Upload model
Browse files- config.json +15 -0
- configuration.py +19 -0
- encoder.py +26 -0
- projector.py +29 -0
- pytorch_model.bin +3 -0
config.json
ADDED
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"ThaiLightWeightEncoderModel"
|
4 |
+
],
|
5 |
+
"auto_map": {
|
6 |
+
"AutoConfig": "configuration.ThaiLightWeightEncoderConfig",
|
7 |
+
"AutoModel": "encoder.ThaiLightWeightEncoderModel"
|
8 |
+
},
|
9 |
+
"dropout": 0.2,
|
10 |
+
"final_embedding_dim": 512,
|
11 |
+
"input_embedding_dim": 300,
|
12 |
+
"torch_dtype": "float32",
|
13 |
+
"transformers_version": "4.28.1",
|
14 |
+
"word_vector_model_name": "thai2fit_wv"
|
15 |
+
}
|
configuration.py
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import PretrainedConfig
|
2 |
+
from typing import List
|
3 |
+
|
4 |
+
|
5 |
+
class ThaiLightWeightEncoderConfig(PretrainedConfig):
|
6 |
+
|
7 |
+
def __init__(
|
8 |
+
self,
|
9 |
+
input_embedding_dim: int = 300,
|
10 |
+
final_embedding_dim: int = 512,
|
11 |
+
dropout: float = 0.2,
|
12 |
+
word_vector_model_name: str = "thai2fit_wv",
|
13 |
+
**kwargs,
|
14 |
+
):
|
15 |
+
self.input_embedding_dim = input_embedding_dim
|
16 |
+
self.final_embedding_dim = final_embedding_dim
|
17 |
+
self.word_vector_model_name = word_vector_model_name
|
18 |
+
self.dropout = dropout
|
19 |
+
super().__init__(**kwargs)
|
encoder.py
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import PreTrainedModel
|
2 |
+
|
3 |
+
from pythainlp import word_vector
|
4 |
+
import torch
|
5 |
+
|
6 |
+
from .configuration import ThaiLightWeightEncoderConfig
|
7 |
+
from .projector import Projector
|
8 |
+
|
9 |
+
|
10 |
+
class ThaiLightWeightEncoderModel(PreTrainedModel):
|
11 |
+
config_class = ThaiLightWeightEncoderConfig
|
12 |
+
|
13 |
+
def __init__(self, config):
|
14 |
+
super().__init__(config)
|
15 |
+
self.wv = word_vector.WordVector(model_name=config.word_vector_model_name)
|
16 |
+
self.projector = Projector(
|
17 |
+
input_embedding_dim=config.input_embedding_dim,
|
18 |
+
final_embedding_dim=config.final_embedding_dim,
|
19 |
+
dropout=config.dropout
|
20 |
+
)
|
21 |
+
|
22 |
+
def forward(self, text: str):
|
23 |
+
embed = self.wv.sentence_vectorizer(text, use_mean=True)[0]
|
24 |
+
proj_embed = self.projector(torch.from_numpy(embed).float())
|
25 |
+
proj_embed = proj_embed.to("cpu").detach().numpy()
|
26 |
+
return proj_embed
|
projector.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
import numpy as np
|
4 |
+
|
5 |
+
|
6 |
+
class Projector(nn.Module):
|
7 |
+
def __init__(
|
8 |
+
self,
|
9 |
+
input_embedding_dim: int = 300,
|
10 |
+
final_embedding_dim: int = 512,
|
11 |
+
dropout: float = 0.2
|
12 |
+
):
|
13 |
+
super().__init__()
|
14 |
+
self.fc1 = nn.Linear(input_embedding_dim, 512)
|
15 |
+
self.fn1 = nn.LeakyReLU()
|
16 |
+
self.fc2 = nn.Linear(512, final_embedding_dim)
|
17 |
+
self.fn2 = nn.LeakyReLU()
|
18 |
+
self.dropout = nn.Dropout(dropout)
|
19 |
+
self.layer_norm = nn.LayerNorm(final_embedding_dim)
|
20 |
+
self.temperature = nn.Parameter(torch.ones([]) * np.log(1 / 0.07))
|
21 |
+
|
22 |
+
def forward(self, x):
|
23 |
+
x = self.fc1(x)
|
24 |
+
x = self.fn1(x)
|
25 |
+
x = self.dropout(x)
|
26 |
+
x = self.fc2(x)
|
27 |
+
x = self.fn2(x)
|
28 |
+
x = self.layer_norm(x)
|
29 |
+
return x
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4161f95136d9f4410be6b2508b60610445bc7d07f84c1f1ae2cc47eae2c755a6
|
3 |
+
size 1673609
|