Hi Xinran,
Thanks for sharing this repo — it's been really helpful!
While reading through alignMAP/training/ppo.py, I noticed this line:
from alignmap.data.loaders import get_dataset_prompts
However, there is no loaders.py under alignmap/data/, and the function get_dataset_prompts appears to be missing entirely.
I’m guessing this function was supposed to load and preprocess the dataset prompts. I’m currently using the Anthropic/hh-rlhf dataset, where the prompts are embedded in the full conversation string like:
"Human: Assistant: "
So we just need to extract the prompts, right?
Best
Hi Xinran,
Thanks for sharing this repo — it's been really helpful!
While reading through alignMAP/training/ppo.py, I noticed this line:
from alignmap.data.loaders import get_dataset_prompts
However, there is no loaders.py under alignmap/data/, and the function get_dataset_prompts appears to be missing entirely.
I’m guessing this function was supposed to load and preprocess the dataset prompts. I’m currently using the Anthropic/hh-rlhf dataset, where the prompts are embedded in the full conversation string like:
"Human: Assistant: "
So we just need to extract the prompts, right?
Best