From 9e629d31fd93f7c4cf0fbee4632f150c8ba62c4b Mon Sep 17 00:00:00 2001
From: YC Yen-Ching Tseng <yctseng@amd.com>
Date: Fri, 20 Mar 2026 02:50:19 +0800
Subject: [PATCH] [AMD] CI - Fix AMD CI (multimodal test, move flaky test to
 non-deterministic group) (#20815)

---
 .github/workflows/pr-test-amd-rocm720.yml       | 4 ++--
 .github/workflows/pr-test-amd.yml               | 6 +++---
 test/registered/lora/test_multi_lora_backend.py | 2 +-
 test/registered/models/test_vlm_models.py       | 2 +-
 4 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/pr-test-amd-rocm720.yml b/.github/workflows/pr-test-amd-rocm720.yml
index 1ea7913a0..fd78f9a8b 100644
--- a/.github/workflows/pr-test-amd-rocm720.yml
+++ b/.github/workflows/pr-test-amd-rocm720.yml
@@ -533,7 +533,7 @@ jobs:
       max-parallel: 1  # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT
       matrix:
         runner: [linux-mi325-1gpu-sglang]
-        part: [0, 1]  # 2 partitions: 11 tests ÷ 2 = ~5-6 tests each
+        part: [0, 1, 2, 3]
     runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
@@ -639,7 +639,7 @@ jobs:
             ci_sglang python3 sglang/multimodal_gen/test/run_suite.py \
               --suite 1-gpu \
               --partition-id ${{ matrix.part }} \
-              --total-partitions 2 \
+              --total-partitions 4 \
               -k "not flux_2"
 
           # Post-test diagnostics
diff --git a/.github/workflows/pr-test-amd.yml b/.github/workflows/pr-test-amd.yml
index 246d294a3..3e821b8d8 100644
--- a/.github/workflows/pr-test-amd.yml
+++ b/.github/workflows/pr-test-amd.yml
@@ -539,7 +539,7 @@ jobs:
       max-parallel: 1  # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT
       matrix:
         runner: [linux-mi325-1gpu-sglang]
-        part: [0, 1]  # 2 partitions: 11 tests ÷ 2 = ~5-6 tests each
+        part: [0, 1, 2, 3]  # 2 partitions: 11 tests ÷ 2 = ~5-6 tests each
     runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout code
@@ -621,7 +621,7 @@ jobs:
           docker exec ci_sglang rocm-smi --showmeminfo vram 2>/dev/null || echo "rocm-smi not available"
 
       - name: Run diffusion server tests (1-GPU)
-        timeout-minutes: 70
+        timeout-minutes: 90
         run: |
           # AMD CI: All 1-GPU tests except FLUX.2 (FLUX.1 covers same code path)
           # Tests: T2V, T2I, I2V, LoRA
@@ -644,7 +644,7 @@ jobs:
             ci_sglang python3 sglang/multimodal_gen/test/run_suite.py \
               --suite 1-gpu \
               --partition-id ${{ matrix.part }} \
-              --total-partitions 2 \
+              --total-partitions 4 \
               -k "not flux_2"
 
           # Post-test diagnostics
diff --git a/test/registered/lora/test_multi_lora_backend.py b/test/registered/lora/test_multi_lora_backend.py
index 9a7465d45..f34b9e622 100644
--- a/test/registered/lora/test_multi_lora_backend.py
+++ b/test/registered/lora/test_multi_lora_backend.py
@@ -26,7 +26,7 @@ from sglang.test.lora_utils import (
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
 register_cuda_ci(est_time=100, suite="stage-b-test-large-1-gpu")
-register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu-amd-nondeterministic")
+register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestMultiLoRABackend(CustomTestCase):
diff --git a/test/registered/models/test_vlm_models.py b/test/registered/models/test_vlm_models.py
index ffcba0acb..a2c6cabc9 100644
--- a/test/registered/models/test_vlm_models.py
+++ b/test/registered/models/test_vlm_models.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import is_in_ci
 
 
 register_cuda_ci(est_time=228, suite="stage-b-test-large-1-gpu")
-register_amd_ci(est_time=850, suite="stage-b-test-small-1-gpu-amd")
+register_amd_ci(est_time=850, suite="stage-b-test-small-1-gpu-amd-nondeterministic")
 
 _is_hip = is_hip()
 # VLM models for testing