Fixing the pipeline optimization by reindexing targets (V2) (#12330)

* Fixing the pipeline optimization by rescaling the logits first.

* Add test for target equivalence

Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>

src/transformers/pipelines/fill_mask.py

@@ -98,9 +98,9 @@ class FillMaskPipeline(Pipeline):
             args (:obj:`str` or :obj:`List[str]`):
                 One or several texts (or one list of prompts) with masked tokens.
             targets (:obj:`str` or :obj:`List[str]`, `optional`):
-                When passed, the model will return the scores for the passed token or tokens rather than the top k
+                When passed, the model will limit the scores to the passed targets instead of looking up in the whole
-                predictions in the entire vocabulary. If the provided targets are not in the model vocab, they will be
+                vocab. If the provided targets are not in the model vocab, they will be tokenized and the first
-                tokenized and the first resulting token will be used (with a warning).
+                resulting token will be used (with a warning, and that might be slower).
             top_k (:obj:`int`, `optional`):
                 When passed, overrides the number of predictions to return.
 
@@ -115,25 +115,56 @@ class FillMaskPipeline(Pipeline):
         inputs = self._parse_and_tokenize(*args, **kwargs)
         outputs = self._forward(inputs, return_tensors=True)
 
+        # top_k must be defined
+        if top_k is None:
+            top_k = self.top_k
+
         results = []
         batch_size = outputs.shape[0] if self.framework == "tf" else outputs.size(0)
 
         if targets is not None:
-            if len(targets) == 0 or len(targets[0]) == 0:
-                raise ValueError("At least one target must be provided when passed.")
             if isinstance(targets, str):
                 targets = [targets]
 
-            targets_proc = []
+            try:
+                vocab = self.tokenizer.get_vocab()
+            except Exception:
+                vocab = {}
+            target_ids = []
             for target in targets:
-                target_enc = self.tokenizer.tokenize(target)
+                id_ = vocab.get(target, None)
-                if len(target_enc) > 1 or target_enc[0] == self.tokenizer.unk_token:
+                if id_ is None:
+                    input_ids = self.tokenizer(
+                        target,
+                        add_special_tokens=False,
+                        return_attention_mask=False,
+                        return_token_type_ids=False,
+                        max_length=1,
+                        truncation=True,
+                    )["input_ids"]
+                    if len(input_ids) == 0:
+                        logger.warning(
+                            f"The specified target token `{target}` does not exist in the model vocabulary. "
+                            f"We cannot replace it with anything meaningful, ignoring it"
+                        )
+                        continue
+                    id_ = input_ids[0]
+                    # XXX: If users encounter this pass
+                    # it becomes pretty slow, so let's make sure
+                    # The warning enables them to fix the input to
+                    # get faster performance.
                     logger.warning(
                         f"The specified target token `{target}` does not exist in the model vocabulary. "
-                        f"Replacing with `{target_enc[0]}`."
+                        f"Replacing with `{self.tokenizer.convert_ids_to_tokens(id_)}`."
                     )
-                targets_proc.append(target_enc[0])
+                target_ids.append(id_)
-            target_inds = np.array(self.tokenizer.convert_tokens_to_ids(targets_proc))
+            target_ids = list(set(target_ids))
+            if len(target_ids) == 0:
+                raise ValueError("At least one target must be provided when passed.")
+            target_ids = np.array(target_ids)
+            # Cap top_k if there are targets
+            if top_k > target_ids.shape[0]:
+                top_k = target_ids.shape[0]
 
         for i in range(batch_size):
             input_ids = inputs["input_ids"][i]
@@ -147,14 +178,11 @@ class FillMaskPipeline(Pipeline):
 
                 logits = outputs[i, masked_index.item(), :]
                 probs = tf.nn.softmax(logits)
-                if targets is None:
+                if targets is not None:
-                    topk = tf.math.top_k(probs, k=top_k if top_k is not None else self.top_k)
+                    probs = tf.gather_nd(probs, tf.reshape(target_ids, (-1, 1)))
-                    values, predictions = topk.values.numpy(), topk.indices.numpy()
+
-                else:
+                topk = tf.math.top_k(probs, k=top_k)
-                    values = tf.gather_nd(probs, tf.reshape(target_inds, (-1, 1)))
+                values, predictions = topk.values.numpy(), topk.indices.numpy()
-                    sort_inds = tf.reverse(tf.argsort(values), [0])
-                    values = tf.gather_nd(values, tf.reshape(sort_inds, (-1, 1))).numpy()
-                    predictions = target_inds[sort_inds.numpy()]
             else:
                 masked_index = torch.nonzero(input_ids == self.tokenizer.mask_token_id, as_tuple=False)
 
@@ -163,16 +191,15 @@ class FillMaskPipeline(Pipeline):
 
                 logits = outputs[i, masked_index.item(), :]
                 probs = logits.softmax(dim=0)
-                if targets is None:
+                if targets is not None:
-                    values, predictions = probs.topk(top_k if top_k is not None else self.top_k)
+                    probs = probs[..., target_ids]
-                else:
+
-                    values = probs[..., target_inds]
+                values, predictions = probs.topk(top_k)
-                    sort_inds = list(reversed(values.argsort(dim=-1)))
-                    values = values[..., sort_inds]
-                    predictions = target_inds[sort_inds]
 
             for v, p in zip(values.tolist(), predictions.tolist()):
                 tokens = input_ids.numpy()
+                if targets is not None:
+                    p = target_ids[p].tolist()
                 tokens[masked_index] = p
                 # Filter padding out:
                 tokens = tokens[np.where(tokens != self.tokenizer.pad_token_id)]

tests/test_pipelines_fill_mask.py

@@ -15,7 +15,7 @@
 import unittest
 
 from transformers import pipeline
-from transformers.testing_utils import require_tf, require_torch, slow
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, slow
 
 from .test_pipelines_common import MonoInputPipelineCommonMixin
 
@@ -78,7 +78,8 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
     @require_torch
     def test_torch_fill_mask_with_targets(self):
         valid_inputs = ["My name is <mask>"]
-        valid_targets = [[" Teven", " Patrick", " Clara"], [" Sam"]]
+        # ' Sam' will yield a warning but work
+        valid_targets = [[" Teven", "ĠPatrick", "ĠClara"], ["ĠSam"], [" Sam"]]
         invalid_targets = [[], [""], ""]
         for model_name in self.small_models:
             unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="pt")
@@ -89,19 +90,77 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
             for targets in invalid_targets:
                 self.assertRaises(ValueError, unmasker, valid_inputs, targets=targets)
 
+    @require_torch
+    @slow
+    def test_torch_fill_mask_targets_equivalence(self):
+        model_name = self.large_models[0]
+        unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="pt")
+        unmasked = unmasker(self.valid_inputs[0])
+        tokens = [top_mask["token_str"] for top_mask in unmasked]
+        scores = [top_mask["score"] for top_mask in unmasked]
+
+        unmasked_targets = unmasker(self.valid_inputs[0], targets=tokens)
+        target_scores = [top_mask["score"] for top_mask in unmasked_targets]
+
+        self.assertEqual(scores, target_scores)
+
+    @require_torch
+    def test_torch_fill_mask_with_targets_and_topk(self):
+        model_name = self.small_models[0]
+        unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="pt")
+        targets = [" Teven", "ĠPatrick", "ĠClara"]
+        top_k = 2
+        outputs = unmasker("My name is <mask>", targets=targets, top_k=top_k)
+        self.assertEqual(
+            nested_simplify(outputs),
+            [
+                {"sequence": "My name is Patrick", "score": 0.0, "token": 3499, "token_str": " Patrick"},
+                {"sequence": "My name is Te", "score": 0.0, "token": 2941, "token_str": " Te"},
+            ],
+        )
+
+    @require_torch
+    def test_torch_fill_mask_with_duplicate_targets_and_topk(self):
+        model_name = self.small_models[0]
+        unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="pt")
+        # String duplicates + id duplicates
+        targets = [" Teven", "ĠPatrick", "ĠClara", "ĠClara", " Clara"]
+        top_k = 10
+        outputs = unmasker("My name is <mask>", targets=targets, top_k=top_k)
+
+        # The target list contains duplicates, so we can't output more
+        # than them
+        self.assertEqual(len(outputs), 3)
+
     @require_tf
     def test_tf_fill_mask_with_targets(self):
         valid_inputs = ["My name is <mask>"]
-        valid_targets = [[" Teven", " Patrick", " Clara"], [" Sam"]]
+        # ' Teven' will yield a warning but work as " Te"
         invalid_targets = [[], [""], ""]
-        for model_name in self.small_models:
+        unmasker = pipeline(
-            unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="tf")
+            task="fill-mask", model=self.small_models[0], tokenizer=self.small_models[0], framework="tf"
-            for targets in valid_targets:
+        )
-                outputs = unmasker(valid_inputs, targets=targets)
+        outputs = unmasker(valid_inputs, targets=[" Teven", "ĠPatrick", "ĠClara"])
-                self.assertIsInstance(outputs, list)
+        self.assertEqual(
-                self.assertEqual(len(outputs), len(targets))
+            nested_simplify(outputs),
-            for targets in invalid_targets:
+            [
-                self.assertRaises(ValueError, unmasker, valid_inputs, targets=targets)
+                {"sequence": "My name is Clara", "score": 0.0, "token": 13606, "token_str": " Clara"},
+                {"sequence": "My name is Patrick", "score": 0.0, "token": 3499, "token_str": " Patrick"},
+                {"sequence": "My name is Te", "score": 0.0, "token": 2941, "token_str": " Te"},
+            ],
+        )
+        # topk
+        outputs = unmasker(valid_inputs, targets=[" Teven", "ĠPatrick", "ĠClara"], top_k=2)
+        self.assertEqual(
+            nested_simplify(outputs),
+            [
+                {"sequence": "My name is Clara", "score": 0.0, "token": 13606, "token_str": " Clara"},
+                {"sequence": "My name is Patrick", "score": 0.0, "token": 3499, "token_str": " Patrick"},
+            ],
+        )
+        for targets in invalid_targets:
+            with self.assertRaises(ValueError):
+                unmasker(valid_inputs, targets=targets)
 
     @require_torch
     @slow
@@ -111,7 +170,7 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
             "My name is <mask>",
             "The largest city in France is <mask>",
         ]
-        valid_targets = [" Patrick", " Clara"]
+        valid_targets = ["ĠPatrick", "ĠClara"]
         for model_name in self.large_models:
             unmasker = pipeline(
                 task="fill-mask",
@@ -184,7 +243,7 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
             "My name is <mask>",
             "The largest city in France is <mask>",
         ]
-        valid_targets = [" Patrick", " Clara"]
+        valid_targets = ["ĠPatrick", "ĠClara"]
         for model_name in self.large_models:
             unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="tf", top_k=2)
 
@@ -242,3 +301,17 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
                     self.assertIn(key, result)
 
             self.assertRaises(Exception, unmasker, [None])
+
+    @require_tf
+    @slow
+    def test_tf_fill_mask_targets_equivalence(self):
+        model_name = self.large_models[0]
+        unmasker = pipeline(task="fill-mask", model=model_name, tokenizer=model_name, framework="tf")
+        unmasked = unmasker(self.valid_inputs[0])
+        tokens = [top_mask["token_str"] for top_mask in unmasked]
+        scores = [top_mask["score"] for top_mask in unmasked]
+
+        unmasked_targets = unmasker(self.valid_inputs[0], targets=tokens)
+        target_scores = [top_mask["score"] for top_mask in unmasked_targets]
+
+        self.assertEqual(scores, target_scores)