Signed-off-by: Lucas Kabela <lucaskabela@meta.com>
This commit is contained in:
@@ -10,7 +10,6 @@ from typing import Callable, Optional
|
||||
|
||||
import torch
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.config import (CompilationLevel, CUDAGraphMode,
|
||||
get_current_vllm_config)
|
||||
from vllm.logger import init_logger
|
||||
@@ -47,11 +46,10 @@ class TorchCompileWrapperWithCustomDispatcher:
|
||||
options = get_current_vllm_config(
|
||||
).compilation_config.inductor_compile_config
|
||||
|
||||
compiled_callable = torch.compile(
|
||||
self.forward,
|
||||
fullgraph=envs.VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE,
|
||||
backend=backend,
|
||||
options=options)
|
||||
compiled_callable = torch.compile(self.forward,
|
||||
fullgraph=True,
|
||||
backend=backend,
|
||||
options=options)
|
||||
|
||||
self.compiled_callable = compiled_callable
|
||||
self.original_code_object = self.__class__.forward.__code__
|
||||
|
||||
Reference in New Issue
Block a user