From 574274660f788174d925cacf99771b939ccba4de Mon Sep 17 00:00:00 2001
From: "jacky.cheng" <yichiche@amd.com>
Date: Wed, 19 Aug 2026 18:49:32 +0800
Subject: [PATCH] [AMD] cookbook: serve Qwen3.5 MXFP4 on MI355X with an
 fp8_e4m3 KV cache (#35445)

---
 docs/src/snippets/autoregressive/qwen35-deployment.jsx | 1 +
 1 file changed, 1 insertion(+)

diff --git a/docs/src/snippets/autoregressive/qwen35-deployment.jsx b/docs/src/snippets/autoregressive/qwen35-deployment.jsx
index 4d1fbd4af..f71c52f15 100644
--- a/docs/src/snippets/autoregressive/qwen35-deployment.jsx
+++ b/docs/src/snippets/autoregressive/qwen35-deployment.jsx
@@ -480,6 +480,7 @@ export const Qwen35Deployment = () => {
         // (this recipe uses quick all-reduce instead of AITER allreduce fusion).
         // Add the FP4-specific flags here.
         cmd += ' \\\n  --disable-radix-cache';
+        cmd += ' \\\n  --kv-cache-dtype fp8_e4m3';
         // Cap concurrency under MTP to avoid OOM at tp=2.
         if (speculative === 'enabled') {
           cmd += ' \\\n  --max-running-requests 128';