Enable safetensors loading for all models (#974)
This commit is contained in:
@@ -279,11 +279,11 @@ class BloomForCausalLM(nn.Module):
|
||||
def load_weights(self,
|
||||
model_name_or_path: str,
|
||||
cache_dir: Optional[str] = None,
|
||||
use_np_cache: bool = False):
|
||||
load_format: str = "auto"):
|
||||
tp_rank = get_tensor_model_parallel_rank()
|
||||
state_dict = self.state_dict()
|
||||
for name, loaded_weight in hf_model_weights_iterator(
|
||||
model_name_or_path, cache_dir, use_np_cache):
|
||||
model_name_or_path, cache_dir, load_format):
|
||||
if name == "lm_head.weight":
|
||||
# Since hidden_states are parallelized, we need to
|
||||
# load lm_head.weight in parallel.
|
||||
|
||||
Reference in New Issue
Block a user