File size: 5,411 Bytes
f378e98 320dfe6 02fe10e f378e98 320dfe6 f378e98 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d f378e98 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 687f82d 320dfe6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 |
---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
datasets:
- sbx/superlim-2
language:
- sv
widget:
- source_sentence: "Mannen åt mat."
sentences:
- "Han förtärde en närande och nyttig måltid."
- "Det var ett sunkigt hak med ganska gott käk."
- "Han inmundigade middagen tillsammans med ett glas rödvin."
- "Potatischips är jättegoda."
- "Tryck på knappen för att få tala med kundsupporten."
example_title: "Mat"
- source_sentence: "Kan jag deklarera digitalt från utlandet?"
sentences:
- "Du som befinner dig i utlandet kan deklarera digitalt på flera olika sätt."
- "Du som har kvarskatt att betala ska göra en inbetalning till ditt skattekonto."
- "Efter att du har deklarerat går vi igenom uppgifterna i din deklaration och räknar ut din skatt."
- "I din deklaration som du får från oss har vi räknat ut vad du ska betala eller få tillbaka."
- "Tryck på knappen för att få tala med kundsupporten."
example_title: "Skatteverket FAQ"
- source_sentence: "Hon kunde göra bakåtvolter."
sentences:
- "Hon var atletisk."
- "Hon var bra på gymnastik."
- "Hon var inte atletisk."
- "Hon var oförmögen att flippa baklänges."
example_title: "Gymnastik"
---
# jzju/sbert-sv-lim2
This model Is trained from [KBLab/bert-base-swedish-cased-new](https://huggingface.co/KBLab/bert-base-swedish-cased-new) with data from [sbx/superlim-2](https://huggingface.co/datasets/sbx/superlim-2)
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 256 dimensional dense vector space and can be used for tasks like clustering or semantic search.
## Usage (Sentence-Transformers)
Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
```
pip install -U sentence-transformers
```
Then you can use the model like this:
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('jzju/sbert-sv-lim2')
embeddings = model.encode(sentences)
print(embeddings)
```
## Training Code
```python
from datasets import load_dataset, concatenate_datasets
from sentence_transformers import (
SentenceTransformer,
InputExample,
losses,
models,
util,
datasets,
)
from torch.utils.data import DataLoader
from torch import nn
import random
word_embedding_model = models.Transformer(
"KBLab/bert-base-swedish-cased-new", max_seq_length=256
)
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension())
dense_model = models.Dense(
in_features=pooling_model.get_sentence_embedding_dimension(),
out_features=256,
activation_function=nn.Tanh(),
)
model = SentenceTransformer(modules=[word_embedding_model, pooling_model, dense_model])
def pair():
def norm(x):
x["label"] = x["label"] / m
return x
dd = []
for sub in ["swepar", "swesim_relatedness", "swesim_similarity"]:
ds = concatenate_datasets(
[d for d in load_dataset("sbx/superlim-2", sub).values()]
)
if "sentence_1" in ds.features:
ds = ds.rename_column("sentence_1", "d1")
ds = ds.rename_column("sentence_2", "d2")
else:
ds = ds.rename_column("word_1", "d1")
ds = ds.rename_column("word_2", "d2")
m = max([d["label"] for d in ds])
dd.append(ds.map(norm))
ds = concatenate_datasets(dd)
train_examples = []
for d in ds:
train_examples.append(InputExample(texts=[d["d1"], d["d2"]], label=d["label"]))
train_dataloader = DataLoader(train_examples, shuffle=True, batch_size=64)
train_loss = losses.CosineSimilarityLoss(model)
model.fit(
train_objectives=[(train_dataloader, train_loss)], epochs=10, warmup_steps=100
)
def nli():
ds = concatenate_datasets(
[d for d in load_dataset("sbx/superlim-2", "swenli").values()]
)
def add_to_samples(sent1, sent2, label):
if sent1 not in train_data:
train_data[sent1] = {0: set(), 1: set(), 2: set()}
train_data[sent1][label].add(sent2)
train_data = {}
for d in ds:
add_to_samples(d["premise"], d["hypothesis"], d["label"])
add_to_samples(d["hypothesis"], d["premise"], d["label"])
train_samples = []
for sent1, others in train_data.items():
if len(others[0]) > 0 and len(others[1]) > 0:
train_samples.append(
InputExample(
texts=[
sent1,
random.choice(list(others[0])),
random.choice(list(others[1])),
]
)
)
train_samples.append(
InputExample(
texts=[
random.choice(list(others[0])),
sent1,
random.choice(list(others[1])),
]
)
)
train_dataloader = datasets.NoDuplicatesDataLoader(train_samples, batch_size=64)
train_loss = losses.MultipleNegativesRankingLoss(model)
model.fit(
train_objectives=[(train_dataloader, train_loss)], epochs=1, warmup_steps=100
)
pair()
nli()
model.save()
``` |