victor70 commited on
Commit
4869aab
·
verified ·
1 Parent(s): d74e952

Add model card

Browse files
Files changed (1) hide show
  1. README.md +50 -0
README.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - ko
4
+ library_name: pytorch
5
+ tags:
6
+ - hybridko
7
+ - korean
8
+ - rnn
9
+ - attention
10
+ - griffin
11
+ ---
12
+
13
+ # HybriKo - Korean Hybrid LLM
14
+
15
+ Griffin-inspired hybrid architecture combining RNN and Attention mechanisms.
16
+
17
+ ## Model Details
18
+
19
+ - **Architecture**: Hybrid RNN + Attention (2:1 ratio)
20
+ - **Parameters**: 117.8M
21
+ - **Training**: Continued pretraining on exp4_plus dataset
22
+ - **Base Model**: exp4 (Wikipedia pretrained)
23
+
24
+ ## Training Data
25
+
26
+ - korean_textbooks_tiny (50K samples)
27
+ - korean_textbooks_edu (50K samples)
28
+ - korean_public_corpus (50K samples)
29
+
30
+ ## Usage
31
+
32
+ ```python
33
+ from hybridko.model import HybriKoModel, HybriKoConfig
34
+
35
+ config = HybriKoConfig.from_yaml("config.yaml")
36
+ model = HybriKoModel(config)
37
+
38
+ # Load checkpoint
39
+ checkpoint = torch.load("checkpoints/checkpoint_step_XXX.pt")
40
+ model.load_state_dict(checkpoint["model_state_dict"])
41
+ ```
42
+
43
+ ## Citation
44
+
45
+ ```bibtex
46
+ @misc{hybridko2024,
47
+ title={HybriKo: Korean Hybrid LLM},
48
+ year={2024},
49
+ }
50
+ ```