[AMD] Fix CI - unit-test-backend-1-gpu-amd-mi35x and unit-test-backend-2-gpu-amd, stage-b-test-small-1-gpu-amd (#16675)

Co-authored-by: Bingxu Chen <bingxche@amd.com>

.github/workflows/pr-test-amd.yml

@@ -224,7 +224,46 @@ jobs:
       - name: Run test
         timeout-minutes: 30
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 12
+          bash scripts/ci/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 12
+
+  stage-b-test-small-1-gpu-amd-mi35x:
+    needs: [check-changes, stage-a-test-1-amd]
+    if: |
+      always() &&
+      (
+        (inputs.target_stage == 'stage-b-test-small-1-gpu-amd-mi35x') ||
+        (
+          !inputs.target_stage &&
+          (!failure() && !cancelled()) &&
+          ((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
+        )
+      )
+    strategy:
+      fail-fast: false
+      matrix:
+        runner: [linux-mi35x-gpu-1]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.pr_head_sha || inputs.ref || github.sha }}
+
+      - name: Ensure VRAM is clear
+        run: bash scripts/ensure_vram_clear.sh rocm
+
+      - name: Start CI container
+        run: bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: bash scripts/ci/amd_ci_install_dependency.sh
+
+      - name: Run test
+        timeout-minutes: 30
+        run: |
+          bash scripts/ci/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu-amd-mi35x
 
   stage-b-test-large-2-gpu-amd:
     needs: [check-changes, stage-a-test-1-amd]
@@ -545,87 +584,8 @@ jobs:
         run: |
           bash scripts/ci/amd_ci_exec.sh python3 run_suite.py --suite per-commit-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
-  unit-test-backend-1-gpu-amd-mi35x:
-    needs: [check-changes, stage-a-test-1-amd]
-    if: |
-      always() &&
-      (
-        (inputs.target_stage == 'unit-test-backend-1-gpu-amd-mi35x') ||
-        (
-          !inputs.target_stage &&
-          (!failure() && !cancelled()) &&
-          ((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
-        )
-      )
-    strategy:
-      fail-fast: false
-      matrix:
-        runner: [linux-mi35x-gpu-1]
-    runs-on: ${{matrix.runner}}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.pr_head_sha || inputs.ref || github.sha }}
-
-      - name: Ensure VRAM is clear
-        run: bash scripts/ensure_vram_clear.sh rocm
-
-      - name: Start CI container
-        run: bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: bash scripts/ci/amd_ci_install_dependency.sh
-
-      - name: Run test
-        timeout-minutes: 15
-        run: |
-          bash scripts/ci/amd_ci_exec.sh python3 run_suite.py --suite per-commit-amd-mi35x
-
-  unit-test-backend-2-gpu-amd:
-    needs: [check-changes, stage-a-test-1-amd]
-    if: |
-      always() &&
-      (
-        (inputs.target_stage == 'unit-test-backend-2-gpu-amd') ||
-        (
-          !inputs.target_stage &&
-          (!failure() && !cancelled()) &&
-          ((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
-        )
-      )
-    strategy:
-      fail-fast: false
-      matrix:
-        runner: [linux-mi325-gpu-2]
-        part: [0, 1]
-    runs-on: ${{matrix.runner}}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.pr_head_sha || inputs.ref || github.sha }}
-
-      - name: Ensure VRAM is clear
-        run: bash scripts/ensure_vram_clear.sh rocm
-
-      - name: Start CI container
-        run: bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: bash scripts/ci/amd_ci_install_dependency.sh
-
-      - name: Run test
-        timeout-minutes: 30
-        run: |
-          bash scripts/ci/amd_ci_exec.sh python3 run_suite.py --suite per-commit-2-gpu-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
-
   unit-test-backend-8-gpu-amd:
-    needs: [check-changes, unit-test-backend-2-gpu-amd]
+    needs: [check-changes, stage-a-test-1-amd]
     if: |
       always() &&
       (
@@ -673,7 +633,7 @@ jobs:
           bash scripts/ci/amd_ci_exec.sh python3 run_suite.py --suite per-commit-8-gpu-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 2 --timeout-per-file 3600
 
   unit-test-backend-8-gpu-amd-mi35x:
-    needs: [check-changes, unit-test-backend-2-gpu-amd]
+    needs: [check-changes, stage-a-test-1-amd]
     if: always() && !failure() && !cancelled() &&
       ((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
     strategy:
@@ -806,7 +766,7 @@ jobs:
           bash scripts/ci/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_default_fp8
 
   performance-test-2-gpu-amd:
-    needs: [check-changes, unit-test-backend-2-gpu-amd]
+    needs: [check-changes, stage-a-test-1-amd]
     if: |
       always() &&
       (
@@ -965,10 +925,9 @@ jobs:
 
         stage-a-test-1-amd,
         stage-b-test-small-1-gpu-amd,
+        stage-b-test-small-1-gpu-amd-mi35x,
         stage-b-test-large-2-gpu-amd,
         unit-test-backend-1-gpu-amd,
-        unit-test-backend-1-gpu-amd-mi35x,
-        unit-test-backend-2-gpu-amd,
         unit-test-backend-8-gpu-amd,
         unit-test-backend-8-gpu-amd-mi35x,
         performance-test-1-gpu-part-1-amd,

scripts/ci/slash_command_handler.py

@@ -177,6 +177,7 @@ def handle_rerun_stage(
         "sgl-kernel-unit-test-amd",
         "stage-a-test-1-amd",
         "stage-b-test-small-1-gpu-amd",
+        "stage-b-test-small-1-gpu-amd-mi35x",
         "stage-b-test-large-2-gpu-amd",
         "unit-test-backend-1-gpu-amd",
         "unit-test-backend-2-gpu-amd",

test/registered/attention/test_create_kvindices.py

@@ -9,7 +9,7 @@ from sglang.test.test_utils import CustomTestCase
 
 # Triton kernel unit test for KV indices creation
 register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestCreateKvIndices(CustomTestCase):

test/registered/attention/test_radix_attention.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 
 # RadixAttention server integration tests
 register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestRadixCacheFCFS(CustomTestCase):

test/registered/attention/test_swa_unittest.py

@@ -10,7 +10,7 @@ from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=8, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestSWA(unittest.TestCase):

test/registered/attention/test_torch_native_attention_backend.py

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
 
 # Torch native attention backend integration test with MMLU eval
 register_cuda_ci(est_time=150, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestTorchNativeAttnBackend(CustomTestCase):

test/registered/attention/test_triton_attention_backend.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
 
 # Triton attention backend integration test with latency benchmark and MMLU eval
 register_cuda_ci(est_time=200, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=1110, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=1110, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestTritonAttnBackend(CustomTestCase):

test/registered/attention/test_triton_attention_kernels.py

@@ -26,7 +26,7 @@ from sglang.test.test_utils import CustomTestCase
 register_cuda_ci(est_time=30, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=30,
-    suite="stage-b-test-small-1-gpu",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="test was never enabled for AMD CI, needs validation",
 )
 

test/registered/attention/test_triton_sliding_window.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 
 # Sliding window attention with Triton backend (Gemma-3 model)
 register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=100, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestSlidingWindowAttentionTriton(CustomTestCase):

test/registered/backends/test_torch_compile.py

@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=190, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=1100, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=1100, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestTorchCompile(CustomTestCase):

test/registered/constrained_decoding/test_constrained_decoding.py

@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=111, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=179, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=179, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 

test/registered/core/test_gpt_oss_1gpu.py

@@ -4,7 +4,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
 register_cuda_ci(est_time=402, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=750, suite="stage-b-test-small-1-gpu-amd")
+register_amd_ci(est_time=750, suite="stage-b-test-small-1-gpu-amd-mi35x")
 
 
 class TestGptOss1Gpu(BaseTestGptOss):

test/registered/debug_utils/test_tensor_dump_forward_hook.py

@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=9, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 

test/registered/dllm/test_llada2_mini_amd.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci
 
-register_amd_ci(est_time=520, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=520, suite="stage-b-test-small-1-gpu-amd")
 
 """
 Test LLaDA2 (Diffusion Language Model) on AMD GPUs.

test/registered/lora/test_lora.py

@@ -25,7 +25,7 @@ from sglang.test.lora_utils import (
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
 register_cuda_ci(est_time=82, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=82, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=82, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestLoRA(CustomTestCase):

test/registered/lora/test_lora_backend.py

@@ -32,7 +32,7 @@ from sglang.test.test_utils import CustomTestCase, is_in_ci
 register_cuda_ci(est_time=200, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=200,
-    suite="stage-b-test-small-1-gpu",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/13107",
 )
 

test/registered/lora/test_lora_eviction.py

@@ -24,7 +24,7 @@ from sglang.test.runners import SRTRunner
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=224, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=224, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=224, suite="stage-b-test-small-1-gpu-amd")
 
 PROMPTS = [
     "AI is a field of computer science focused on",

test/registered/lora/test_multi_lora_backend.py

@@ -25,7 +25,7 @@ from sglang.test.lora_utils import (
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
 register_cuda_ci(est_time=60, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
 
 # All prompts are used at once in a batch.
 PROMPTS = [

test/registered/metrics/test_metrics.py

@@ -6,7 +6,7 @@ import requests
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=32, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=32, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=32, suite="stage-b-test-small-1-gpu-amd")
 from prometheus_client.parser import text_string_to_metric_families
 from prometheus_client.samples import Sample
 

test/registered/mla/test_mla.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 register_cuda_ci(est_time=194, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=242,
-    suite="stage-a-test-1",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/13107",
 )
 

test/registered/mla/test_mla_deepseek_v3.py

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
 register_cuda_ci(est_time=442, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=221,
-    suite="stage-a-test-1",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/12574",
 )
 

test/registered/models/test_compressed_tensors_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Model tests for compressed tensors (FP8)
 register_cuda_ci(est_time=42, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=42, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=42, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 from types import SimpleNamespace

test/registered/models/test_cross_encoder_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Cross encoder model tests
 register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=150, suite="stage-b-test-small-1-gpu-amd")
 
 import multiprocessing as mp
 import random

test/registered/models/test_embedding_models.py

@@ -4,7 +4,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 register_cuda_ci(est_time=73, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=73,
-    suite="stage-b-test-small-1-gpu",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/11127",
 )
 

test/registered/models/test_qwen_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Qwen model tests
 register_cuda_ci(est_time=90, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=130, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=130, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 from types import SimpleNamespace

test/registered/models/test_reward_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Reward model tests
 register_cuda_ci(est_time=103, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=132, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=132, suite="stage-b-test-small-1-gpu-amd")
 
 # Copyright 2023-2024 SGLang Team
 # Licensed under the Apache License, Version 2.0 (the "License");

test/registered/models/test_transformers_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Transformers fallback model tests
 register_cuda_ci(est_time=245, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=320, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=320, suite="stage-b-test-small-1-gpu-amd")
 
 import dataclasses
 import multiprocessing as mp

test/registered/models/test_vlm_models.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # VLM (Vision Language Model) tests
 register_cuda_ci(est_time=270, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=420, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=420, suite="stage-b-test-small-1-gpu-amd")
 
 import argparse
 import random

test/registered/moe/test_fused_moe.py

@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=80, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=30, suite="stage-a-test-1")
+register_amd_ci(est_time=30, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 

test/registered/openai_server/basic/test_openai_embedding.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=70, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=141, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=141, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestOpenAIEmbedding(CustomTestCase):

test/registered/openai_server/basic/test_openai_server.py

@@ -29,7 +29,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=184, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=149, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=149, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestOpenAIServer(CustomTestCase):

test/registered/openai_server/basic/test_protocol.py

@@ -31,7 +31,7 @@ from sglang.srt.entrypoints.openai.protocol import (
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=3, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestModelCard(unittest.TestCase):

test/registered/openai_server/basic/test_serving_chat.py

@@ -24,7 +24,7 @@ from sglang.srt.utils import get_or_create_event_loop
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 class _MockTokenizerManager:

test/registered/openai_server/basic/test_serving_completions.py

@@ -14,7 +14,7 @@ from sglang.srt.managers.tokenizer_manager import TokenizerManager
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 class _MockTemplateManager:

test/registered/openai_server/basic/test_serving_embedding.py

@@ -17,7 +17,7 @@ from sglang.srt.managers.io_struct import EmbeddingReqInput
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
 
 
 # Mock TokenizerManager for embedding tests

test/registered/openai_server/features/test_enable_thinking.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=70, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=200, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=200, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestEnableThinking(CustomTestCase):

test/registered/openai_server/features/test_json_mode.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=109, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=180, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=180, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestJSONModeMixin:

test/registered/openai_server/features/test_openai_server_ebnf.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=7, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=20, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=20, suite="stage-b-test-small-1-gpu-amd")
 
 
 # -------------------------------------------------------------------------

test/registered/openai_server/features/test_openai_server_hidden_states.py

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
 register_cuda_ci(est_time=186, suite="stage-b-test-small-1-gpu")
 register_amd_ci(
     est_time=186,
-    suite="stage-b-test-small-1-gpu",
+    suite="stage-b-test-small-1-gpu-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/11127",
 )
 

test/registered/openai_server/features/test_reasoning_content.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=89, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=89, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=89, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestReasoningContentAPI(CustomTestCase):

test/registered/openai_server/function_call/test_openai_function_calling.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=60, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=73, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=73, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestOpenAIServerFunctionCalling(CustomTestCase):

test/registered/openai_server/function_call/test_tool_choice.py

@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=120, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=258, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=258, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestToolChoiceLlama32(CustomTestCase):

test/registered/openai_server/validation/test_large_max_new_tokens.py

@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=41, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=41, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=41, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestLargeMaxNewTokens(CustomTestCase):

test/registered/openai_server/validation/test_matched_stop.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=40, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestMatchedStop(CustomTestCase, MatchedStopMixin):

test/registered/openai_server/validation/test_openai_server_ignore_eos.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=6, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=47, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=47, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestOpenAIServerIgnoreEOS(CustomTestCase):

test/registered/openai_server/validation/test_request_length_validation.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=38, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=31, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=31, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestRequestLengthValidation(CustomTestCase):

test/registered/ops/test_repeat_interleave.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Ops - Repeat Interleave tests (1-GPU)
 register_cuda_ci(est_time=60, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=75, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=75, suite="stage-b-test-small-1-gpu-amd")
 
 import time
 from typing import Tuple

test/registered/profiling/test_start_profile.py

@@ -30,7 +30,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=41, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
 
 OUTPUT_DIR = "./profiler_dir"
 

test/registered/quant/test_block_int8.py

@@ -11,7 +11,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=44, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=22, suite="stage-a-test-1")
+register_amd_ci(est_time=22, suite="stage-b-test-small-1-gpu-amd")
 
 
 # For test

test/registered/quant/test_eval_fp8_accuracy.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=250, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=303, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=303, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestEvalFP8Accuracy(CustomTestCase):

test/registered/quant/test_triton_scaled_mm.py

@@ -9,7 +9,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=8, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=12, suite="stage-a-test-1")
+register_amd_ci(est_time=12, suite="stage-b-test-small-1-gpu-amd")
 
 
 def torch_scaled_mm(

test/registered/rl/test_fp32_lm_head.py

@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=9, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 from types import SimpleNamespace

test/registered/rl/test_update_weights_from_disk.py

@@ -1,7 +1,9 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=210, suite="stage-b-test-small-1-gpu", disabled="see #14021")
-register_amd_ci(est_time=210, suite="stage-b-test-small-1-gpu", disabled="see #14021")
+register_amd_ci(
+    est_time=210, suite="stage-b-test-small-1-gpu-amd", disabled="see #14021"
+)
 
 import json
 import random

test/registered/rotary/test_mrope.py

@@ -2,7 +2,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Rotary Embedding - MRoPE tests (1-GPU)
 register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=15, suite="stage-b-test-small-1-gpu-amd")
 
 from typing import NamedTuple
 

test/registered/sampling/test_original_logprobs.py

@@ -18,7 +18,7 @@ import os
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=41, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
 import random
 import unittest
 

test/registered/sampling/test_penalty.py

@@ -10,7 +10,7 @@ from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=82, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=82, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=82, suite="stage-b-test-small-1-gpu-amd")
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,

test/registered/sampling/test_pytorch_sampling_backend.py

@@ -7,7 +7,7 @@ from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=66, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=66, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=66, suite="stage-b-test-small-1-gpu-amd")
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_TEST,

test/registered/tokenizer/test_multi_tokenizer.py

@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=230, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=345, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=345, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestMultiTokenizer(CustomTestCase):

test/registered/tokenizer/test_skip_tokenizer_init.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=77, suite="stage-b-test-small-1-gpu")
-register_amd_ci(est_time=117, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=117, suite="stage-b-test-small-1-gpu-amd")
 
 
 class TestSkipTokenizerInit(CustomTestCase):

test/run_suite.py

@@ -20,7 +20,8 @@ PER_COMMIT_SUITES = {
     HWBackend.CPU: ["default", "stage-a-cpu-only"],
     HWBackend.AMD: [
         "stage-a-test-1",
-        "stage-b-test-small-1-gpu",
+        "stage-b-test-small-1-gpu-amd",
+        "stage-b-test-small-1-gpu-amd-mi35x",
         "stage-b-test-large-2-gpu-amd",
     ],
     HWBackend.CUDA: [