use skip_move_to_device for all cases (#3015)
* use skip_move_to_device for all cases * use experimental option for skip move
This commit is contained in:
@@ -845,6 +845,9 @@ class ModelLoader:
|
||||
self.model._tp_size = self.cfg.tensor_parallel_size
|
||||
self.model._device_mesh = self.model_kwargs["device_mesh"]
|
||||
|
||||
if self.cfg.experimental_skip_move_to_device is not None:
|
||||
skip_move_to_device = self.cfg.experimental_skip_move_to_device
|
||||
|
||||
return skip_move_to_device
|
||||
|
||||
def _set_z3_leaf_modules(self):
|
||||
|
||||
@@ -62,6 +62,14 @@ class ModelInputConfig(BaseModel):
|
||||
json_schema_extra={"description": "Trust remote code for untrusted source"},
|
||||
)
|
||||
|
||||
experimental_skip_move_to_device: bool | None = Field(
|
||||
default=None,
|
||||
json_schema_extra={
|
||||
"description": "Don't move the model to the device before sharding. "
|
||||
"This is an experimental feature that may be included in the future as the default."
|
||||
},
|
||||
)
|
||||
|
||||
@field_validator("trust_remote_code")
|
||||
@classmethod
|
||||
def hint_trust_remote_code(cls, trust_remote_code):
|
||||
|
||||
Reference in New Issue
Block a user