日本語

Evolve-Instruct

Overview

The Evolve Instruct method creates more diverse and complex instructions by modifying (augmenting) existing instruction data. To achieve this, this technique utilizes LLMs such as GPT-4o to rewrite or transform existing instructions. In particular, It uses two strategies to make instructions more complex or create new instructions: In-depth Evolving and In-breadth Evolving.

• In-depth Evolving: Make an instruction more difficult by adding constraints to it, making it more specific, increasing the logical reasoning steps, or complicating the input.
• In-breadth Evolving: Create completely new commands based on existing commands to expand the scope of topics and technologies and increase the diversity of your datasets.

Implementation

This open-source implementation is based on the WizardLM paper and h2o-wizardlm. We added the following features to the original implementation:

• Modified it to be able to call Azure OpenAI by adding the AzureGPTPipeline class.
• The prompt has been refined and modified to support multiple languages. Use --language argument for other language. (e.g., --language Korean)
• Made it possible to create questions only when necessary. A better strategy is to create questions and answers separately. Use --question_only argument. (e.g., --questioin_only True)
• Prevented infinite loop. mutate() in the original implementation determines the validity of the augmented statement and repeats the loop until it is valid. However, this process takes a very long time and there is a problem in that the loop repeats infinitely in certain situations.

How to create dataset

Option 1. If you want to generate your own seed dataset through this lab (Please check ../seed)

Example datasets are placed in this folder. Please try the minimal example first and configure your dataset by referring to the tunable parameters.

Debug for test

chmod +x run_debug.sh
./run_debug.sh

Option 2. If you already have your own dataset

Example datasets are placed in this folder. Please try the minimal example first and configure your dataset by referring to the tunable parameters.

Debug for test

python evolve.py --seed_file xxx.jsonl --column_names Instruction --num_rows 50 --max_len_chars 512 --language English

Tunable parameters

parser.add_argument("--seed_file", type=str)
parser.add_argument("--column_names", nargs='+', default="Instruction")
parser.add_argument("--num_rows", type=int, default=5)
parser.add_argument("--min_len_chars", type=int, default=32)
parser.add_argument("--max_len_chars", type=int, default=512)
parser.add_argument("--temperature", type=int, default=0.7)
parser.add_argument("--top_p", type=int, default=0.95)
parser.add_argument("--model_name", type=str, default="gpt-4o")
parser.add_argument("--language", type=str, default="English")
parser.add_argument("--question_only", type=bool, default=True)

Run example: