Fix 2-gpu CI test and suppress some warnings (#4930)

python/sglang/srt/models/deepseek_v2.py

@@ -30,9 +30,6 @@ from sglang.srt.distributed import (
     tensor_model_parallel_all_reduce,
 )
 from sglang.srt.layers.activation import SiluAndMul
-from sglang.srt.layers.attention.triton_ops.rocm_mla_decode_rope import (
-    decode_attention_fwd_grouped_rope,
-)
 from sglang.srt.layers.dp_attention import (
     dp_gather_partial,
     dp_scatter,
@@ -83,6 +80,11 @@ if _is_cuda:
 else:
     from vllm import _custom_ops as ops
 
+if _is_hip:
+    from sglang.srt.layers.attention.triton_ops.rocm_mla_decode_rope import (
+        decode_attention_fwd_grouped_rope,
+    )
+
 expert_distribution_recorder = ExpertDistributionRecorder()
 
 

python/sglang/srt/utils.py

@@ -564,6 +564,10 @@ def load_image(image_file: Union[str, bytes]) -> tuple[Image, tuple[int, int]]:
 
 
 def suppress_other_loggers():
+    warnings.filterwarnings(
+        "ignore", category=UserWarning, message="The given NumPy array is not writable"
+    )
+
     try:
         from vllm.logger import logger as vllm_default_logger
     except ImportError:
@@ -578,10 +582,6 @@ def suppress_other_loggers():
     )
     logging.getLogger("vllm.config").setLevel(logging.ERROR)
 
-    warnings.filterwarnings(
-        "ignore", category=UserWarning, message="The given NumPy array is not writable"
-    )
-
 
 def assert_pkg_version(pkg: str, min_version: str, message: str):
     try:

test/srt/run_suite.py

@@ -15,7 +15,7 @@ suites = {
     "per-commit": [
         TestFile("models/lora/test_lora.py", 76),
         TestFile("models/lora/test_lora_backend.py", 420),
-        TestFile("models/lora/test_multi_lora_backend.py", 144),
+        TestFile("models/lora/test_multi_lora_backend.py", 60),
         TestFile("models/test_embedding_models.py", 35),
         TestFile("models/test_generation_models.py", 103),
         TestFile("models/test_grok_models.py", 60),
@@ -28,15 +28,15 @@ suites = {
         TestFile("test_chunked_prefill.py", 336),
         TestFile("test_eagle_infer.py", 500),
         TestFile("test_ebnf_constrained.py"),
-        TestFile("test_fp8_kernel.py", 2),
+        TestFile("test_fp8_kernel.py", 8),
         TestFile("test_embedding_openai_server.py", 36),
         TestFile("test_hidden_states.py", 55),
-        TestFile("test_int8_kernel.py", 1),
+        TestFile("test_int8_kernel.py", 8),
         TestFile("test_input_embeddings.py", 38),
         TestFile("test_json_constrained.py", 98),
         TestFile("test_large_max_new_tokens.py", 41),
         TestFile("test_metrics.py", 32),
-        TestFile("test_mla.py", 92),
+        TestFile("test_mla.py", 162),
         TestFile("test_mla_deepseek_v3.py", 221),
         TestFile("test_mla_int8_deepseek_v3.py", 522),
         TestFile("test_mla_flashinfer.py", 395),
@@ -68,23 +68,23 @@ suites = {
         TestFile("test_vertex_endpoint.py", 31),
         TestFile("test_vision_chunked_prefill.py", 223),
         TestFile("test_vlm_accuracy.py", 60),
-        TestFile("test_vision_openai_server.py", 344),
-        TestFile("test_fim_completion.py", 120),
+        TestFile("test_vision_openai_server.py", 537),
+        TestFile("test_fim_completion.py", 40),
         TestFile("test_w8a8_quantization.py", 46),
-        TestFile("test_eval_fp8_accuracy.py", 172),
+        TestFile("test_eval_fp8_accuracy.py", 303),
         TestFile("test_create_kvindices.py", 2),
         TestFile("test_hicache.py", 60),
         TestFile("test_hicache_mla.py", 90),
     ],
     "per-commit-2-gpu": [
+        TestFile("models/lora/test_lora_tp.py", 300),
         TestFile("test_data_parallelism.py", 90),
         TestFile("test_dp_attention.py", 90),
+        TestFile("test_mla_tp.py", 420),
+        TestFile("test_moe_ep.py", 220),
+        TestFile("test_patch_torch.py", 30),
         TestFile("test_update_weights_from_distributed.py", 100),
         TestFile("test_verl_engine.py", 100),
-        TestFile("test_patch_torch.py", 30),
-        TestFile("test_moe_ep.py", 220),
-        TestFile("test_mla_tp.py", 420),
-        TestFile("test_lora_tp.py", 300),
     ],
     "nightly": [
         TestFile("test_nightly_gsm8k_eval.py"),

test/srt/test_eagle_infer.py

@@ -567,7 +567,7 @@ class TestEAGLEServerPageSize(TestEAGLEServer):
                 "--max-running-requests",
                 8,
                 "--page-size",
-                4,
+                8,
             ],
         )