Chang-chih/leviathan-10.5b-final
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: Chang-chih/leviathan-10.5b-final
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-11-01 04:14:18 +08:00
commit eacb329930
11 changed files with 499523 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
⚙️ 蒸餾的實際流程 Normal file
View File

@@ -0,0 +1,4 @@
graph LR
A[16B 導師模型<br>Chang-chih/leviathan-16B-final] -->|產生 Soft Labels| B[蒸餾損失函數<br>KL Divergence]
C[10.5B 學生模型<br>繼承 7B 靈魂 + DeepSeek 骨架] -->|產生 Student Logits| B
B -->|反向傳播| C