[2/N] executor pass the complete config to worker/modelrunner (#9938)

Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
youkaichao
2024-11-02 07:35:05 -07:00
committed by GitHub
co-authored by Nick Hill
parent 1d4cfe2be1
commit e893795443
44 changed files with 249 additions and 579 deletions
+1 -8
View File
@@ -138,18 +138,11 @@ class CPUExecutor(ExecutorBase):
assert self.distributed_init_method is not None
kwargs = dict(
model_config=self.model_config,
parallel_config=self.parallel_config,
scheduler_config=self.scheduler_config,
device_config=self.device_config,
cache_config=self.cache_config,
load_config=self.load_config,
vllm_config=self.vllm_config,
local_rank=local_rank,
rank=rank,
distributed_init_method=self.distributed_init_method,
lora_config=self.lora_config,
kv_cache_dtype=self.cache_config.cache_dtype,
prompt_adapter_config=self.prompt_adapter_config,
is_driver_worker=rank == 0,
)
wrapper.init_worker(**kwargs)
+2 -2
View File
@@ -1,7 +1,7 @@
from abc import ABC, abstractmethod
from typing import List, Optional, Set, Tuple
from vllm.config import EngineConfig
from vllm.config import VllmConfig
from vllm.lora.request import LoRARequest
from vllm.model_executor.layers.sampler import SamplerOutput
from vllm.prompt_adapter.request import PromptAdapterRequest
@@ -20,7 +20,7 @@ class ExecutorBase(ABC):
def __init__(
self,
vllm_config: EngineConfig,
vllm_config: VllmConfig,
) -> None:
self.vllm_config = vllm_config
self.model_config = vllm_config.model_config
+1 -10
View File
@@ -49,21 +49,12 @@ class GPUExecutor(ExecutorBase):
distributed_init_method = get_distributed_init_method(
get_ip(), get_open_port())
return dict(
model_config=self.model_config,
parallel_config=self.parallel_config,
scheduler_config=self.scheduler_config,
device_config=self.device_config,
cache_config=self.cache_config,
load_config=self.load_config,
vllm_config=self.vllm_config,
local_rank=local_rank,
rank=rank,
distributed_init_method=distributed_init_method,
lora_config=self.lora_config,
speculative_config=self.speculative_config,
prompt_adapter_config=self.prompt_adapter_config,
is_driver_worker=(not self.parallel_config)
or (rank % self.parallel_config.tensor_parallel_size == 0),
observability_config=self.observability_config,
)
def _get_worker_module_and_class(
+1 -5
View File
@@ -29,11 +29,7 @@ class NeuronExecutor(ExecutorBase):
distributed_init_method = get_distributed_init_method(
get_ip(), get_open_port())
self.driver_worker = NeuronWorker(
model_config=self.model_config,
parallel_config=self.parallel_config,
scheduler_config=self.scheduler_config,
device_config=self.device_config,
cache_config=self.cache_config,
vllm_config=self.vllm_config,
local_rank=0,
rank=0,
distributed_init_method=distributed_init_method)
+1 -7
View File
@@ -48,16 +48,10 @@ class OpenVINOExecutor(ExecutorBase):
get_ip(), get_open_port())
self.driver_worker = OpenVINOWorker(
ov_core=self.ov_core,
model_config=self.model_config,
parallel_config=self.parallel_config,
scheduler_config=self.scheduler_config,
device_config=self.device_config,
cache_config=self.cache_config,
load_config=self.load_config,
vllm_config=self.vllm_config,
local_rank=0,
rank=0,
distributed_init_method=distributed_init_method,
lora_config=self.lora_config,
kv_cache_dtype=self.cache_config.cache_dtype,
is_driver_worker=True,
)
+1 -6
View File
@@ -44,12 +44,7 @@ class TPUExecutor(ExecutorBase):
distributed_init_method = get_distributed_init_method(
get_ip(), get_open_port())
return dict(
model_config=self.model_config,
parallel_config=self.parallel_config,
scheduler_config=self.scheduler_config,
device_config=self.device_config,
cache_config=self.cache_config,
load_config=self.load_config,
vllm_config=self.vllm_config,
local_rank=local_rank,
rank=rank,
distributed_init_method=distributed_init_method,