Fix weight tying (#51665)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -560,12 +560,7 @@ class ExaoneMoeForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
|
||||
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
|
||||
loader = AutoWeightsLoader(
|
||||
self,
|
||||
# With tie_word_embeddings, we can skip lm_head.weight
|
||||
# The weight might appear unnecessarily in the files if the model is
|
||||
# processed with quantization, LoRA, fine-tuning, etc.
|
||||
skip_prefixes=(
|
||||
["lm_head.", "mtp."] if self.config.tie_word_embeddings else ["mtp."]
|
||||
),
|
||||
skip_prefixes=["mtp."],
|
||||
# Skip loading extra parameters for GPTQ/modelopt models.
|
||||
ignore_unexpected_suffixes=[
|
||||
".bias",
|
||||
|
||||
Reference in New Issue
Block a user