Scripted tokenizer support for DocModel (facebookresearch#1314)

Summary:
Pull Request resolved: facebookresearch#1314

Adding scripted tokenization support to the most widely used model

OSS Test failures
Waiting for a TorchScript diff to land: https://fb.workplace.com/groups/329222650990087/permalink/632527153992967/

Differential Revision: D20955370

fbshipit-source-id: 6e002136fc0113dfe87a24ca3f82be9a73d1d0bc

pytext/models/doc_model.py

@@ -14,6 +14,7 @@
     TokenTensorizer,
     UidTensorizer,
 )
+from pytext.data.tokenizers import DoNothingTokenizer
 from pytext.data.utils import PAD, UNK
 from pytext.exporters.exporter import ModelExporter
 from pytext.loss import BinaryCrossEntropyLoss, MultiLabelSoftMarginLoss
@@ -119,6 +120,13 @@ def torchscriptify(self, tensorizers, traced_model):
 
         input_vocab = tensorizers["tokens"].vocab
         max_seq_len = tensorizers["tokens"].max_seq_len or -1
+        scripted_tokenizer = None
+        try:
+            scripted_tokenizer = tensorizers["tokens"].tokenizer.torchscriptify()
+        except NotImplementedError:
+            pass
+        if scripted_tokenizer and isinstance(scripted_tokenizer, DoNothingTokenizer):
+            scripted_tokenizer = None
 
         """
         The input tensor packing memory is allocated/cached for different shapes,
@@ -139,6 +147,7 @@ def __init__(self):
                 self.output_layer = output_layer
                 self.pad_idx = jit.Attribute(input_vocab.get_pad_index(), int)
                 self.max_seq_len = jit.Attribute(max_seq_len, int)
+                self.tokenizer = scripted_tokenizer
 
             @jit.script_method
             def forward(
@@ -148,6 +157,16 @@ def forward(
                 tokens: Optional[List[List[str]]] = None,
                 languages: Optional[List[str]] = None,
             ):
+                # PyTorch breaks with 2 'not None' checks right now.
+                if texts is not None:
+                    if tokens is not None:
+                        raise RuntimeError("Can't set both tokens and texts")
+                    if self.tokenizer is not None:
+                        tokens = [
+                            [t[0] for t in self.tokenizer.tokenize(text)]
+                            for text in texts
+                        ]
+
                 if tokens is None:
                     raise RuntimeError("tokens is required")
 
@@ -171,6 +190,7 @@ def __init__(self):
                 self.output_layer = output_layer
                 self.pad_idx = jit.Attribute(input_vocab.get_pad_index(), int)
                 self.max_seq_len = jit.Attribute(max_seq_len, int)
+                self.tokenizer = scripted_tokenizer
 
             @jit.script_method
             def forward(
@@ -181,6 +201,16 @@ def forward(
                 languages: Optional[List[str]] = None,
                 dense_feat: Optional[List[List[float]]] = None,
             ):
+                # PyTorch breaks with 2 'not None' checks right now.
+                if texts is not None:
+                    if tokens is not None:
+                        raise RuntimeError("Can't set both tokens and texts")
+                    if self.tokenizer is not None:
+                        tokens = [
+                            [t[0] for t in self.tokenizer.tokenize(text)]
+                            for text in texts
+                        ]
+
                 if tokens is None:
                     raise RuntimeError("tokens is required")
                 if dense_feat is None: