[CI] Add mteb testing for rerank models (#19344)
This commit is contained in:
@@ -432,12 +432,7 @@ class BertWithRope(nn.Module, SupportsV0Only, SupportsQuant):
|
||||
else:
|
||||
hidden_states = self.embeddings(input_ids=input_ids,
|
||||
token_type_ids=token_type_ids)
|
||||
hidden_states = self.encoder(positions, hidden_states)
|
||||
|
||||
# convert the embedding output to float32,
|
||||
# otherwise precision will be lost significantly
|
||||
hidden_states = hidden_states.to(torch.float32)
|
||||
return hidden_states
|
||||
return self.encoder(positions, hidden_states)
|
||||
|
||||
def load_weights(self, weights: Iterable[tuple[str,
|
||||
torch.Tensor]]) -> set[str]:
|
||||
|
||||
Reference in New Issue
Block a user