Skip to content

Commit

Permalink
deploy: 58b8fca
Browse files Browse the repository at this point in the history
  • Loading branch information
p208p2002 committed Dec 13, 2023
1 parent e209b26 commit 828531a
Show file tree
Hide file tree
Showing 3 changed files with 5 additions and 5 deletions.
4 changes: 2 additions & 2 deletions _version.txt
Original file line number Diff line number Diff line change
@@ -1,3 +1,3 @@
Git Hash: 20a63ac5e063454b1886944a9c81b5afdb6be828
Build Time: 2023/12/13 03:01:46
Git Hash: 58b8fcad17058ed5f6515e61089b279559e5e317
Build Time: 2023/12/13 03:12:16
Build Platform: Linux-6.2.0-1018-azure-x86_64-with-glibc2.35
2 changes: 1 addition & 1 deletion docs/fill-in-middle/document.md
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,7 @@ Fill-in-the-middle(FIM)訓練模型的實用性、簡單性和效率,作

![image](./1.png)
![image](./2.png)
黃色線段(0.5)表示有50%的語料使用 FIM方法 (),紫色線段表示常規的預訓練。
黃色線段(0.5)表示有50%的語料使用 FIM方法 (FIM rate),紫色線段表示常規的預訓練。

僅僅考慮測試損失通常是不足夠的。為了加強上述結果,作者在一套標準下游基準測試中評估了作者的模型,其結果呈現在上圖。作者再次發現,Joint FIM pre-training 不會導致在標準AR基準測試中出現性能降級,因為性能在自然語言和 code 方面都在誤差範圍內匹配。

Expand Down
4 changes: 2 additions & 2 deletions index.json
Original file line number Diff line number Diff line change
Expand Up @@ -15,9 +15,9 @@
"token",
"訓練",
"prefix",
"rate",
"預訓練",
"學習",
"suffix"
"學習"
],
"page_link": "https://blog.philip-huang.tech?page=fill-in-middle",
"file_link": "https://blog.philip-huang.tech/docs/fill-in-middle/document.md",
Expand Down

0 comments on commit 828531a

Please sign in to comment.