apunkt/MetaGPT
1
0
Fork 0
mirror of https://github.com/FoundationAgents/MetaGPT.git synced 2026-04-25 16:56:26 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

add annotations

Browse source
This commit is contained in:
黄伟韬 2024-08-12 20:07:35 +08:00
parent be6c3b4455
commit 90e1e53bb6
1 changed files with 17 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

17
metagpt/utils/repair_llm_raw_output.py
View file

@ -353,6 +353,23 @@ def repair_escape_error(commands):
"""
Repaires escape errors in command responses.
When role-zero parses a command, the command may contain unknown escape characters.
This function has two steps:
1. Transform unescaped substrings like "\d" and "\(" to "\\\\d" and "\\\\(".
2. Transform escaped characters like '\f' to substrings like "\\\\f".
Example:
When the original JSON string is " {"content":"\\\\( \\\\frac{1}{2} \\\\)"} ",
The "content" will be parsed correctly to "\( \frac{1}{2} \)".
When there is a wrong JSON string like: " {"content":"\( \frac{1}{2} \)"}",
It will cause a parsing error.
To repair the wrong JSON string, the following transformations will be used:
"\(" ---> "\\\\("
'\f' ---> "\\\\f"
"\)" ---> "\\\\)"
"""
escape_repair_map = {
"\a": "\\\\a",