added pretrained pipeline

config.json

@@ -0,0 +1,44 @@
+{
+  "_name_or_path": "DennisOneHealth/mesh-encoder",
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "custom_pipelines": {
+    "disease-normalization": {
+      "default": {
+        "model": {
+          "pt": [
+            "DennisOneHealth/mesh-encoder",
+            "166966f"
+          ]
+        }
+      },
+      "impl": "mesh_pipeline.MeshPipeline",
+      "pt": [
+        "AutoModelForSequenceClassification"
+      ],
+      "tf": [],
+      "type": "text"
+    }
+  },
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.33.3",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 28998
+}

mesh_embeddings.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a4b0571afbed3415179ff6a86544e0db088289ae3b4ef28269527b4efd1862d
+size 26902763

mesh_pipeline.py

@@ -0,0 +1,44 @@
+import torch
+from transformers import Pipeline
+
+
+class MeshPipeline(Pipeline):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        mesh_ordered_dictionary = torch.load('./mesh_embeddings.pt')
+
+        mesh_records = mesh_ordered_dictionary.keys()
+        self.mesh_records = list(mesh_records)
+
+        mesh_embeddings = mesh_ordered_dictionary.values()
+        mesh_embeddings = torch.stack(list(mesh_embeddings), dim=0)
+        self.mesh_embeddings = mesh_embeddings
+
+    def _sanitize_parameters(self, **kwargs):
+        preprocess_kwargs = {}
+        _forward_kwargs = {}
+        postprocess_kwargs = {}
+        return preprocess_kwargs, _forward_kwargs, postprocess_kwargs
+
+    def preprocess(self, annotation):
+        text, start, end = annotation.values()
+        model_inputs = self.tokenizer(text, return_tensors='pt')
+        token_indices = [
+            model_inputs.char_to_token(char_idx)
+            for char_idx in range(start, end)
+            if model_inputs.char_to_token(char_idx) is not None
+        ]
+        model_inputs['token_type_ids'][0, token_indices] = 1
+
+        return model_inputs
+
+    def _forward(self, model_inputs):
+        return self.model(**model_inputs)
+
+    def postprocess(self, model_outputs):
+        text_embeddings = model_outputs.last_hidden_state[:, 0, :]
+        mesh_embeddings = self.mesh_embeddings.to(text_embeddings.device)
+        scores = text_embeddings.matmul(mesh_embeddings.T)
+        labels = torch.argmax(scores, dim=1)
+        labels = [self.mesh_records[label] for label in labels]
+        return {'labels': labels, 'scores': scores}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25aa68a3276a2cd7c4a6990f1d659984d1e84a7a5d9e19730d2b9237746c7407
+size 433312179

special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "max_length": 200,
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_to_multiple_of": null,
+  "pad_token": "[PAD]",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}