poperskop commited on
Commit
27d130e
·
verified ·
1 Parent(s): 14d8e8c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +33 -1
README.md CHANGED
@@ -7,4 +7,36 @@ sdk: static
7
  pinned: false
8
  ---
9
 
10
- Edit this `README.md` markdown file to author your organization card.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  pinned: false
8
  ---
9
 
10
+ ## Description:
11
+ Our goal was to make PoC solution for matching masseges from telegramm marketplaces.
12
+
13
+ There is 2 models:
14
+ - RoSBERTa-hermes-ru trained for **location recognition**, **catogories labeling** and **inside-outside location classififcation**;
15
+ - rubert-tiny-separater trained for **supply and demand** classification
16
+
17
+
18
+ ## Architecture and Pretraining
19
+
20
+ # (RoSBERTa-hermes-ru)[https://huggingface.co/poc-embeddings/RoSBERTa-hermes-ru]
21
+
22
+ Rosberta has (ai-forever/ru-en-RoSBERTa)[https://huggingface.co/ai-forever/ru-en-RoSBERTa] as backbone with multiple heads for downstream tasks on top
23
+ - **backbone** fully unfreezed and **NER head** was fine-tuned for location recognition;
24
+ - **alocator head** was trained for decide whether or not message contain actuall location of user;
25
+ - **adapter** and **tags head** was train to mark message with deffirent categories describing the messsage whereabouts such as tools, medecine, cloth and etc.
26
+
27
+ # (rubert-tiny-separater)[https://huggingface.co/poc-embeddings/rubert-tiny-separater]
28
+
29
+ rubert has (sergeyzh/rubert-tiny-turbo)[https://huggingface.co/sergeyzh/rubert-tiny-turbo] as backbone with Linear layer on top.
30
+ Whole model was trained for classification messages type from telegram marketplaces.
31
+
32
+ Labels:
33
+
34
+ - **supply**: somebody willing to sell something or provide service
35
+ - **demand**: somebody wants to buy something or hire somebody
36
+ - **noise**: messages unrelated to topic.
37
+
38
+ ## Supported Languages
39
+
40
+ Russian with English in it.
41
+
42
+