Add scibert model

category_classification/models/HibiscusMaximus__scibert_paper_classification/model.py

@@ -0,0 +1,65 @@
+import typing as tp
+
+import torch
+from transformers import pipeline, Pipeline, AutoModelForSequenceClassification
+from transformers.pipelines import PIPELINE_REGISTRY
+
+name = "HibiscusMaximus/scibert_paper_classification"
+
+
+class SciBertPaperClassifierPipeline(Pipeline):
+    def _sanitize_parameters(self, **kwargs):
+        return {}, {}, {}
+
+    def preprocess(self, inputs):
+        if not isinstance(inputs, tp.Iterable):
+            inputs = [inputs]
+        texts = [
+            f"AUTHORS: {' '.join(paper.authors) if isinstance(paper.authors, list) else paper.authors} "
+            f"TITLE: {paper.title} ABSTRACT: {paper.abstract}"
+            for paper in inputs
+        ]
+        inputs = self.tokenizer(
+            texts, truncation=True, padding=True, max_length=256, return_tensors="pt"
+        ).to(self.device)
+        return inputs
+
+    def _forward(self, model_inputs):
+        with torch.no_grad():
+            outputs = self.model(**model_inputs)
+        return outputs
+
+    def postprocess(self, model_outputs):
+        probs = torch.nn.functional.softmax(model_outputs.logits, dim=-1)
+        results = []
+        for prob in probs:
+            result = [
+                {"label": self.model.config.id2label[label_idx], "score": score.item()}
+                for label_idx, score in enumerate(prob)
+            ]
+            results.append(result)
+        if 1 == len(results):
+            return results[0]
+        return results
+
+
+PIPELINE_REGISTRY.register_pipeline(
+    "paper-classification",
+    pipeline_class=SciBertPaperClassifierPipeline,
+    pt_model=AutoModelForSequenceClassification,
+)
+
+
+class SciBertPaperClassifier:
+    def __init__(self):
+        self.pipeline = pipeline("paper-classification", model=name)
+
+    def __call__(self, input):
+        return self.pipeline(input)
+
+
+def get_model():
+    return SciBertPaperClassifier()
+
+
+supported_langs = ["en"]