From 7d515c6d1f2217fee8924c4ad4def7f675ee640e Mon Sep 17 00:00:00 2001
From: zijiexia <37504505+zijiexia@users.noreply.github.com>
Date: Wed, 13 May 2026 00:50:15 -0700
Subject: [PATCH] docs: prepend SGLANG_JIT_DEEPGEMM_PRECOMPILE=0 for H200 FP8
 Flash max-throughput (#25152)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 .../src/snippets/autoregressive/deepseek-v4-deployment.jsx     | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx b/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx
index 95c6eeb5d..080a8e7ec 100644
--- a/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx
+++ b/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx
@@ -391,6 +391,9 @@ export const DeepSeekV4Deployment = () => {
       }
     } else if (recipe === "max-throughput") {
       if (hardware === "h200") {
+        if (!isBig) {
+          recipeEnv.push("SGLANG_JIT_DEEPGEMM_PRECOMPILE=0");
+        }
         recipeEnv.push(isBig
           ? "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=128"
           : "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=256");