dearwolf/aider
1
0
Fork 0
mirror of https://github.com/Aider-AI/aider.git synced 2025-05-25 14:55:00 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

copy

Browse source
This commit is contained in:
Paul Gauthier 2023-12-17 18:23:45 -08:00
parent 9807dbdeb1
commit 8c9536bcdd
1 changed files with 4 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

15
docs/unified-diffs.md
View file

@ -27,9 +27,9 @@ This new laziness benchmark produced the following results with `gpt-4-1106-prev
- **Aider's new unified diff edit format raised the score to 65%**. - **Aider's new unified diff edit format raised the score to 65%**.
- **No benefit from the user being blind, without hands, tipping $2000 or fearing truncated code trauma.** These widely circulated folk remedies performed no better than baseline when added to the system prompt with aider's SEARCH/REPLACE edit format. Including *all* of them only scored at 15% - **No benefit from the user being blind, without hands, tipping $2000 or fearing truncated code trauma.** These widely circulated folk remedies performed no better than baseline when added to the system prompt with aider's SEARCH/REPLACE edit format. Including *all* of them only scored at 15%
The older `gpt-4-0613` also did better on the laziness benchmark by using unified diffs. The older `gpt-4-0613` also did better on the laziness benchmark using unified diffs.
The benchmark was designed to work with large source code files, and The benchmark was designed to work with large source code files, and
28% of them are too large to use with June GPT-4's 8k context window. 28% of them are too large to fit in June GPT-4's 8k context window.
This significantly harmed the benchmark results. This significantly harmed the benchmark results.
- **The June GPT-4's baseline was 26%** using aider's existing "SEARCH/REPLACE block" edit format. - **The June GPT-4's baseline was 26%** using aider's existing "SEARCH/REPLACE block" edit format.
@ -86,7 +86,6 @@ default output format of `git diff`:
@@ -1,5 +1,5 @@ @@ -1,5 +1,5 @@
def main(args): def main(args):
# show a greeting # show a greeting
- print("Hello!") - print("Hello!")
+ print("Goodbye!") + print("Goodbye!")
return return
@ -149,7 +148,7 @@ numbers in editing formats,
backed up by many quantitative benchmark experiments. backed up by many quantitative benchmark experiments.
You've probably ignored the line numbers in every diff you've seen? You've probably ignored the line numbers in every diff you've seen?
So aider tells GPT not to include line numbers, So aider tells GPT not to include them,
and just interprets each hunk from the unified diffs and just interprets each hunk from the unified diffs
as a search and replace operation: as a search and replace operation:
@ -159,19 +158,17 @@ This diff:
@@ ... @@ @@ ... @@
def main(args): def main(args):
# show a greeting # show a greeting
- print("Hello!") - print("Hello!")
+ print("Goodbye!") + print("Goodbye!")
return return
``` ```
Means we want to search the original source file for all the Means we want to search the file for all the
*space* ` ` and *minus* `-` lines from the hunk: *space* ` ` and *minus* `-` lines from the hunk:
```python ```python
def main(args): def main(args):
# show a greeting # show a greeting
print("Hello!") print("Hello!")
return return
``` ```
@ -181,7 +178,6 @@ And then replace them with all the *space* ` ` and *plus* `+` lines:
```python ```python
def main(args): def main(args):
# show a greeting # show a greeting
print("Goodbye!") print("Goodbye!")
return return
``` ```
@ -267,7 +263,6 @@ import sys
def main(args): def main(args):
# show a greeting # show a greeting
print("Hello!") print("Hello!")
return return
@ -284,11 +279,9 @@ because of the missing comment.
```diff ```diff
@@ ... @@ @@ ... @@
-def main(args): -def main(args):
-
- print("Hello!") - print("Hello!")
- return - return
+def main(args): +def main(args):
+
+ print("Goodbye!") + print("Goodbye!")
+ return + return
``` ```