data-generation-hyper 0.0.5.4

Library for synthetic data generation

Code Generation using main_script.py

This script is designed to generate code snippets using various parameters. You can customize the parameters directly in the script or by modifying the JSON file in the config folder. The generated codes are less than the requested because the algorithm drops duplicated ones. Inside prompts folder/prompt_list_and_probabilities can be found the probabilities with which the four different prompts templates are selected, the proportions can be changed in this file.

Parameters

• api_key: Your API key for accessing the code generation service.
• share: A dictionary specifying the share of snippets for each language.
• Total_number: Total number of snippets to extract.
• batch_size: Batch size for parallel processing.
• n_jobs: Number of parallel jobs to run (null for automatic detection).
• model: The model to use for code generation.
• temperature_problem: Temperature for creating the problem text.
• temperature_solution: Temperature for creating the solution text.
• test: Whether to run in test mode (produces only 6 example generated samples).

Default Values

Here are the default values for each parameter:

• api_key: "aaa"
• share: {"Python" : 0.40, "C++" : 0.05, ... }
• Total_number: 100
• batch_size: 3
• n_jobs: null
• model: "meta-llama/Llama-3-70b-chat-hf"
• temperature_problem: 0.7
• temperature_solution: 0.5
• test: false

Running the Script

To run the script with the default parameters, use the following command:

python app/main_coder.py

## Author
- Onur Alp Güvercin
- Cesare Bidini