minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

docs(DeepSeek-V4): add h200|big verified recipes + tune H200 Pro parameters (#23742)

Browse Source
This commit is contained in:
Ethan (Yusheng) Su
2026-04-25 21:44:51 -07:00
committed by GitHub
parent 282b47fcfd
commit 3cfd1561df
1 changed files with 22 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx
+22 -8
View File
@@ -184,6 +184,9 @@ export const DeepSeekV4Deployment = () => {
"gb300|small|max-throughput", "gb300|small|max-throughput",
"h200|small|cp", "h200|small|cp",
"h200|small|pd-disagg", "h200|small|pd-disagg",
"h200|big|low-latency",
"h200|big|balanced",
"h200|big|max-throughput",
"h200|big|pd-disagg", "h200|big|pd-disagg",
"gb300|small|cp", "gb300|small|cp",
"gb300|big|cp", "gb300|big|cp",
@@ -272,7 +275,9 @@ export const DeepSeekV4Deployment = () => {
} }
} else if (recipe === "balanced") { } else if (recipe === "balanced") {
if (hardware === "h200") { if (hardware === "h200") {
recipeEnv.push("SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256"); recipeEnv.push(isBig
? "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=128"
: "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256");
} else { } else {
// Blackwell: small=1024, big=256 (allinone ternary). // Blackwell: small=1024, big=256 (allinone ternary).
recipeEnv.push(isBig recipeEnv.push(isBig
@@ -281,7 +286,9 @@ export const DeepSeekV4Deployment = () => {
} }
} else if (recipe === "max-throughput") { } else if (recipe === "max-throughput") {
if (hardware === "h200") { if (hardware === "h200") {
recipeEnv.push("SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256"); recipeEnv.push(isBig
? "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=128"
: "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256");
} else { } else {
recipeEnv.push(isBig recipeEnv.push(isBig
? "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256" ? "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256"
@@ -328,8 +335,8 @@ export const DeepSeekV4Deployment = () => {
flags.push(" --moe-runner-backend flashinfer_mxfp4"); flags.push(" --moe-runner-backend flashinfer_mxfp4");
} }
if (hardware === "h200" && isBig) { if (hardware === "h200" && isBig) {
flags.push(" --cuda-graph-max-bs 32"); flags.push(" --cuda-graph-max-bs 8");
flags.push(" --max-running-requests 64"); flags.push(" --max-running-requests 32");
} }
// MTP 3/4 // MTP 3/4
flags.push(" --speculative-algo EAGLE"); flags.push(" --speculative-algo EAGLE");
@@ -340,7 +347,7 @@ export const DeepSeekV4Deployment = () => {
flags.push(" --chunked-prefill-size 4096"); flags.push(" --chunked-prefill-size 4096");
flags.push(" --disable-flashinfer-autotune"); flags.push(" --disable-flashinfer-autotune");
} }
if (isBig) flags.push(" --mem-fraction-static 0.82"); if (isBig) flags.push(" --mem-fraction-static 0.88");
} else if (recipe === "balanced") { } else if (recipe === "balanced") {
// allinone balanced: TP + DP + DP-attn + DeepEP + MTP_112. // allinone balanced: TP + DP + DP-attn + DeepEP + MTP_112.
// H200 small: cg=128 max-run=128 | H200 big: cg=128 max-run=128 (same) // H200 small: cg=128 max-run=128 | H200 big: cg=128 max-run=128 (same)
@@ -355,12 +362,17 @@ export const DeepSeekV4Deployment = () => {
flags.push(" --speculative-num-steps 1"); flags.push(" --speculative-num-steps 1");
flags.push(" --speculative-eagle-topk 1"); flags.push(" --speculative-eagle-topk 1");
flags.push(" --speculative-num-draft-tokens 2"); flags.push(" --speculative-num-draft-tokens 2");
if (isBig && hardware === "gb200") { if (hardware === "h200" && isBig) {
flags.push(" --mem-fraction-static 0.88");
} else if (isBig && hardware === "gb200") {
flags.push(" --mem-fraction-static 0.78"); flags.push(" --mem-fraction-static 0.78");
} else if (isBig) { } else if (isBig) {
flags.push(" --mem-fraction-static 0.82"); flags.push(" --mem-fraction-static 0.82");
} }
if (hardware === "h200") { if (hardware === "h200" && isBig) {
flags.push(" --cuda-graph-max-bs 8");
flags.push(" --max-running-requests 32");
} else if (hardware === "h200") {
flags.push(" --cuda-graph-max-bs 128"); flags.push(" --cuda-graph-max-bs 128");
flags.push(" --max-running-requests 128"); flags.push(" --max-running-requests 128");
} else if (isBig && hardware === "b200") { } else if (isBig && hardware === "b200") {
@@ -386,7 +398,9 @@ export const DeepSeekV4Deployment = () => {
flags.push(" --enable-dp-attention"); flags.push(" --enable-dp-attention");
if (multinode) flags.push(...multiNodeFlags(nnodes)); if (multinode) flags.push(...multiNodeFlags(nnodes));
flags.push(" --moe-a2a-backend deepep"); flags.push(" --moe-a2a-backend deepep");
if (isBig && hardware === "gb200") { if (hardware === "h200" && isBig) {
flags.push(" --mem-fraction-static 0.88");
} else if (isBig && hardware === "gb200") {
flags.push(" --mem-fraction-static 0.78"); flags.push(" --mem-fraction-static 0.78");
} else if (isBig) { } else if (isBig) {
flags.push(" --mem-fraction-static 0.82"); flags.push(" --mem-fraction-static 0.82");