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

docs: fix Nemotron Super MTP deployment command (spec-v2 + B200) (#27184)

Browse Source 
Co-authored-by: zijiexia <37504505+zijiexia@users.noreply.github.com>
This commit is contained in:
Jiajun Li
2026-06-03 14:52:47 -07:00
committed by GitHub
co-authored by zijiexia
parent 978fb6ed1a
commit 90985117a5
1 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs_new/src/snippets/autoregressive/nemotron3-super-deployment.jsx
+7 -3
View File
@@ -39,7 +39,10 @@ export const Nemotron3SuperDeployment = () => {
{ id: 'enabled', label: 'Enabled', default: false }, { id: 'enabled', label: 'Enabled', default: false },
{ id: 'disabled', label: 'Disabled', default: true } { id: 'disabled', label: 'Disabled', default: true }
], ],
commandRule: (value) => value === 'enabled' ? '--speculative-algorithm EAGLE \\\n --speculative-num-steps 3 \\\n --speculative-eagle-topk 1 \\\n --speculative-num-draft-tokens 4 \\\n --disable-radix-cache' : null // trtllm_mha is Blackwell-only; on B200 it replaces the flashinfer default,
// whose per-step plan() host-sync breaks the spec-v2 overlap scheduler. H200
// defaults to fa3 (no such sync), so no override is needed there.
commandRule: (value, state) => value === 'enabled' ? '--speculative-algorithm EAGLE \\\n --speculative-num-steps 3 \\\n --speculative-eagle-topk 1 \\\n --speculative-num-draft-tokens 4 \\\n --mamba-scheduler-strategy extra_buffer' + (state.hardware === 'b200' ? ' \\\n --attention-backend trtllm_mha' : '') : null
}, },
kvcache: { kvcache: {
name: 'kvcache', name: 'kvcache',
@@ -75,7 +78,8 @@ export const Nemotron3SuperDeployment = () => {
const modelPath = MODEL_PATHS[model] || MODEL_PATHS['bf16']; const modelPath = MODEL_PATHS[model] || MODEL_PATHS['bf16'];
let cmd = 'python3 -m sglang.launch_server \\\n'; const specV2Env = values.mtp === 'enabled' ? 'SGLANG_ENABLE_SPEC_V2=1 ' : '';
let cmd = `${specV2Env}sglang serve \\\n`;
cmd += ` --model-path ${modelPath} \\\n`; cmd += ` --model-path ${modelPath} \\\n`;
cmd += ` --trust-remote-code \\\n`; cmd += ` --trust-remote-code \\\n`;
cmd += ` --tp ${tp} \\\n`; cmd += ` --tp ${tp} \\\n`;
@@ -86,7 +90,7 @@ export const Nemotron3SuperDeployment = () => {
for (const [key, option] of Object.entries(options)) { for (const [key, option] of Object.entries(options)) {
if (option.commandRule) { if (option.commandRule) {
const rule = option.commandRule(values[key]); const rule = option.commandRule(values[key], values);
if (rule) { if (rule) {
cmd += ` ${rule} \\\n`; cmd += ` ${rule} \\\n`;
} }