[platform] support custom torch.compile backend key (#11318)
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
@@ -45,6 +45,7 @@ from vllm.model_executor.model_loader.weight_utils import (
|
||||
row_parallel_weight_loader)
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.model_executor.utils import set_weight_attrs
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors
|
||||
|
||||
from .interfaces import SupportsLoRA, SupportsPP
|
||||
@@ -53,7 +54,7 @@ from .utils import (extract_layer_index, is_pp_missing_parameter,
|
||||
maybe_prefix)
|
||||
|
||||
|
||||
@torch.compile
|
||||
@torch.compile(backend=current_platform.simple_compile_backend)
|
||||
def layer_norm_func(hidden_states, weight, variance_epsilon):
|
||||
input_dtype = hidden_states.dtype
|
||||
hidden_states = hidden_states.to(torch.float32)
|
||||
|
||||
@@ -20,6 +20,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
|
||||
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import IntermediateTensors
|
||||
|
||||
from .interfaces import SupportsPP
|
||||
@@ -54,12 +55,12 @@ class HeadMajorColumnParallelLinear(MergedColumnParallelLinear):
|
||||
return load_column_parallel_weight(param, loaded_weight)
|
||||
|
||||
|
||||
@torch.compile(dynamic=True)
|
||||
@torch.compile(dynamic=True, backend=current_platform.simple_compile_backend)
|
||||
def quick_gelu(x):
|
||||
return x * torch.sigmoid(1.702 * x)
|
||||
|
||||
|
||||
@torch.compile(dynamic=True)
|
||||
@torch.compile(dynamic=True, backend=current_platform.simple_compile_backend)
|
||||
def gegelu(input, limit: Optional[float] = None):
|
||||
a_gelu, a_linear = input[..., ::2], input[..., 1::2]
|
||||
if limit is not None:
|
||||
|
||||
Reference in New Issue
Block a user