From 01db8a78290a47636111135260935db58a221adb Mon Sep 17 00:00:00 2001
From: "Wang, FangYuan" <39615225+At1a8@users.noreply.github.com>
Date: Tue, 25 Aug 2026 13:13:59 +0800
Subject: [PATCH] [BugFix][Qwen3.8] Support Qwen3.8-MXFP4 DCP by registing
 Qwen3_5 text-only archs in mamba radix cache whitelists (#35297)

---
 python/sglang/srt/arg_groups/overrides.py | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/python/sglang/srt/arg_groups/overrides.py b/python/sglang/srt/arg_groups/overrides.py
index 66c83d463..6f29998b3 100644
--- a/python/sglang/srt/arg_groups/overrides.py
+++ b/python/sglang/srt/arg_groups/overrides.py
@@ -1680,6 +1680,10 @@ _MAMBA_RADIX_CACHE_ARCHS = frozenset(
         "InternS2PreviewForConditionalGeneration",
         "InternS2MobiusForConditionalGeneration",
         "Qwen3_5ForConditionalGeneration",
+        # Text-only entries of the same hybrid stack (models/qwen3_5_text.py);
+        # Qwen3.8-2.4T-A95B ships as Qwen3_5MoeForCausalLM.
+        "Qwen3_5MoeForCausalLM",
+        "Qwen3_5ForCausalLM",
         "MiniCPMV4_6ForConditionalGeneration",
         "NemotronHForCausalLM",
         "NemotronHPuzzleForCausalLM",
@@ -1700,6 +1704,10 @@ _MAMBA_EXTRA_BUFFER_ARCHS = frozenset(
         "KimiLinearForCausalLM",
         "Qwen3_5ForConditionalGeneration",
         "Qwen3_5MoeForConditionalGeneration",
+        # Text-only entries of the same hybrid stack (models/qwen3_5_text.py);
+        # Qwen3.8-2.4T-A95B ships as Qwen3_5MoeForCausalLM.
+        "Qwen3_5MoeForCausalLM",
+        "Qwen3_5ForCausalLM",
         "Qwen3NextForCausalLM",
         "InternS2PreviewForConditionalGeneration",
         "MiniCPMV4_6ForConditionalGeneration",