From 90985117a5f7b13fdd4374e2115716b0e18ec448 Mon Sep 17 00:00:00 2001 From: Jiajun Li <48857426+guapisolo@users.noreply.github.com> Date: Wed, 3 Jun 2026 14:52:47 -0700 Subject: [PATCH] docs: fix Nemotron Super MTP deployment command (spec-v2 + B200) (#27184) Co-authored-by: zijiexia <37504505+zijiexia@users.noreply.github.com> --- .../autoregressive/nemotron3-super-deployment.jsx | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/docs_new/src/snippets/autoregressive/nemotron3-super-deployment.jsx b/docs_new/src/snippets/autoregressive/nemotron3-super-deployment.jsx index c9c623695..9afd87862 100644 --- a/docs_new/src/snippets/autoregressive/nemotron3-super-deployment.jsx +++ b/docs_new/src/snippets/autoregressive/nemotron3-super-deployment.jsx @@ -39,7 +39,10 @@ export const Nemotron3SuperDeployment = () => { { id: 'enabled', label: 'Enabled', default: false }, { 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: { name: 'kvcache', @@ -75,7 +78,8 @@ export const Nemotron3SuperDeployment = () => { 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 += ` --trust-remote-code \\\n`; cmd += ` --tp ${tp} \\\n`; @@ -86,7 +90,7 @@ export const Nemotron3SuperDeployment = () => { for (const [key, option] of Object.entries(options)) { if (option.commandRule) { - const rule = option.commandRule(values[key]); + const rule = option.commandRule(values[key], values); if (rule) { cmd += ` ${rule} \\\n`; }