Enable dense features in ByteTokensDocumentModel (#763)

Summary:
Pull Request resolved: #763

Update a couple model methods to support dense features, as the parent model `DocModel` already supports dense features.

Differential Revision: D16187410

fbshipit-source-id: b7fc7f4e753a4bea4aaa1f3161a454d54c791958

Author: Michael Wu

pytext/models/doc_model.py

@@ -225,7 +225,9 @@ def create_embedding(cls, config, tensorizers: Dict[str, Tensorizer]):
         assert word_tensorizer.column == byte_tensorizer.column
 
         word_embedding = create_module(
-            config.embedding, tensorizer=tensorizers["tokens"]
+            config.embedding,
+            tensorizer=tensorizers["tokens"],
+            init_from_saved_state=config.init_from_saved_state,
         )
         byte_embedding = CharacterEmbedding(
             ByteTokenTensorizer.NUM_BYTES,
@@ -241,10 +243,16 @@ def arrange_model_inputs(self, tensor_dict):
         tokens, seq_lens, _ = tensor_dict["tokens"]
         token_bytes, byte_seq_lens, _ = tensor_dict["token_bytes"]
         assert (seq_lens == byte_seq_lens).all().item()
-        return tokens, token_bytes, seq_lens
+        model_inputs = tokens, token_bytes, seq_lens
+        if "dense" in tensor_dict:
+            model_inputs += (tensor_dict["dense"],)
+        return model_inputs
 
     def get_export_input_names(self, tensorizers):
-        return ["tokens", "token_bytes", "tokens_lens"]
+        names = ["tokens", "token_bytes", "tokens_lens"]
+        if "dense" in tensorizers:
+            names.append("float_vec_vals")
+        return names
 
     def torchscriptify(self, tensorizers, traced_model):
         output_layer = self.output_layer.torchscript_predictions()
@@ -277,7 +285,35 @@ def forward(self, tokens: List[List[str]]):
                 )
                 return self.output_layer(logits)
 
-        return Model()
+        class ModelWithDenseFeat(jit.ScriptModule):
+            def __init__(self):
+                super().__init__()
+                self.vocab = Vocabulary(input_vocab, unk_idx=input_vocab.idx[UNK])
+                self.max_byte_len = jit.Attribute(max_byte_len, int)
+                self.byte_offset_for_non_padding = jit.Attribute(
+                    byte_offset_for_non_padding, int
+                )
+                self.pad_idx = jit.Attribute(input_vocab.idx[PAD], int)
+                self.model = traced_model
+                self.output_layer = output_layer
+
+            @jit.script_method
+            def forward(self, tokens: List[List[str]], dense_feat: List[List[float]]):
+                seq_lens = make_sequence_lengths(tokens)
+                word_ids = self.vocab.lookup_indices_2d(tokens)
+                word_ids = pad_2d(word_ids, seq_lens, self.pad_idx)
+                token_bytes, _ = make_byte_inputs(
+                    tokens, self.max_byte_len, self.byte_offset_for_non_padding
+                )
+                logits = self.model(
+                    torch.tensor(word_ids),
+                    token_bytes,
+                    torch.tensor(seq_lens),
+                    torch.tensor(dense_feat),
+                )
+                return self.output_layer(logits)
+
+        return ModelWithDenseFeat() if "dense" in tensorizers else Model()
 
 
 class DocRegressionModel(DocModel):