[TRTLLM-8933][chore] remove unused update_executor_config function (#8678)

Signed-off-by: junq <[email protected]>

Author: QiJune

tensorrt_llm/_torch/pyexecutor/config.py

@@ -3,11 +3,8 @@
 
 from tensorrt_llm._torch.models.checkpoints.base_checkpoint_loader import \
     BaseCheckpointLoader
-from tensorrt_llm.bindings.executor import ExecutorConfig
 
 from ...llmapi.llm_args import LoadFormat, SamplerType
-from ...logger import logger
-from ...mapping import Mapping
 from ..model_config import MoeLoadBalancerConfig
 from .resource_manager import BaseResourceManager
 
@@ -117,57 +114,6 @@ class PyTorchConfig:
     _limit_torch_cuda_mem_fraction: bool = True
 
 
-EXETENDED_EXECUTOR_CONFIG_FIELDS = [
-    'backend',
-    'pytorch_backend_config',
-    'max_seq_len',
-    'mapping',
-    'hf_model_dir',
-    'mm_encoder_only',
-]
-
-
-def update_executor_config(
-        executor_config: ExecutorConfig,
-        backend: Optional[str] = None,
-        pytorch_backend_config: Optional[PyTorchConfig] = None,
-        mapping: Optional[Mapping] = None,
-        speculative_config: Optional["DecodingBaseConfig"] = None,
-        hf_model_dir: Optional[str] = None,
-        max_input_len: Optional[int] = None,
-        max_seq_len: Optional[int] = None,
-        checkpoint_format: Optional[str] = None,
-        checkpoint_loader: Optional[BaseCheckpointLoader] = None,
-        mm_encoder_only: bool = False):
-    if backend is None:
-        return
-
-    for field_name in EXETENDED_EXECUTOR_CONFIG_FIELDS:
-        if hasattr(executor_config, field_name):
-            raise AttributeError(
-                f"{field_name} should be dynamically assigned.")
-        setattr(executor_config, field_name, None)
-
-    executor_config.backend = backend
-    executor_config.pytorch_backend_config = pytorch_backend_config
-    executor_config.mapping = mapping
-    executor_config.speculative_config = speculative_config
-    executor_config.mm_encoder_only = mm_encoder_only
-
-    logger.info(f"{executor_config.pytorch_backend_config}")
-
-    executor_config.hf_model_dir = hf_model_dir
-
-    if max_input_len is not None:
-        executor_config.max_input_len = max_input_len
-
-    if max_seq_len is not None:
-        executor_config.max_seq_len = max_seq_len
-
-    executor_config.checkpoint_loader = _construct_checkpoint_loader(
-        backend, checkpoint_loader, checkpoint_format)
-
-
 def _construct_checkpoint_loader(
         backend: str, checkpoint_loader: Optional[BaseCheckpointLoader],
         checkpoint_format: Optional[str]) -> Optional[BaseCheckpointLoader]:

tests/unittest/executor/test_base_worker.py

@@ -6,7 +6,6 @@
 import torch
 
 from tensorrt_llm._utils import mpi_comm, mpi_rank, mpi_world_size
-from tensorrt_llm.bindings import executor as tllm
 from tensorrt_llm.llmapi.mpi_session import MpiPoolSession
 
 # isort: off
@@ -15,7 +14,6 @@
 from utils.util import skip_single_gpu
 # isort: on
 
-from tensorrt_llm._torch.pyexecutor.config import update_executor_config
 from tensorrt_llm.executor.base_worker import BaseWorker
 from tensorrt_llm.executor.request import GenerationRequest
 from tensorrt_llm.llmapi.llm_args import TorchLlmArgs
@@ -28,12 +26,16 @@
 class FakeWorker(BaseWorker):
 
     def __init__(self, engine: str, tp_size: int = 1):
-        llm_args, executor_config = create_fake_executor_config(engine, tp_size)
+        llm_args = TorchLlmArgs(
+            model=model_path,
+            tensor_parallel_size=tp_size,
+            backend='pytorch',
+            enable_iter_perf_stats=True,
+        )
         super().__init__(
             engine=engine,
             llm_args=llm_args,
             hf_model_dir=engine,
-            executor_config=executor_config,
         )
         # Note: BaseWorker doesn't call setup_engine() automatically,
         # unlike GenerationExecutorWorker, so we need to call it manually
@@ -116,34 +118,6 @@ def test_fetch_responses_timeout(self, timeout: float):
             assert timeout / 2 <= elapsed <= timeout * 2, f"Latency out of expected range: {elapsed}"
 
 
-def create_fake_executor_config(model_path, tp_size=1):
-    # Use TorchLlmArgs for PyTorch backend tests
-    llm_args = TorchLlmArgs(
-        model=model_path,
-        tensor_parallel_size=tp_size,
-        backend='pytorch',
-        enable_iter_perf_stats=True,
-    )
-
-    executor_config = tllm.ExecutorConfig(1)
-    executor_config.max_batch_size = 1
-    executor_config.model_world_size = tp_size
-
-    update_executor_config(
-        executor_config,
-        pytorch_backend_config=llm_args.get_pytorch_backend_config(),
-        mapping=llm_args.parallel_config.to_mapping(),
-        speculative_config=llm_args.speculative_config,
-        hf_model_dir=model_path,
-        max_input_len=20,
-        max_seq_len=40,
-        checkpoint_format=llm_args.checkpoint_format,
-        checkpoint_loader=llm_args.checkpoint_loader,
-    )
-
-    return llm_args, executor_config
-
-
 class TestRpcWorkerBaseTP2:
 
     def setup_method(self):

tests/unittest/executor/test_rpc_worker.py

@@ -6,12 +6,12 @@
 from concurrent.futures import ProcessPoolExecutor
 
 import pytest
-from test_base_worker import create_fake_executor_config
 
 from tensorrt_llm.executor.request import GenerationRequest
 from tensorrt_llm.executor.rpc import RPCClient
 from tensorrt_llm.executor.rpc.rpc_common import get_unique_ipc_addr
 from tensorrt_llm.executor.rpc_worker import RpcWorker
+from tensorrt_llm.llmapi.llm_args import TorchLlmArgs
 from tensorrt_llm.llmapi.mpi_session import MpiPoolSession
 from tensorrt_llm.sampling_params import SamplingParams
 
@@ -28,8 +28,12 @@
 class TestRpcWorkerTP1:
 
     def setup_method(self):
-        self.llm_args, self.executor_config = create_fake_executor_config(
-            model_path)
+        self.llm_args = TorchLlmArgs(
+            model=model_path,
+            tensor_parallel_size=1,
+            backend='pytorch',
+            enable_iter_perf_stats=True,
+        )
         self.pool, self.addr = self.create_worker_pool()
         self.client = self.create_rpc_client(self.addr)
         self.client.setup_engine().remote()
@@ -50,7 +54,6 @@ def create_worker_pool(self):
             RpcWorker.main_task,
             engine=model_path,
             rpc_addr=addr,
-            executor_config=self.executor_config,
             llm_args=self.llm_args,
             hf_model_dir=model_path,
         )
@@ -200,8 +203,12 @@ async def consume_stats():
 class TestRpcWorkerTP2:
 
     def setup_method(self):
-        self.llm_args, self.executor_config = create_fake_executor_config(
-            model_path, tp_size=2)
+        self.llm_args = TorchLlmArgs(
+            model=model_path,
+            tensor_parallel_size=2,
+            backend='pytorch',
+            enable_iter_perf_stats=True,
+        )
         self.session, self.addr, self.futures = self.create_worker_session()
         self.client = self.create_rpc_client(self.addr)
         self.client.setup_engine().remote()
@@ -218,7 +225,6 @@ def create_worker_session(self):
         futures = session.submit(RpcWorker.main_task,
                                  engine=model_path,
                                  rpc_addr=addr,
-                                 executor_config=self.executor_config,
                                  llm_args=self.llm_args,
                                  hf_model_dir=model_path,
                                  model_world_size=2)