Upload model

config.json

@@ -0,0 +1,17 @@
+{
+  "_name_or_path": "/content/drive/MyDrive/project/multilingual-sentence-embedder/ckpt",
+  "architectures": [
+    "SentenceEmbedderModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_sentence_embedder.SentenceEmbedderConfig",
+    "AutoModel": "modeling_sentence_embedder.SentenceEmbedderModel"
+  },
+  "backbone_name": "xlm-roberta-base",
+  "base_model_name": "nreimers/mMiniLMv2-L6-H384-distilled-from-XLMR-Large",
+  "init_backbone": true,
+  "model_type": "sentence_embedder",
+  "output_size": 768,
+  "torch_dtype": "float32",
+  "transformers_version": "4.25.1"
+}

configuration_sentence_embedder.py

@@ -0,0 +1,16 @@
+from transformers import PretrainedConfig
+
+class SentenceEmbedderConfig(PretrainedConfig):
+    model_type = 'sentence_embedder'
+
+    def __init__(
+        self, 
+        backbone_name: str = 'xlm-roberta-base',
+        output_size: int = 768,
+        init_backbone: bool = False,
+        **kwargs
+    ):
+        self.backbone_name = backbone_name
+        self.output_size = output_size
+        self.init_backbone = init_backbone
+        super().__init__(**kwargs)

modeling_sentence_embedder.py

@@ -0,0 +1,23 @@
+import torch.nn as nn
+from transformers import PreTrainedModel, AutoConfig, AutoModel
+
+from .configuration_sentence_embedder import SentenceEmbedderConfig
+
+class SentenceEmbedderModel(PreTrainedModel):
+    config_class = SentenceEmbedderConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        if config.init_backbone:
+            self.backbone = AutoModel.from_pretrained(config.backbone_name)
+        else:
+            backbone_config = AutoConfig.from_pretrained(config.backbone_name)
+            self.backbone = AutoModel.from_config(backbone_config)
+        self.projection = nn.Linear(self.backbone.config.hidden_size, config.output_size)
+
+
+    def forward(self, input_ids, attention_mask, head=None):
+        outputs = self.backbone(input_ids, attention_mask)
+        last_hidden_state = self.projection(outputs.last_hidden_state)
+        outputs.last_hidden_state = last_hidden_state
+        return outputs

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6971b13c8c8869e2eacf90547fb3f8da4010ff984496245c85eb8a56d5ec9335
+size 1114607797