Make the remaining DeepSeek-V4.1 NVIDIA cells start (#38861)

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

docs/src/snippets/configs/deepseek-ai/deepseek-v4_1.jsx

@@ -1,8 +1,16 @@
 // Single `export const config` literal — no spreads/calls/IIFE (Mintlify re-evals at hydration).
 //
 // Cells marked `verified` are transcribed from recorded runs on that hardware with
-// real weights. DP-Attention, DeepEP and MegaMoE are absent by design: they have
-// never been enabled on this model. EP is set equal to TP on every shape here.
+// real weights.
+//
+// Every DSpark cell caps --cuda-graph-max-bs-decode: the derived batch list does
+// not fit while capturing the DSpark decode graphs, on any NVIDIA platform. The
+// MI350X cell already carried the equivalent --cuda-graph-max-bs. H200 is the
+// tightest board at 140 GiB and needs the cap on both cells plus a lower memory
+// fraction; the three other High-Throughput cells start without either.
+//
+// DP-Attention, DeepEP and MegaMoE are absent by design: they have never been
+// enabled on this model. EP is set equal to TP on every shape here.
 
 export const config = {
   modelName: "DeepSeek-V4.1",
@@ -182,6 +190,8 @@ export const config = {
         "--mem-fraction-static 0.8",
         "--speculative-algorithm DSPARK",
         "--speculative-dspark-block-size 5",
+        // Decode CUDA graphs: the derived batch list does not fit here.
+        "--cuda-graph-max-bs-decode 64",
         "--reasoning-parser auto",
         "--tool-call-parser auto",
         "--host {{HOST_IP}}",
@@ -209,20 +219,24 @@ export const config = {
       ],
     },
 
-    // ---------- H200: 8x H200, TP8 + EP8. No MXFP8 dense path on Hopper;
-    // verification round open. ----------
+    // ---------- H200: 8x H200, TP8 + EP8. No MXFP8 dense path on Hopper. The
+    // tightest board here at 140 GiB, so both cells also need the memory
+    // fraction pulled back; the cap alone still leaves the graphs short. ------
     {
       match: { hw: "h200", strategy: "low-latency" },
       nnodes: 1,
-      verificationStatus: "in-progress",
+      verified: true,
       flags: [
         "--trust-remote-code",
         "--model-path {{MODEL_NAME}}",
         "--tp 8",
         "--ep-size 8",
+        "--mem-fraction-static 0.8",
         "--attention-backend dsv4",
         "--moe-runner-backend flashinfer_mxfp4",
         "--enable-decoder-swa-bounded-replay",
+        // Decode CUDA graphs: the derived batch list does not fit here.
+        "--cuda-graph-max-bs-decode 64",
         "--reasoning-parser auto",
         "--tool-call-parser auto",
         "--host {{HOST_IP}}",
@@ -232,15 +246,18 @@ export const config = {
     {
       match: { hw: "h200", strategy: "high-throughput" },
       nnodes: 1,
-      verificationStatus: "in-progress",
+      verified: true,
       flags: [
         "--trust-remote-code",
         "--model-path {{MODEL_NAME}}",
         "--tp 8",
         "--ep-size 8",
+        "--mem-fraction-static 0.8",
         "--attention-backend dsv4",
         "--moe-runner-backend flashinfer_mxfp4",
         "--max-running-requests 256",
+        // Decode CUDA graphs: the derived batch list does not fit here.
+        "--cuda-graph-max-bs-decode 64",
         "--reasoning-parser auto",
         "--tool-call-parser auto",
         "--host {{HOST_IP}}",
@@ -248,12 +265,12 @@ export const config = {
       ],
     },
 
-    // ---------- B200: verification round open. Mirrors the GB300 recipe because
-    // the kernels dispatch by architecture family. ----------
+    // ---------- B200: 4x B200, TP4 + EP4. Mirrors the GB300 recipe — the
+    // kernels dispatch by architecture family. ----------
     {
       match: { hw: "b200", strategy: "low-latency" },
       nnodes: 1,
-      verificationStatus: "in-progress",
+      verified: true,
       flags: [
         "--trust-remote-code",
         "--model-path {{MODEL_NAME}}",
@@ -262,6 +279,8 @@ export const config = {
         "--mem-fraction-static 0.8",
         "--speculative-algorithm DSPARK",
         "--speculative-dspark-block-size 5",
+        // Decode CUDA graphs: the derived batch list does not fit here.
+        "--cuda-graph-max-bs-decode 64",
         "--reasoning-parser auto",
         "--tool-call-parser auto",
         "--host {{HOST_IP}}",
@@ -271,7 +290,7 @@ export const config = {
     {
       match: { hw: "b200", strategy: "high-throughput" },
       nnodes: 1,
-      verificationStatus: "in-progress",
+      verified: true,
       flags: [
         "--trust-remote-code",
         "--model-path {{MODEL_NAME}}",
@@ -286,8 +305,7 @@ export const config = {
     },
 
     // ---------- B300: 4x B300, TP4 + EP4. Same recipe as GB300 — the kernels
-    // dispatch by architecture family — with one extra memory knob on
-    // Low-Latency that GB300 does not need. ----------
+    // dispatch by architecture family. ----------
     {
       match: { hw: "b300", strategy: "low-latency" },
       nnodes: 1,
@@ -300,8 +318,7 @@ export const config = {
         "--mem-fraction-static 0.8",
         "--speculative-algorithm DSPARK",
         "--speculative-dspark-block-size 5",
-        // GB300 does not need this, B300 does: the derived value (256) runs
-        // out of memory capturing the DSpark decode graphs at this fraction.
+        // Decode CUDA graphs: the derived batch list does not fit here.
         "--cuda-graph-max-bs-decode 64",
         "--reasoning-parser auto",
         "--tool-call-parser auto",