From 6dd0cefb2a36323d96b207736d1a432b985eccb0 Mon Sep 17 00:00:00 2001
From: Xiaoyu Zhang <1182563586@qq.com>
Date: Sat, 4 Jul 2026 21:17:32 +0800
Subject: [PATCH] [CI] Revert ModelOpt NVFP4 threshold relax (#29844)

---
 .github/workflows/pr-test-multimodal-gen.yml  |  2 ++
 .../server/consistency_thresholds/h100.json   | 12 ---------
 scripts/ci/cuda/ci_install_dependency.sh      | 27 +++++++++++++++++++
 3 files changed, 29 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/pr-test-multimodal-gen.yml b/.github/workflows/pr-test-multimodal-gen.yml
index b831c06e3..cc280aa61 100644
--- a/.github/workflows/pr-test-multimodal-gen.yml
+++ b/.github/workflows/pr-test-multimodal-gen.yml
@@ -352,6 +352,8 @@ jobs:
 
       - name: Install dependencies
         timeout-minutes: 20
+        env:
+          SGLANG_CI_EARLY_LD_LIBRARY_PATH: "1"
         run: |
           CUSTOM_BUILD_SGL_KERNEL=${{inputs.sgl_kernel}} bash scripts/ci/cuda/ci_install_dependency.sh diffusion
 
diff --git a/python/sglang/multimodal_gen/test/server/consistency_thresholds/h100.json b/python/sglang/multimodal_gen/test/server/consistency_thresholds/h100.json
index d32fa4d4a..29705f212 100644
--- a/python/sglang/multimodal_gen/test/server/consistency_thresholds/h100.json
+++ b/python/sglang/multimodal_gen/test/server/consistency_thresholds/h100.json
@@ -19,12 +19,6 @@
             "psnr_threshold": 28.0,
             "mean_abs_diff_threshold": 6.3
         },
-        "flux2_modelopt_nvfp4_t2i": {
-            "clip_threshold": 0.98,
-            "ssim_threshold": 0.95,
-            "psnr_threshold": 17.0,
-            "mean_abs_diff_threshold": 4.0
-        },
         "ideogram4_nvfp4_t2i": {
             "clip_threshold": 0.98,
             "ssim_threshold": 0.95,
@@ -163,12 +157,6 @@
             "psnr_threshold": 27.1,
             "mean_abs_diff_threshold": 7.2
         },
-        "wan22_modelopt_nvfp4_t2v": {
-            "clip_threshold": 0.98,
-            "ssim_threshold": 0.95,
-            "psnr_threshold": 24.0,
-            "mean_abs_diff_threshold": 4.0
-        },
         "fastwan2_2_ti2v_5b": {
             "clip_threshold": 0.97,
             "ssim_threshold": 0.92,
diff --git a/scripts/ci/cuda/ci_install_dependency.sh b/scripts/ci/cuda/ci_install_dependency.sh
index a7cdd6535..8b87c9005 100755
--- a/scripts/ci/cuda/ci_install_dependency.sh
+++ b/scripts/ci/cuda/ci_install_dependency.sh
@@ -196,6 +196,27 @@ setup_pip_toolchain() {
     mark_step_done "${FUNCNAME[0]}"
 }
 
+remove_stale_cuda12_nvidia_wheels() {
+    if [ "$CU_MAJOR" != "13" ]; then
+        mark_step_done "${FUNCNAME[0]}"
+        return
+    fi
+
+    mapfile -t STALE_CUDA12_NVIDIA_WHEELS < <(
+        python3 -m pip list --format=freeze | sed -n 's/^\(nvidia-.*-cu12\)==.*/\1/p'
+    )
+    if [ ${#STALE_CUDA12_NVIDIA_WHEELS[@]} -eq 0 ]; then
+        echo "No stale CUDA 12 NVIDIA wheels found for ${CU_VERSION} job"
+        mark_step_done "${FUNCNAME[0]}"
+        return
+    fi
+
+    echo "Removing stale CUDA 12 NVIDIA wheels from ${CU_VERSION} job: ${STALE_CUDA12_NVIDIA_WHEELS[*]}"
+    $PIP_UNINSTALL_CMD "${STALE_CUDA12_NVIDIA_WHEELS[@]}" $PIP_UNINSTALL_SUFFIX
+
+    mark_step_done "${FUNCNAME[0]}"
+}
+
 uninstall_stale_flashinfer() {
     # Keep flashinfer packages if version matches to avoid re-downloading:
     # - flashinfer-cubin: 150+ MB
@@ -552,8 +573,14 @@ main() {
     install_apt_packages
     clean_site_packages
     setup_pip_toolchain
+    remove_stale_cuda12_nvidia_wheels
     uninstall_stale_flashinfer
     install_sglang
+    # Diffusion B200 CI imports torch inside install_sglang_kernel after removing
+    # stale CUDA 12 NVIDIA wheels, so opt into one early LD_LIBRARY_PATH refresh.
+    if [ "${SGLANG_CI_EARLY_LD_LIBRARY_PATH:-0}" = "1" ]; then
+        setup_ld_library_path
+    fi
     install_sglang_kernel
     install_sglang_router
     download_flashinfer_cache