Format fixes, add Attanasio et al. (2023) to readme

README.md

@@ -268,6 +268,7 @@ Inseq has been used in various research projects. A list of known publications t
     <li> <a href="https://arxiv.org/abs/2302.14220">Are Character-level Translations Worth the Wait? Comparing Character- and Subword-level Models for Machine Translation</a> (Edman et al., 2023) </li>
     <li> <a href="https://aclanthology.org/2023.nlp4convai-1.1/">Response Generation in Longitudinal Dialogues: Which Knowledge Representation Helps?</a> (Mousavi et al., 2023)  </li>
     <li> <a href="https://arxiv.org/abs/2310.01188">Quantifying the Plausibility of Context Reliance in Neural Machine Translation</a> (Sarti et al., 2023)</li>
+    <li> <a href="https://arxiv.org/abs/2310.12127">A Tale of Pronouns: Interpretability Informs Gender Bias Mitigation for Fairer Instruction-Tuned Machine Translation</a> (Attanasio et al., 2023)</li>
   </ol>
 
 </details>

inseq/models/attribution_model.py

@@ -404,11 +404,10 @@ def attribute(
             generated_texts = self.generate(
                 encoded_input, return_generation_output=False, batch_size=batch_size, **generation_args
             )
-        else:
-            if generation_args:
-                logger.warning(
-                    f"Generation arguments {generation_args} are provided, but will be ignored (constrained decoding)."
-                )
+        elif generation_args:
+            logger.warning(
+                f"Generation arguments {generation_args} are provided, but will be ignored (constrained decoding)."
+            )
         logger.debug(f"reference_texts={generated_texts}")
         attribution_method = self.get_attribution_method(method, override_default_attribution)
         attributed_fn = self.get_attributed_fn(attributed_fn)

inseq/utils/alignment_utils.py

@@ -74,10 +74,10 @@ def _get_aligner_subword_aligns(
 ) -> torch.Tensor:
     aligner = get_aligner_model()
     tokenizer = get_aligner_tokenizer()
-    tok_aenized = [tokenizer.tokenize(word) for word in src]
-    tok_benized = [tokenizer.tokenize(word) for word in tgt]
-    ids_src, sub2word_map_src = _preprocess_sequence_for_alignment(tok_aenized)
-    ids_tgt, sub2word_map_tgt = _preprocess_sequence_for_alignment(tok_benized)
+    tokenized_src = [tokenizer.tokenize(word) for word in src]
+    tokenized_tgt = [tokenizer.tokenize(word) for word in tgt]
+    ids_src, sub2word_map_src = _preprocess_sequence_for_alignment(tokenized_src)
+    ids_tgt, sub2word_map_tgt = _preprocess_sequence_for_alignment(tokenized_tgt)
     aligner.eval()
     with torch.no_grad():
         out_src = aligner(ids_src.unsqueeze(0), output_hidden_states=True)[2][align_layer][0, 1:-1]