[3/N] model runner pass the whole config to model (#9958)
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
@@ -453,13 +453,7 @@ class CPUModelRunner(ModelRunnerBase[ModelInputForCPU]):
|
||||
return uses_mrope(self.model_config.hf_config)
|
||||
|
||||
def load_model(self) -> None:
|
||||
self.model = get_model(model_config=self.model_config,
|
||||
load_config=self.load_config,
|
||||
device_config=self.device_config,
|
||||
lora_config=self.lora_config,
|
||||
parallel_config=self.parallel_config,
|
||||
scheduler_config=self.scheduler_config,
|
||||
cache_config=self.cache_config)
|
||||
self.model = get_model(vllm_config=self.vllm_config)
|
||||
|
||||
def make_model_input_from_broadcasted_tensor_dict(
|
||||
self,
|
||||
|
||||
@@ -1051,13 +1051,7 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
|
||||
def load_model(self) -> None:
|
||||
logger.info("Starting to load model %s...", self.model_config.model)
|
||||
with DeviceMemoryProfiler() as m:
|
||||
self.model = get_model(model_config=self.model_config,
|
||||
device_config=self.device_config,
|
||||
load_config=self.load_config,
|
||||
lora_config=self.lora_config,
|
||||
parallel_config=self.parallel_config,
|
||||
scheduler_config=self.scheduler_config,
|
||||
cache_config=self.cache_config)
|
||||
self.model = get_model(vllm_config=self.vllm_config)
|
||||
|
||||
self.model_memory_usage = m.consumed_memory
|
||||
logger.info("Loading model weights took %.4f GB",
|
||||
|
||||
@@ -137,15 +137,7 @@ class TPUModelRunner(ModelRunnerBase[ModelInputForTPU]):
|
||||
"vllm.model_executor.layers.vocab_parallel_embedding."
|
||||
"get_tensor_model_parallel_rank",
|
||||
return_value=xm_tp_rank):
|
||||
model = get_model(
|
||||
model_config=self.model_config,
|
||||
load_config=self.load_config,
|
||||
device_config=self.device_config,
|
||||
parallel_config=self.parallel_config,
|
||||
cache_config=self.cache_config,
|
||||
scheduler_config=self.scheduler_config,
|
||||
lora_config=None,
|
||||
)
|
||||
model = get_model(vllm_config=self.vllm_config)
|
||||
model = model.eval()
|
||||
xm.wait_device_ops()
|
||||
self.model = ModelWrapper(model)
|
||||
|
||||
@@ -405,15 +405,7 @@ class XPUModelRunner(ModelRunnerBase[ModelInputForXPUWithSamplingMetadata]):
|
||||
|
||||
def load_model(self) -> None:
|
||||
with DeviceMemoryProfiler() as m:
|
||||
self.model = get_model(
|
||||
model_config=self.model_config,
|
||||
device_config=self.device_config,
|
||||
load_config=self.load_config,
|
||||
lora_config=self.lora_config,
|
||||
parallel_config=self.parallel_config,
|
||||
scheduler_config=self.scheduler_config,
|
||||
cache_config=self.cache_config,
|
||||
)
|
||||
self.model = get_model(vllm_config=self.vllm_config)
|
||||
|
||||
self.model_memory_usage = m.consumed_memory
|
||||
logger.info("Loading model weights took %.4f GB",
|
||||
|
||||
Reference in New Issue
Block a user