add model
Browse files- config.json +31 -0
- modeling_roberta.py +24 -0
- pytorch_model.bin +3 -0
config.json
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "/content/drive/MyDrive/ColabModels/XROBERTA_USE_QA/pytorch_model/",
|
3 |
+
"architectures": [
|
4 |
+
"XLMRobertaModel"
|
5 |
+
],
|
6 |
+
"attention_probs_dropout_prob": 0.1,
|
7 |
+
"auto_map": {
|
8 |
+
"AutoModel": "modeling_roberta.XLMRobertaModel"
|
9 |
+
},
|
10 |
+
"bos_token_id": 0,
|
11 |
+
"classifier_dropout": null,
|
12 |
+
"eos_token_id": 2,
|
13 |
+
"hidden_act": "gelu",
|
14 |
+
"hidden_dropout_prob": 0.1,
|
15 |
+
"hidden_size": 768,
|
16 |
+
"initializer_range": 0.02,
|
17 |
+
"intermediate_size": 3072,
|
18 |
+
"layer_norm_eps": 1e-12,
|
19 |
+
"max_position_embeddings": 514,
|
20 |
+
"model_type": "xlm-roberta",
|
21 |
+
"num_attention_heads": 12,
|
22 |
+
"num_hidden_layers": 12,
|
23 |
+
"pad_token_id": 1,
|
24 |
+
"position_embedding_type": "absolute",
|
25 |
+
"roberta": 1,
|
26 |
+
"torch_dtype": "float32",
|
27 |
+
"transformers_version": "4.18.0",
|
28 |
+
"type_vocab_size": 1,
|
29 |
+
"use_cache": true,
|
30 |
+
"vocab_size": 250002
|
31 |
+
}
|
modeling_roberta.py
ADDED
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
|
3 |
+
from transformers import XLMRobertaModel as XLMRobertaModelBase
|
4 |
+
|
5 |
+
|
6 |
+
class XLMRobertaModel(XLMRobertaModelBase):
|
7 |
+
def __init__(self, config):
|
8 |
+
super().__init__(config)
|
9 |
+
self.question_projection = torch.nn.Linear(768, 512)
|
10 |
+
self.answer_projection = torch.nn.Linear(768, 512)
|
11 |
+
|
12 |
+
def _embed(self, input_ids, attention_mask, projection):
|
13 |
+
outputs = super().__call__(input_ids, attention_mask=attention_mask)
|
14 |
+
sequence_output = outputs[0]
|
15 |
+
|
16 |
+
input_mask_expanded = attention_mask.unsqueeze(-1).expand(sequence_output.size()).float()
|
17 |
+
embeddings = torch.sum(sequence_output * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
|
18 |
+
return torch.tanh(projection(embeddings))
|
19 |
+
|
20 |
+
def question(self, input_ids, attention_mask):
|
21 |
+
return self._embed(input_ids, attention_mask, self.question_projection)
|
22 |
+
|
23 |
+
def answer(self, input_ids, attention_mask):
|
24 |
+
return self._embed(input_ids, attention_mask, self.answer_projection)
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1e3ded1f396d3f20cdb3249faefa44c373f3f59d5adaa19542eb4d20ffc3b908
|
3 |
+
size 1115392297
|