From 8d4a22c5af1fa41b30f51fe2d31f1302c6b84d18 Mon Sep 17 00:00:00 2001
From: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
Date: Thu, 18 Jun 2026 12:42:00 +0800
Subject: [PATCH] [Docs] Add fp8 kv cache for tokenspeed mla docs (#28201)

---
 .../src/snippets/autoregressive/kimi-k25-deployment.jsx   | 8 +++++---
 .../src/snippets/autoregressive/kimi-k26-deployment.jsx   | 6 ++++--
 .../snippets/autoregressive/kimi-k27-code-deployment.jsx  | 6 ++++--
 3 files changed, 13 insertions(+), 7 deletions(-)

diff --git a/docs_new/src/snippets/autoregressive/kimi-k25-deployment.jsx b/docs_new/src/snippets/autoregressive/kimi-k25-deployment.jsx
index e847c3469..982fd443d 100644
--- a/docs_new/src/snippets/autoregressive/kimi-k25-deployment.jsx
+++ b/docs_new/src/snippets/autoregressive/kimi-k25-deployment.jsx
@@ -196,13 +196,15 @@ export const KimiK25Deployment = () => {
       cmd += ' \\\n  --speculative-algorithm EAGLE3 \\\n  --speculative-num-steps 3 \\\n  --speculative-eagle-topk 1 \\\n  --speculative-num-draft-tokens 4 \\\n  --speculative-draft-model-path lightseekorg/kimi-k2.5-eagle3-mla';
     }
 
+    const usesTokenspeedMla = hardware === 'b300' || hardware === 'gb300';
+
     // Blackwell (B300/GB300): tokenspeed MLA attention backend
-    if (hardware === 'b300' || hardware === 'gb300') {
+    if (usesTokenspeedMla) {
       cmd += ' \\\n  --attention-backend tokenspeed_mla';
     }
 
-    // AMD: FP8 KV cache for memory efficiency
-    if (isAMD) {
+    // FP8 KV cache for AMD memory efficiency and tokenspeed MLA compatibility
+    if (isAMD || usesTokenspeedMla) {
       cmd += ' \\\n  --kv-cache-dtype fp8_e4m3';
     }
 
diff --git a/docs_new/src/snippets/autoregressive/kimi-k26-deployment.jsx b/docs_new/src/snippets/autoregressive/kimi-k26-deployment.jsx
index 3dc2e820e..3e7b998bc 100644
--- a/docs_new/src/snippets/autoregressive/kimi-k26-deployment.jsx
+++ b/docs_new/src/snippets/autoregressive/kimi-k26-deployment.jsx
@@ -194,11 +194,13 @@ export const KimiK26Deployment = () => {
       cmd += ' \\\n  --speculative-draft-model-path lightseekorg/kimi-k2.6-eagle3.1-mla';
     }
 
-    if (hardware === 'b300' || hardware === 'gb300') {
+    const usesTokenspeedMla = hardware === 'b300' || hardware === 'gb300';
+
+    if (usesTokenspeedMla) {
       cmd += ' \\\n  --attention-backend tokenspeed_mla';
     }
 
-    if (isAMD) {
+    if (isAMD || usesTokenspeedMla) {
       cmd += ' \\\n  --kv-cache-dtype fp8_e4m3';
     }
 
diff --git a/docs_new/src/snippets/autoregressive/kimi-k27-code-deployment.jsx b/docs_new/src/snippets/autoregressive/kimi-k27-code-deployment.jsx
index e2f6c695f..e90c74e03 100644
--- a/docs_new/src/snippets/autoregressive/kimi-k27-code-deployment.jsx
+++ b/docs_new/src/snippets/autoregressive/kimi-k27-code-deployment.jsx
@@ -144,11 +144,13 @@ export const KimiK27CodeDeployment = () => {
       cmd += ' \\\n  --tool-call-parser kimi_k2';
     }
 
-    if (hardware === 'b300' || hardware === 'gb300') {
+    const usesTokenspeedMla = hardware === 'b300' || hardware === 'gb300';
+
+    if (usesTokenspeedMla) {
       cmd += ' \\\n  --attention-backend tokenspeed_mla';
     }
 
-    if (isAMD) {
+    if (isAMD || usesTokenspeedMla) {
       cmd += ' \\\n  --kv-cache-dtype fp8_e4m3';
     }