minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Default MegaMoE to W4A8 for Max-Throughput recipe (#26004)

Browse Source
This commit is contained in:
Yuhao Yang
2026-05-21 11:54:16 -07:00
committed by GitHub
parent b765faee30
commit 81d686d9fa
1 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx
+13 -2
View File
@@ -177,6 +177,16 @@ export const DeepSeekV4Deployment = () => {
) { ) {
next.megamoe = "disabled"; next.megamoe = "disabled";
} }
// Switching to max-throughput on supported hardware: default MegaMoE to
// W4A8 if it's currently disabled (best throughput config).
if (
(optionName === "recipe" || optionName === "hardware") &&
next.recipe === "max-throughput" &&
next.megamoe === "disabled" &&
!isMegamoeUnsupported(next)
) {
next.megamoe = "w4a8";
}
return next; return next;
}); });
}; };
@@ -605,7 +615,8 @@ export const DeepSeekV4Deployment = () => {
} }
// allinone H200 gates DEEPEP_LARGE_SMS_FLAG on !multinode — only H200 big // allinone H200 gates DEEPEP_LARGE_SMS_FLAG on !multinode — only H200 big
// is multi-node; all Blackwell cells get the flag unconditionally. // is multi-node; all Blackwell cells get the flag unconditionally.
if (!multinode) flags.push(DEEPEP_LARGE_SMS_FLAG); // Skip when MegaMoE is enabled (uses its own backend, not DeepEP).
if (!multinode && megamoe === "disabled") flags.push(DEEPEP_LARGE_SMS_FLAG);
} else if (recipe === "max-throughput") { } else if (recipe === "max-throughput") {
// allinone max-throughput: TP + DP + DP-attn + DeepEP (NO MTP). // allinone max-throughput: TP + DP + DP-attn + DeepEP (NO MTP).
// H200 small: cg=128 max-run=256 | H200 big: cg=128 max-run=256 (same) // H200 small: cg=128 max-run=256 | H200 big: cg=128 max-run=256 (same)
@@ -642,7 +653,7 @@ export const DeepSeekV4Deployment = () => {
flags.push(" --cuda-graph-max-bs 64"); flags.push(" --cuda-graph-max-bs 64");
flags.push(" --max-running-requests 256"); flags.push(" --max-running-requests 256");
} }
if (!multinode) flags.push(DEEPEP_LARGE_SMS_FLAG); if (!multinode && megamoe === "disabled") flags.push(DEEPEP_LARGE_SMS_FLAG);
} else if (recipe === "cp") { } else if (recipe === "cp") {
// allinone cp: TP (NO --dp) + DeepEP + _CP_FLAGS (mem-frac 0.78, max-run 1024). // allinone cp: TP (NO --dp) + DeepEP + _CP_FLAGS (mem-frac 0.78, max-run 1024).
// Blackwell big additionally: mem-frac 0.70 (overrides), cg=256, max-run=256. // Blackwell big additionally: mem-frac 0.70 (overrides), cg=256, max-run=256.