chenjh356 commited on
Commit
124f987
·
verified ·
1 Parent(s): 71d08bf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +106 -3
README.md CHANGED
@@ -1,3 +1,106 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - biology
5
+ ---
6
+
7
+ <div align="center">
8
+ <img src="https://raw.githubusercontent.com/BGI-HangzhouAI/Genos/main/images/Genos_model.png" width="100%" />
9
+ </div>
10
+
11
+ # Genos
12
+
13
+ Genos, as a foundational model in the field of human genomics, trained on hundreds of high-quality genome reference data, has achieved the ability to contextually model human genome sequences up to millions of base pairs. Through single-base resolution learning, this model possesses the capability to identify hidden deep sequence patterns and functional features within genomes, providing scientists with a new research method that connects genetic information with life activities.
14
+
15
+ For instructions, details, and examples, please refer to the [Genos GitHub](https://github.com/BGI-HangzhouAI/Genos).
16
+
17
+ Below are the data volume of our model training and related parameters.
18
+
19
+ <table align="center">
20
+ <tr>
21
+ <th>Model Specification</th>
22
+ <th>Genos 1.2B</th>
23
+ <th>Genos 10B</th>
24
+ </tr>
25
+
26
+ <!-- Model Scale category title - span 3 columns -->
27
+ <tr>
28
+ <td colspan="3" align="center"><b>Model Scale</b></td>
29
+ </tr>
30
+ <tr>
31
+ <td>Total Parameters</td>
32
+ <td>1.2B</td>
33
+ <td>10B</td>
34
+ </tr>
35
+ <tr>
36
+ <td>Activated Parameters</td>
37
+ <td>0.33B</td>
38
+ <td>2.87B</td>
39
+ </tr>
40
+ <tr>
41
+ <td>Trained Tokens</td>
42
+ <td>1600 B</td>
43
+ <td>2200 B</td>
44
+ </tr>
45
+
46
+ <!-- Architecture category title - span 3 columns -->
47
+ <tr>
48
+ <td colspan="3" align="center"><b>Architecture</b></td>
49
+ </tr>
50
+ <tr>
51
+ <td>Architecture Type</td>
52
+ <td>MoE</td>
53
+ <td>MoE</td>
54
+ </tr>
55
+ <tr>
56
+ <td>Number of Experts</td>
57
+ <td>8</td>
58
+ <td>8</td>
59
+ </tr>
60
+ <tr>
61
+ <td>Selected Experts per Token</td>
62
+ <td>2</td>
63
+ <td>2</td>
64
+ </tr>
65
+ <tr>
66
+ <td>Number of Layers</td>
67
+ <td>12</td>
68
+ <td>12</td>
69
+ </tr>
70
+ <tr>
71
+ <td>Attention Hidden Dimension</td>
72
+ <td>1024</td>
73
+ <td>4096</td>
74
+ </tr>
75
+ <tr>
76
+ <td>Number of Attention Heads</td>
77
+ <td>16</td>
78
+ <td>16</td>
79
+ </tr>
80
+ <tr>
81
+ <td>MoE Hidden Dimension (per Expert)</td>
82
+ <td>4096</td>
83
+ <td>8192</td>
84
+ </tr>
85
+ <tr>
86
+ <td>Vocabulary Size</td>
87
+ <td>128 (padded)</td>
88
+ <td>256 (padded)</td>
89
+ </tr>
90
+ <tr>
91
+ <td>Context Length</td>
92
+ <td>up to 1M</td>
93
+ <td>up to 1M</td>
94
+ </tr>
95
+ </table>
96
+
97
+
98
+ Genos 1.2B and 10B checkpoints are available here:
99
+
100
+ - [Genos-1.2B](https://huggingface.co/BGI-HangzhouAI/Genos-1.2B)
101
+ - [Genos-10B](https://huggingface.co/BGI-HangzhouAI/Genos-10B)
102
+
103
+ We also provide checkpoints trained under the [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) framework:
104
+
105
+ - [Genos-Megatron-1.2B](https://huggingface.co/BGI-HangzhouAI/Genos-Megatron-1.2B)
106
+ - [Genos-Megatron-10B](https://huggingface.co/BGI-HangzhouAI/Genos-Megatron-10B)