fix: renaming test file and job names + skip blocking llama4 nightly (#16971)

.github/workflows/nightly-test-nvidia.yml

@@ -168,13 +168,16 @@ jobs:
         run: |
           IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
 
-      - name: Run test
+      - name: Run common 8-GPU model tests
-        timeout-minutes: 120
+        if: always()
+        timeout-minutes: 300
         env:
+          TRACE_BASE_URL: https://raw.githubusercontent.com/sglang-bot/sglang-ci-data/main/traces/${{ github.run_id }}
+          PERFETTO_RELAY_URL: ${{ vars.PERFETTO_RELAY_URL }}
           GPU_CONFIG: "8-gpu-b200"
         run: |
           cd test
-          python3 run_suite.py --hw cuda --suite nightly-8-gpu-b200 --nightly --continue-on-error --timeout-per-file 2400
+          IS_BLACKWELL=1 python3 run_suite.py --hw cuda --suite nightly-8-gpu-common --nightly --timeout-per-file=12000 --continue-on-error
 
   # Text model accuracy tests
   nightly-test-text-accuracy-2-gpu-runner:
@@ -368,8 +371,8 @@ jobs:
           cd test
           python3 run_suite.py --hw cuda --suite nightly-4-gpu-b200 --nightly --continue-on-error
 
-  # B200 Performance tests - 8 GPU
+  # Specialized B200 tests - 8 GPU, for specific backends and configs
-  nightly-test-perf-8-gpu-b200:
+  nightly-test-specialized-8-gpu-b200:
     if: github.repository == 'sgl-project/sglang' && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-perf-8-gpu-b200')
     runs-on: 8-gpu-b200
     env:
@@ -384,16 +387,13 @@ jobs:
         run: |
           IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
 
-      - name: Run common 8-GPU model tests
+      - name: Run test
-        if: always()
+        timeout-minutes: 120
-        timeout-minutes: 300
         env:
-          TRACE_BASE_URL: https://raw.githubusercontent.com/sglang-bot/sglang-ci-data/main/traces/${{ github.run_id }}
-          PERFETTO_RELAY_URL: ${{ vars.PERFETTO_RELAY_URL }}
           GPU_CONFIG: "8-gpu-b200"
         run: |
           cd test
-          IS_BLACKWELL=1 python3 run_suite.py --hw cuda --suite nightly-8-gpu-common --nightly --timeout-per-file=12000 --continue-on-error
+          python3 run_suite.py --hw cuda --suite nightly-8-gpu-b200 --nightly --continue-on-error --timeout-per-file 2400
 
   # Final check job
   check-all-jobs:
@@ -411,7 +411,7 @@ jobs:
       - nightly-test-multimodal-server-1-gpu
       - nightly-test-multimodal-server-2-gpu
       - nightly-test-perf-4-gpu-b200
-      - nightly-test-perf-8-gpu-b200
+      - nightly-test-specialized-8-gpu-b200
     runs-on: ubuntu-latest
     steps:
       - name: Check if any job failed

test/registered/8-gpu-models/test_deepseek_v31.py

@@ -12,7 +12,7 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 DEEPSEEK_V31_MODEL_PATH = "deepseek-ai/DeepSeek-V3.1"
 
 
-class TestDeepseekV31Unified(unittest.TestCase):
+class TestDeepseekV31(unittest.TestCase):
     """Unified test class for DeepSeek-V3.1 performance and accuracy.
 
     Two variants:

test/registered/8-gpu-models/test_deepseek_v32.py

@@ -27,7 +27,7 @@ GSM8K_BASELINE = 0.935
 GPQA_BASELINE = 0.835
 
 
-class TestDeepseekV32Unified(unittest.TestCase):
+class TestDeepseekV32(unittest.TestCase):
     """Unified test class for DeepSeek V3.2 performance and accuracy.
 
     Tests multiple variants with both performance and accuracy tests:

test/registered/8-gpu-models/test_glm_46.py

@@ -12,7 +12,7 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 GLM_4_6_MODEL_PATH = "zai-org/GLM-4.6"
 
 
-class TestGLM46Unified(unittest.TestCase):
+class TestGLM46(unittest.TestCase):
     """Unified test class for GLM-4.6 performance and accuracy.
 
     Single variant with simple TP=8 configuration.

test/registered/8-gpu-models/test_glm_46_fp8.py

@@ -12,7 +12,7 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 GLM_4_6_FP8_MODEL_PATH = "zai-org/GLM-4.6-FP8"
 
 
-class TestGLM46FP8Unified(unittest.TestCase):
+class TestGLM46FP8(unittest.TestCase):
     """Unified test class for GLM-4.6-FP8 performance and accuracy.
 
     Single variant with simple TP=8 configuration.

test/registered/8-gpu-models/test_kimi_k2.py

@@ -12,7 +12,7 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 KIMI_K2_THINKING_MODEL_PATH = "moonshotai/Kimi-K2-Thinking"
 
 
-class TestKimiK2Unified(unittest.TestCase):
+class TestKimiK2(unittest.TestCase):
     """Unified test class for Kimi-K2-Thinking performance and accuracy.
 
     Single variant with TP=8 + tool/reasoning parsers.

test/registered/8-gpu-models/test_llama4.py

@@ -12,7 +12,8 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 LLAMA4_MODEL_PATH = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
 
 
-class TestLlama4Unified(unittest.TestCase):
+@unittest.skip("Blocked: Missing HF token permission for Llama 4 model")
+class TestLlama4(unittest.TestCase):
     """Unified test class for Llama-4-Scout performance and accuracy.
 
     Llama4 has local attention mechanism with hybrid sliding window attention.

test/registered/8-gpu-models/test_minimax_m2.py

@@ -12,7 +12,7 @@ register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
 MINIMAX_M2_MODEL_PATH = "MiniMaxAI/MiniMax-M2"
 
 
-class TestMiniMaxM2Unified(unittest.TestCase):
+class TestMiniMaxM2(unittest.TestCase):
     """Unified test class for MiniMax-M2 performance and accuracy.
 
     Single variant with TP=8 + EP=8 configuration.

test/registered/8-gpu-models/test_mistral_large3.py

@@ -16,7 +16,7 @@ MISTRAL_LARGE3_EAGLE_MODEL_PATH = "mistralai/Mistral-Large-3-675B-Instruct-2512-
 
 
 @unittest.skipIf(not is_blackwell_system(), "Requires B200")
-class TestMistralLarge3Unified(unittest.TestCase):
+class TestMistralLarge3(unittest.TestCase):
     """Unified test class for Mistral-Large-3 performance and accuracy.
 
     Two variants:

test/registered/8-gpu-models/test_qwen3_235b.py

@@ -13,7 +13,7 @@ QWEN3_235B_MODEL_PATH = "Qwen/Qwen3-235B-A22B-Instruct-2507"
 
 
 @unittest.skipIf(not is_blackwell_system(), "Requires B200")
-class TestQwen3235BUnified(unittest.TestCase):
+class TestQwen3235B(unittest.TestCase):
     """Unified test class for Qwen3-235B performance and accuracy.
 
     Single variant with simple TP=8 configuration.