[Doc] cookbook: Refresh SemiAnalysis AgentX launches with seven open InferenceX PRs

docs/src/snippets/_agentx.jsx

@@ -170,7 +170,8 @@ export const AgentX = ({ data }) => {
       const parts = [];
       if (b.workers > 1) parts.push(`Start ${b.workers} ${b.role} workers${n > 1 ? `, each spanning ${n} nodes` : b.gpu_pinning ? " across the shared nodes" : ", one per node"}.`);
       if (n > 1) parts.push(`Run on every node of a worker with NODE_RANK=0..${n - 1}; ${ROLE_HEAD[b.role]} is that worker's first node.`);
-      if (b.gpu_pinning) parts.push(`Workers share nodes: pin GPUs per worker with CUDA_VISIBLE_DEVICES = ${b.gpu_pinning.map((w, i) => `worker ${i}: ${w.filter(Boolean).join(" / ") || "all"}`).join(" · ")}.`);
+      if (b.local_ranks) parts.push(`Workers share nodes: set LOCAL_WORKER_RANK per worker (${b.local_ranks.map((r, i) => `worker ${i}: ${r}`).join(" · ")}); the first lines pick its GPUs and ports.`);
+      else if (b.gpu_pinning) parts.push(`Workers share nodes: pin GPUs per worker with CUDA_VISIBLE_DEVICES = ${b.gpu_pinning.map((w, i) => `worker ${i}: ${w.filter(Boolean).join(" / ") || "all"}`).join(" · ")}.`);
       if (router === "sglang" && workerCount === 1 && b.role === "agg" && n === 1) parts.push("Serves the OpenAI API on port 30000.");
       if (router === "dynamo") parts.push("Workers register with Dynamo through etcd; start them after etcd/NATS.");
       return parts.join(" ") || `Run on the ${b.role} node.`;
@@ -214,7 +215,7 @@ export const AgentX = ({ data }) => {
   };
   const PLACEHOLDERS = {
     NODE_RANK: "rank of this node within its worker (0 on the worker's first node)",
-    LOCAL_WORKER_RANK: "prefill worker index on this node (0 or 1; selects GPUs and distinct ports)",
+    LOCAL_WORKER_RANK: "this worker's index on its node (selects its GPUs and ports)",
     NODE_IP: "IP of the node the command runs on",
     ETCD_IP: "node running etcd",
     NATS_IP: "node running NATS",

docs/src/snippets/agentx/Qwen/qwen3.5.jsx

@@ -1,37 +1,47 @@
 // GENERATED by the sa-infx-cookbook sync scripts — do not edit by hand.
 // Launch commands mirrored from SemiAnalysis InferenceX AgentX submissions
-// (SemiAnalysisAI/InferenceX main@746150a2786f plus open PRs #3930, #4303, #4333, #4346, #4389), with SA-specific settings removed.
+// (SemiAnalysisAI/InferenceX main@39b1aab8c009 plus open PRs #4333, #4346, #4398, #4400, #4401, #4404, #4417), with SA-specific settings removed.
 export const agentx = {
  "model": "Qwen3.5",
- "generatedOn": "2026-10-21",
+ "generatedOn": "2026-10-22",
  "source": {
   "repo": "https://github.com/SemiAnalysisAI/InferenceX",
-  "main": "746150a2786f5cfe6fc474220d4a3335365f14c4",
+  "main": "39b1aab8c009ac29bafd2dbe75a99c2273dc3ac3",
   "prs": {
-   "4389": {
-    "run": null,
-    "sha": "e390de596bf5bae7a561308095ad44cf88b74f51",
-    "status": "in-progress"
+   "4333": {
+    "run": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/39355392627",
+    "sha": "02b72dc0bf88eca6f6638300739732d4c7640ddf",
+    "status": "verified-pr"
    },
    "4346": {
     "run": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/39534932794",
     "sha": "566c6e105131c0cbfd4b668b51cbc8a861336397",
     "status": "verified-pr"
    },
-   "4333": {
-    "run": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/39355392627",
-    "sha": "02b72dc0bf88eca6f6638300739732d4c7640ddf",
-    "status": "verified-pr"
+   "4398": {
+    "run": null,
+    "sha": "4d4e4c7392c088d16ba83b0b9db165b7b752a99e",
+    "status": "in-progress"
    },
-   "4303": {
+   "4400": {
     "run": null,
-    "sha": "a0e5be74248f09552f88904f90dc71e8f190a3e9",
+    "sha": "11358f4843ec3e086014d3bba1a102e400944a59",
     "status": "in-progress"
    },
-   "3930": {
-    "run": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/39551262792",
-    "sha": "c093d2052d2e4f6aeaa8162e9e81219f31556fe4",
-    "status": "verified-pr"
+   "4401": {
+    "run": null,
+    "sha": "cd647ee0072122bbef9f53ed6fe502eb504813ff",
+    "status": "in-progress"
+   },
+   "4404": {
+    "run": null,
+    "sha": "316905a060115a9a393f6a464d3646b830a3c74e",
+    "status": "in-progress"
+   },
+   "4417": {
+    "run": null,
+    "sha": "ecde73b01707bef866f46711eadada1838fcf032",
+    "status": "in-progress"
    }
   }
  },
@@ -112,8 +122,8 @@ export const agentx = {
   }
  ],
  "kvMooncake": {
-  "donor": "dsv4-fp4-gb300-dynamo-sglang-agentic-disagg",
-  "donorHardware": "gb300",
+  "donor": "dsv4-fp4-gb200-dynamo-sglang-agentic-disagg",
+  "donorHardware": "gb200",
   "master": {
    "dynamo": {
     "kind": "mooncake-master",
@@ -2452,7 +2462,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/agg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -2950,7 +2960,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/agg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -3072,7 +3082,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 1 decode worker on 1 node. The prefill and decode workers share a node. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -3094,7 +3110,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/disagg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -3498,9 +3514,9 @@ export const agentx = {
         ["--model-path", "nvidia/Qwen3.5-397B-A17B-NVFP4"],
         ["--served-model-name", "nvidia/Qwen3.5-397B-A17B-NVFP4"],
         ["--host", "0.0.0.0"],
-        ["--port", "30000"],
+        ["--port", "30032"],
         ["--disaggregation-mode", "decode"],
-        ["--disaggregation-bootstrap-port", "8998"],
+        ["--disaggregation-bootstrap-port", "8999"],
         ["--request-plane", "tcp"],
         ["--attention-backend", "trtllm_mha"],
         ["--chunked-prefill-size", "4096"],
@@ -3585,7 +3601,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_0_HEAD_IP:30032"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
         ["--host", "0.0.0.0"],
@@ -3609,7 +3625,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 2 decode workers on 1 node. LOCAL_WORKER_RANK gives each worker on a node its own GPUs and ports. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -3631,7 +3653,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/disagg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -3703,8 +3725,12 @@ export const agentx = {
         ["0,1"],
         ["2,3"]
        ],
+       "local_ranks": [0, 1],
+       "preamble": ["# LOCAL_WORKER_RANK is this worker's index on its node (0 or 1).", "case \"$LOCAL_WORKER_RANK\" in 0) export CUDA_VISIBLE_DEVICES=0,1 ;; 1) export CUDA_VISIBLE_DEVICES=2,3 ;; *) echo \"Set LOCAL_WORKER_RANK to 0 or 1\" >&2; exit 1 ;; esac", "PORT=$((30000 + 32 * LOCAL_WORKER_RANK))", "BOOTSTRAP_PORT=$((8998 + LOCAL_WORKER_RANK))"],
        "diff": {
         "args": {
+         "--port": ["$PORT"],
+         "--disaggregation-bootstrap-port": ["$BOOTSTRAP_PORT"],
          "--cuda-graph-max-bs-decode": ["3"],
          "--max-running-requests": ["3"],
          "--pp-max-micro-batch-size": ["3"],
@@ -3761,7 +3787,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_1_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_1_HEAD_IP:30032"],
         ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
@@ -3808,7 +3834,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/disagg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -4092,7 +4118,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/disagg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -4360,7 +4386,7 @@ export const agentx = {
     "recipe": "inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/disagg-variants.yaml",
     "run": "39355392627",
     "pr": "4333",
-    "comment": "6331658889",
+    "comment": "6334247978",
     "status": "verified-pr"
    },
    "points": [
@@ -5963,7 +5989,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 1 decode worker on 1 node. The prefill and decode workers share a node. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -6130,7 +6162,7 @@ export const agentx = {
         ["--model-path", "Qwen/Qwen3.5-397B-A17B-FP8"],
         ["--served-model-name", "Qwen/Qwen3.5-397B-A17B-FP8"],
         ["--host", "0.0.0.0"],
-        ["--port", "30000"],
+        ["--port", "30032"],
         ["--disaggregation-mode", "decode"],
         ["--request-plane", "tcp"],
         ["--attention-backend", "trtllm_mha"],
@@ -6208,7 +6240,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_0_HEAD_IP:30032"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
         ["--host", "0.0.0.0"],
@@ -6563,7 +6595,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 1 decode worker on 1 node. The prefill and decode workers share a node. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -6817,7 +6855,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_0_HEAD_IP:30032"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
         ["--host", "0.0.0.0"],
@@ -6841,7 +6879,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 1 decode worker on 1 node. The prefill and decode workers share a node. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -6963,7 +7007,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_0_HEAD_IP:30032"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
         ["--host", "0.0.0.0"],
@@ -9810,7 +9854,13 @@ export const agentx = {
    "imageStatus": "ok",
    "imageCommit": null,
    "submitted": "dynamo",
-   "setup": [],
+   "setup": [
+    {
+     "kind": "note",
+     "script": "Placement",
+     "reason": "Workers share nodes: 1 prefill worker on 1 node, 1 decode worker on 1 node. The prefill and decode workers share a node. The GPU count above counts the GPUs the workers use; reserve whole nodes when planning capacity."
+    }
+   ],
    "disclosures": ["synthetic-acceptance"],
    "kv": {
     "available": ["none", "hicache", "mooncake"],
@@ -10129,7 +10179,7 @@ export const agentx = {
         ["--model-path", "Qwen/Qwen3.5-397B-A17B-FP8"],
         ["--served-model-name", "Qwen/Qwen3.5-397B-A17B-FP8"],
         ["--host", "0.0.0.0"],
-        ["--port", "30000"],
+        ["--port", "30032"],
         ["--disaggregation-mode", "decode"],
         ["--request-plane", "tcp"],
         ["--attention-backend", "trtllm_mha"],
@@ -10204,7 +10254,7 @@ export const agentx = {
        "cmd": "python3 -m sglang_router.launch_router",
        "env": {},
        "args": [
-        ["--decode", "http://$DECODE_0_HEAD_IP:30000"],
+        ["--decode", "http://$DECODE_0_HEAD_IP:30032"],
         ["--prefill", "http://$PREFILL_0_HEAD_IP:30000", "8998"],
         ["--pd-disaggregation"],
         ["--host", "0.0.0.0"],