This pull request introduces a new app called gongwen_english, designed to improve the time efficiency of official document generation.
Key Features
1. Cooperation with TAIDE and Llama3.1
TAIDE: Excellent for refining official documents.
TAIDE is particularly useful in the final stage for revising plain language into formal vocabulary in Chinese. We will detail the format improvements in the few-shot section.
Llama3.1: Ideal for generating content and summaries.
-Llama3.1 is responsible for expanding user input, generating the main topics, and providing relevant information.
2. Flexible Stages
Function List:
After the initial conversation with the bot, users will receive a message showing the correspondence between numbers and functions.
-If a generated result does not meet user expectations, steps can be skipped to quickly regenerate a new result.
Lower Cost for Mistakes:
During development, we noticed that canceling a selected choice was time-consuming. Therefore, we designed a method to improve the user experience by allowing easy retraction of choices (using '/n').
3. Few-shot Learning for Officialization
We experimented with multiple prompts for better official document formatting throughout development. Instead of giving detailed instructions, showing examples proved to be more effective.
For each different process, three examples will be randomly selected from ten available examples.
4. Future Improvement Path
The primary improvement focus will be on the final officialization stage.
Provide better few-shot examples to ensure more formal language and structure.
Explore the use of Multi-Agent systems to generate necessary content and eliminate redundant information.
Consider parallelizing the Main Topic and Information stages.
This pull request introduces a new app called gongwen_english, designed to improve the time efficiency of official document generation.
Key Features
1. Cooperation with TAIDE and Llama3.1
2. Flexible Stages
3. Few-shot Learning for Officialization
4. Future Improvement Path
Flowchart for
gongwen_eng