#26 New models pipeline

keytotext/newmodels.py

@@ -0,0 +1,72 @@
+import torch
+from transformers import (
+    AutoModelForSeq2SeqLM,
+    AutoTokenizer,
+    PreTrainedModel,
+    PreTrainedTokenizer,
+)
+
+
+class NMPipeline:
+    def __init__(
+            self, model: PreTrainedModel, tokenizer: PreTrainedTokenizer, use_cuda: bool
+    ):
+        self.model = model
+        self.tokenizer = tokenizer
+        self.device = "cuda" if torch.cuda.is_available() and use_cuda else "cpu"
+        self.model.to(self.device)
+
+        assert self.model.__class__.__name__ in ["T5ForConditionalGeneration"]
+
+        if "T5ForConditionalGeneration" in self.model.__class__.__name__:
+            self.model_type = "t5"
+
+        self.default_generate_kwargs = {
+            "max_length": 1024,
+            "num_beams": 4,
+            "length_penalty": 1.5,
+            "no_repeat_ngram_size": 3,
+            "early_stopping": True,
+        }
+
+    def __call__(self, keywords, **kwargs):
+        inputs = self._prepare_inputs_for_k2t(keywords)
+        result = ""
+        if not kwargs:
+            kwargs = self.default_generate_kwargs
+
+        for txt in inputs:
+            input_ids = self._tokenize("{} </s>".format(txt), padding=False)
+            outputs = self.model.generate(input_ids.to(self.device), **kwargs)
+            result += self.tokenizer.decode(outputs[0])
+
+        result = re.sub("<pad>|</s>", "", result)
+        return result.strip()
+
+    def _prepare_inputs_for_k2t(self, keywords):
+        text = str(keywords)
+        text = text.replace(",", " ")
+        text = text.replace("'", "")
+        text = text.replace("[", "")
+        text = text.replace("]", "")
+        texts = text.split(".")
+        return texts
+
+    def _tokenize(
+            self,
+            inputs,
+            padding=True,
+            truncation=True,
+            add_special_tokens=True,
+            max_length=1024,
+    ):
+        inputs = self.tokenizer.encode(
+            inputs,
+            max_length=max_length,
+            add_special_tokens=add_special_tokens,
+            truncation=truncation,
+            padding="max_length" if padding else False,
+            pad_to_max_length=padding,
+            return_tensors="pt",
+        )
+        return inputs