[TPU][Core] Enable Pipeline Parallelism on TPU backend (#28506)
Signed-off-by: Chenyaaang <chenyangli@google.com>
This commit is contained in:
@@ -103,7 +103,7 @@ try:
|
||||
output = self.worker.model_runner.execute_model(
|
||||
scheduler_output, intermediate_tensors
|
||||
)
|
||||
if isinstance(output, IntermediateTensors):
|
||||
if self._is_intermediate_tensors(output):
|
||||
return scheduler_output, grammar_output, output
|
||||
|
||||
if isinstance(output, AsyncModelRunnerOutput):
|
||||
@@ -125,6 +125,9 @@ try:
|
||||
def override_env_vars(self, vars: dict[str, str]):
|
||||
os.environ.update(vars)
|
||||
|
||||
def _is_intermediate_tensors(self, output) -> bool:
|
||||
return isinstance(output, IntermediateTensors)
|
||||
|
||||
ray_import_err = None
|
||||
|
||||
except ImportError as e:
|
||||
|
||||
Reference in New Issue
Block a user