philipp-zettl
commited on
Commit
•
6cc2b88
1
Parent(s):
62e2378
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,9 +1,22 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
3 |
tags: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
---
|
5 |
|
6 |
-
# Model Card for
|
7 |
|
8 |
<!-- Provide a quick summary of what the model is/does. -->
|
9 |
|
@@ -15,21 +28,38 @@ tags: []
|
|
15 |
|
16 |
<!-- Provide a longer summary of what this model is. -->
|
17 |
|
18 |
-
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
|
19 |
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
- **Funded by [optional]:** [More Information Needed]
|
22 |
- **Shared by [optional]:** [More Information Needed]
|
23 |
- **Model type:** [More Information Needed]
|
24 |
-
- **Language(s) (NLP):**
|
25 |
-
- **License:**
|
26 |
-
- **Finetuned from model [optional]:**
|
27 |
|
28 |
### Model Sources [optional]
|
29 |
|
30 |
<!-- Provide the basic links for the model. -->
|
31 |
|
32 |
-
- **Repository:** [
|
33 |
- **Paper [optional]:** [More Information Needed]
|
34 |
- **Demo [optional]:** [More Information Needed]
|
35 |
|
@@ -41,7 +71,7 @@ This is the model card of a 🤗 transformers model that has been pushed on the
|
|
41 |
|
42 |
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
43 |
|
44 |
-
|
45 |
|
46 |
### Downstream Use [optional]
|
47 |
|
@@ -53,7 +83,11 @@ This is the model card of a 🤗 transformers model that has been pushed on the
|
|
53 |
|
54 |
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
55 |
|
56 |
-
|
|
|
|
|
|
|
|
|
57 |
|
58 |
## Bias, Risks, and Limitations
|
59 |
|
@@ -71,7 +105,28 @@ Users (both direct and downstream) should be made aware of the risks, biases and
|
|
71 |
|
72 |
Use the code below to get started with the model.
|
73 |
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
## Training Details
|
77 |
|
@@ -79,7 +134,11 @@ Use the code below to get started with the model.
|
|
79 |
|
80 |
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
81 |
|
82 |
-
|
|
|
|
|
|
|
|
|
83 |
|
84 |
### Training Procedure
|
85 |
|
@@ -87,7 +146,24 @@ Use the code below to get started with the model.
|
|
87 |
|
88 |
#### Preprocessing [optional]
|
89 |
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
|
93 |
#### Training Hyperparameters
|
@@ -136,7 +212,79 @@ Use the code below to get started with the model.
|
|
136 |
|
137 |
<!-- Relevant interpretability work for the model goes here -->
|
138 |
|
139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
|
141 |
## Environmental Impact
|
142 |
|
|
|
1 |
---
|
2 |
+
language: multilingual
|
3 |
+
license: mit
|
4 |
+
library_name: torch
|
5 |
tags: []
|
6 |
+
base_model: BAAI/bge-m3
|
7 |
+
datasets: philipp-zettl/GGU-xx
|
8 |
+
metrics:
|
9 |
+
- accuracy
|
10 |
+
- f1
|
11 |
+
- recall
|
12 |
+
model_name: GGU-CLF
|
13 |
+
pipeline_tag: text-classification
|
14 |
+
widget:
|
15 |
+
- name: test1
|
16 |
+
text: hello world
|
17 |
---
|
18 |
|
19 |
+
# Model Card for GGU-CLF
|
20 |
|
21 |
<!-- Provide a quick summary of what the model is/does. -->
|
22 |
|
|
|
28 |
|
29 |
<!-- Provide a longer summary of what this model is. -->
|
30 |
|
|
|
31 |
|
32 |
+
This is a simple classification model trained on a custom dataset.
|
33 |
+
|
34 |
+
Please note that this model, although it is implemented in the `transformers` library. Is not a usual transformer.
|
35 |
+
It combines the underlying embedding model with the required tokenizer into a simple-to-use pipeline for sequence classification.
|
36 |
+
|
37 |
+
It is used to classify user text into the following classes:
|
38 |
+
- 0: Greeting
|
39 |
+
- 1: Gratitude
|
40 |
+
- 2: Unknown
|
41 |
+
|
42 |
+
**Note**: To use this model please remember the following things
|
43 |
+
|
44 |
+
1. The model is an XLMRoberta model based on [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3).
|
45 |
+
2. The required tokenizer is baked into the classifier implementation.
|
46 |
+
|
47 |
+
|
48 |
+
|
49 |
+
|
50 |
+
- **Developed by:** [philipp-zettl](https://huggingface.co/philipp-zettl/)
|
51 |
- **Funded by [optional]:** [More Information Needed]
|
52 |
- **Shared by [optional]:** [More Information Needed]
|
53 |
- **Model type:** [More Information Needed]
|
54 |
+
- **Language(s) (NLP):** multilingual
|
55 |
+
- **License:** mit
|
56 |
+
- **Finetuned from model [optional]:** BAAI/bge-m3
|
57 |
|
58 |
### Model Sources [optional]
|
59 |
|
60 |
<!-- Provide the basic links for the model. -->
|
61 |
|
62 |
+
- **Repository:** [philipp-zettl/GGU-CLF](https://huggingface.co/philipp-zettl/GGU-CLF)
|
63 |
- **Paper [optional]:** [More Information Needed]
|
64 |
- **Demo [optional]:** [More Information Needed]
|
65 |
|
|
|
71 |
|
72 |
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
73 |
|
74 |
+
Use this model to classify messages from natural language chats.
|
75 |
|
76 |
### Downstream Use [optional]
|
77 |
|
|
|
83 |
|
84 |
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
85 |
|
86 |
+
|
87 |
+
The model was not trained on multi-sentence samples. **You should avoid those.**
|
88 |
+
|
89 |
+
Oficially tested and supported languages are **english and german** any other language is considered out of scope.
|
90 |
+
|
91 |
|
92 |
## Bias, Risks, and Limitations
|
93 |
|
|
|
105 |
|
106 |
Use the code below to get started with the model.
|
107 |
|
108 |
+
|
109 |
+
```python
|
110 |
+
from transformers import AutoModel
|
111 |
+
|
112 |
+
model = AutoModel.from_pretrained("philipp-zettl/GGU-xx").to(torch.float16).to('cuda')
|
113 |
+
|
114 |
+
model([
|
115 |
+
'Hi wie gehts?',
|
116 |
+
'Dannke dir mein freund!',
|
117 |
+
'Merci freundchen, send mir mal ein paar Machine Learning jobs.',
|
118 |
+
'Works as expected, cheers!',
|
119 |
+
'How you doin my boy',
|
120 |
+
'send me immediately some matching jobs, thanks',
|
121 |
+
'wer's eigentlich tom selleck?',
|
122 |
+
'sprichst du deutsch?',
|
123 |
+
'sprechen sie deutsch sie hurensohn?',
|
124 |
+
'vergeltsgott',
|
125 |
+
'heidenei dank dir recht herzlich',
|
126 |
+
'grazie mille bambino, come estas'
|
127 |
+
]).argmax(dim=1)
|
128 |
+
```
|
129 |
+
|
130 |
|
131 |
## Training Details
|
132 |
|
|
|
134 |
|
135 |
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
136 |
|
137 |
+
|
138 |
+
This model was trained using the [philipp-zettl/GGU-xx](https://huggingface.co/dataset/philipp-zettl/GGU-xx) dataset.
|
139 |
+
|
140 |
+
You can find it's performance metrics under [Evaluation Results](#evaluation-results).
|
141 |
+
|
142 |
|
143 |
### Training Procedure
|
144 |
|
|
|
146 |
|
147 |
#### Preprocessing [optional]
|
148 |
|
149 |
+
|
150 |
+
The following code was used to create the data set as well as split the data set into training and validation sets.
|
151 |
+
|
152 |
+
```python
|
153 |
+
from datasets import load_dataset
|
154 |
+
|
155 |
+
class Dataset:
|
156 |
+
def __init__(self, dataset, target_names=None):
|
157 |
+
self.data = list(map(lambda x: x[0], dataset))
|
158 |
+
self.target = list(map(lambda x: x[1], dataset))
|
159 |
+
self.target_names = target_names
|
160 |
+
|
161 |
+
|
162 |
+
ds = load_dataset('philipp-zettl/GGU-xx')
|
163 |
+
data = Dataset([[e['sample'], e['label']] for e in ds['train']], ['greeting', 'gratitude', 'unknown'])
|
164 |
+
X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.2, random_state=42)
|
165 |
+
```
|
166 |
+
|
167 |
|
168 |
|
169 |
#### Training Hyperparameters
|
|
|
212 |
|
213 |
<!-- Relevant interpretability work for the model goes here -->
|
214 |
|
215 |
+
|
216 |
+
You can find the initial implementation of the classification model here:
|
217 |
+
|
218 |
+
```python
|
219 |
+
from transformers import PreTrainedModel, PretrainedConfig, AutoModel, AutoTokenizer
|
220 |
+
import torch
|
221 |
+
import torch.nn as nn
|
222 |
+
|
223 |
+
|
224 |
+
class EmbeddingClassifierConfig(PretrainedConfig):
|
225 |
+
model_type = 'xlm-roberta'
|
226 |
+
|
227 |
+
def __init__(self, num_classes=3, base_model='BAAI/bge-m3', tokenizer='BAAI/bge-m3', dropout=0.0, l2_reg=0.01, torch_dtype=torch.float16, **kwargs):
|
228 |
+
self.num_classes = num_classes
|
229 |
+
self.base_model = base_model
|
230 |
+
self.tokenizer = tokenizer
|
231 |
+
self.dropout = dropout
|
232 |
+
self.l2_reg = l2_reg
|
233 |
+
self.torch_dtype = torch_dtype
|
234 |
+
super().__init__(**kwargs)
|
235 |
+
|
236 |
+
|
237 |
+
class EmbeddingClassifier(PreTrainedModel):
|
238 |
+
config_class = EmbeddingClassifierConfig
|
239 |
+
|
240 |
+
def __init__(self, config):
|
241 |
+
super().__init__(config)
|
242 |
+
base_model = config.base_model
|
243 |
+
tokenizer = config.tokenizer
|
244 |
+
|
245 |
+
if base_model is None or isinstance(tokenizer, str):
|
246 |
+
base_model = AutoModel.from_pretrained(base_model)#, torch_dtype=config.torch_dtype)
|
247 |
+
if tokenizer is None or isinstance(tokenizer, str):
|
248 |
+
tokenizer = AutoTokenizer.from_pretrained(tokenizer)
|
249 |
+
|
250 |
+
self.tokenizer = tokenizer
|
251 |
+
self.base = base_model
|
252 |
+
self.fc = nn.Linear(base_model.config.hidden_size, config.num_classes)#, torch_dtype=config.torch_dtype)
|
253 |
+
self.do = nn.Dropout(config.dropout)#, torch_dtype=config.torch_dtype)
|
254 |
+
self.l2_reg = config.l2_reg
|
255 |
+
|
256 |
+
self.to(config.torch_dtype)
|
257 |
+
|
258 |
+
def forward(self, X):
|
259 |
+
encoding = self.tokenizer(
|
260 |
+
X, return_tensors='pt',
|
261 |
+
padding=True, truncation=True
|
262 |
+
).to(self.device)
|
263 |
+
input_ids = encoding['input_ids']
|
264 |
+
attention_mask = encoding['attention_mask']
|
265 |
+
emb = self.base(
|
266 |
+
input_ids,
|
267 |
+
attention_mask=attention_mask,
|
268 |
+
return_dict=True,
|
269 |
+
output_hidden_states=True
|
270 |
+
).last_hidden_state[:, 0, :]
|
271 |
+
return self.fc(self.do(emb))
|
272 |
+
|
273 |
+
def train(self, set_val=True):
|
274 |
+
self.base.train(False)
|
275 |
+
for param in self.base.parameters():
|
276 |
+
param.requires_grad = False
|
277 |
+
for param in self.fc.parameters():
|
278 |
+
param.requires_grad = set_val
|
279 |
+
|
280 |
+
def get_l2_loss(self):
|
281 |
+
l2_loss = torch.tensor(0.).to('cuda')
|
282 |
+
for param in self.parameters():
|
283 |
+
if param.requires_grad:
|
284 |
+
l2_loss += torch.norm(param, 2)
|
285 |
+
return self.l2_reg * l2_loss
|
286 |
+
```
|
287 |
+
|
288 |
|
289 |
## Environmental Impact
|
290 |
|