initial persona data gen 2 commit #489
Conversation
|
@natolambert good idea! removed the big files and moved them to hugging face dataset at: |
| @@ -0,0 +1,48 @@ | |||
| To start make sure you have your OpenAI and Anthropic API keys and have installed the libraries listed in `requirements.txt`: | |||
There was a problem hiding this comment.
Maybe give it a title like
Persona Data Generation
This folder contains scripts on generating persona data used for preference tuning in the Tulu 3 recipe [citation, section xxxx)
|
|
||
| ``` | ||
| # Generate Instruction Following prompts | ||
| python persona_driven_generate_ifdata.py --model "gpt-4o" --start_index 0 --end_index <SAMPLE_SIZE> --output_path if_prompts.jsonl --openai_key Z --org_id YYY --dataset ai2-adapt-dev/personahub_personas --template instruction_following |
There was a problem hiding this comment.
can we make the command copy-paste and run? E.g., remove <SAMPLE_SIZE> and replace it with a sensible default value. --openai_key should be read from the environment possibly? What is this --org_id?
| python persona_driven_generate_ifdata.py --model "gpt-4o" --start_index 0 --end_index <SAMPLE_SIZE> --output_path if_solutions.jsonl --openai_key Z --org_id YYY --dataset if_prompts.jsonl --template instruction_following_solution | ||
|
|
||
| # Rewrite prompts to form Rejected Response (used for Presona-IF DPO data) | ||
| python persona_driven_generate_ifdata.py --model "gpt-4o" --start_index 0 --end_index <SAMPLE_SIZE> --output_path if_solutions.jsonl --openai_key Z --org_id YYY --dataset if_prompts.jsonl --template rewrite_if_prompt |
There was a problem hiding this comment.
Maybe include some example outputs to give the readers on what to expect after running the script. Kind of like how I included Here is an example created dataset: https://huggingface.co/datasets/vwxyzjn/rejection_sampling_1727887719 in https://github.com/allenai/open-instruct/blob/main/docs/algorithms/rejection_sampling.md#scoring-completions
This is the main scripts for persona-driven synthetic data generation for Math, Coding and Precise Instruction Following.