From c82d20d48ecc643ab2a84a0961b8b71901eeffb1 Mon Sep 17 00:00:00 2001
From: Baizhou Zhang <sobereddiezhang@gmail.com>
Date: Fri, 20 Mar 2026 01:04:32 -0700
Subject: [PATCH] Fix DeepSeek V32 FP4 test (#20984)

---
 python/sglang/test/test_utils.py                        | 3 ++-
 test/registered/quant/test_deepseek_v32_fp4_4gpu.py     | 9 +++++++++
 test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py | 9 +++++++++
 3 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/python/sglang/test/test_utils.py b/python/sglang/test/test_utils.py
index 32db3c07d..a987f3e9a 100644
--- a/python/sglang/test/test_utils.py
+++ b/python/sglang/test/test_utils.py
@@ -865,6 +865,7 @@ def popen_launch_server(
         Started subprocess.Popen object
     """
     other_args = other_args or []
+    offline_force_disabled = env.get("HF_HUB_OFFLINE") == "0"
 
     # Auto-detect device if needed
     if device == "auto":
@@ -885,7 +886,7 @@ def popen_launch_server(
     try:
         from sglang.utils import is_in_ci
 
-        if is_in_ci():
+        if is_in_ci() and not offline_force_disabled:
             per_run_marker_path = _try_enable_offline_mode_if_cache_complete(
                 model, env, other_args
             )
diff --git a/test/registered/quant/test_deepseek_v32_fp4_4gpu.py b/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
index 4bdcd1367..cc59c0a4c 100644
--- a/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
+++ b/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
@@ -1,3 +1,4 @@
+import os
 import unittest
 from types import SimpleNamespace
 
@@ -46,6 +47,10 @@ class TestDeepseekV32FP4DP(CustomTestCase):
             cls.base_url,
             timeout=SERVER_LAUNCH_TIMEOUT,
             other_args=other_args,
+            env={
+                **os.environ,
+                "HF_HUB_OFFLINE": "0",
+            },
         )
 
     @classmethod
@@ -113,6 +118,10 @@ class TestDeepseekV32FP4TP(CustomTestCase):
             cls.base_url,
             timeout=SERVER_LAUNCH_TIMEOUT,
             other_args=other_args,
+            env={
+                **os.environ,
+                "HF_HUB_OFFLINE": "0",
+            },
         )
 
     @classmethod
diff --git a/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py b/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
index 3a99f6ad9..e3f0fdabd 100644
--- a/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
+++ b/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
@@ -1,3 +1,4 @@
+import os
 import unittest
 from types import SimpleNamespace
 
@@ -60,6 +61,10 @@ class TestDeepseekV32FP4DPSpecV2(CustomTestCase):
                 cls.base_url,
                 timeout=SERVER_LAUNCH_TIMEOUT,
                 other_args=other_args,
+                env={
+                    **os.environ,
+                    "HF_HUB_OFFLINE": "0",
+                },
             )
 
     @classmethod
@@ -150,6 +155,10 @@ class TestDeepseekV32FP4TPSpecV2(CustomTestCase):
                 cls.base_url,
                 timeout=SERVER_LAUNCH_TIMEOUT,
                 other_args=other_args,
+                env={
+                    **os.environ,
+                    "HF_HUB_OFFLINE": "0",
+                },
             )
 
     @classmethod