via fastsafetensors)"
     - "Model pre-staged at /lustre/fsw/models/Kimi-K3 (moonshotai/Kimi-K3); launcher launch_b200-dgxc.sh gains the kimik3/fp4 model-path mapping, pins the agentic srt-slurm base to upstream NVIDIA/srt-slurm v1.0.36 (validated in #2302/#2341; replaces the cquil11/srt-slurm-nv fork branch whose older srtctl schema rejects newer recipe fields such as benchmark.aiperf_server_metrics), overlays the kimi-k3 agentic recipes onto the clone, and adds the agentic default_mounts (/aiperf_mmap_cache, /hf_hub_cache) already used by the GB200/GB300 agentic paths"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
-  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2355
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2357