refactor: call model forward directly instead of base model backbone

Author: ciaoyizhen

trl/experimental/utils.py

@@ -372,7 +372,7 @@ def get_reward(
         input_ids=input_ids,
         attention_mask=attention_mask,
         position_ids=position_ids
-        ).last_hidden_state
+        ).logits
     sequence_lengths = first_true_indices(query_responses[:, context_length:] == pad_token_id) - 1 + context_length
     # https://github.com/huggingface/transformers/blob/dc68a39c8111217683bf49a4912d0c9018bab33d/src/transformers/models/gpt2/modeling_gpt2.py#L1454
     return (