Retire the per-runner parallel record (#40343)

test/manual/test_forward_split_prefill.py

@@ -15,7 +15,6 @@ import torch
 
 from sglang.benchmark.one_batch import TreeCacheNamespace
 from sglang.srt.configs.model_config import ModelConfig
-from sglang.srt.distributed.parallel_state_wrapper import ParallelState
 from sglang.srt.managers.schedule_batch import Req, ScheduleBatch
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode
 from sglang.srt.model_executor.forward_context import (
@@ -65,7 +64,6 @@ class TestForwardSplitPrefill(CustomTestCase):
             model_config=cls.model_config,
             mem_fraction_static=cls.server_args.mem_fraction_static,
             gpu_id=0,
-            ps=ParallelState.trivial(tp_size=cls.tp_size),
             nccl_port=cls.port_args.nccl_port,
             server_args=cls.server_args,
         )

test/manual/test_vlm_accuracy.py

@@ -9,7 +9,6 @@ import torch.nn.functional as F
 from transformers import AutoModel, AutoProcessor, AutoTokenizer
 
 from sglang.srt.configs.model_config import ModelConfig
-from sglang.srt.distributed.parallel_state_wrapper import ParallelState
 from sglang.srt.entrypoints.openai.protocol import ChatCompletionRequest
 from sglang.srt.managers.mm_utils import embed_mm_inputs, init_mm_embedding_cache
 from sglang.srt.managers.schedule_batch import (
@@ -151,7 +150,6 @@ class VisionLLMLogitsBase(unittest.IsolatedAsyncioTestCase):
             model_config=ModelConfig(self.model_path, model_override_args="{}"),
             mem_fraction_static=0.8,
             gpu_id=0,
-            ps=ParallelState.trivial(),
             nccl_port=12435,
             server_args=server_args,
         )