set z3 leaf for deepseek v2 (#1809) [skip ci]
* set z3 leaf for deepseek v2 * add deepseek v2 chat template
This commit is contained in:
@@ -25,12 +25,12 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
|
||||
]
|
||||
|
||||
|
||||
def patch_for_multipack(model_type, model_name=None):
|
||||
def patch_for_multipack(model_type, model_name=None, is_remote_code=False):
|
||||
if model_type == "gemmoe":
|
||||
patch_remote(model_name, ".configuration_gemmoe", ".modeling_gemmoe")
|
||||
elif model_type == "deepseek_v2":
|
||||
patch_remote(model_name, ".configuration_deepseek", ".modeling_deepseek")
|
||||
elif hasattr(transformers, "modeling_flash_attention_utils"):
|
||||
elif hasattr(transformers, "modeling_flash_attention_utils") and not is_remote_code:
|
||||
transformers.modeling_flash_attention_utils._get_unpad_data = ( # pylint: disable=protected-access
|
||||
get_unpad_data
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user