Data voorbereiden voor RLHF

Reinforcement Learning from Human Feedback (RLHF)

Mina Parham

AI Engineer

Preferentie- vs. prompt-datasets

Het complete RLHF-proces.

Reinforcement Learning from Human Feedback (RLHF)

Preferentie- vs. prompt-datasets

Het complete RLHF-proces met initiële prompt-dataset.

Reinforcement Learning from Human Feedback (RLHF)

Preferentie- vs. prompt-datasets

Het complete RLHF-proces met initiële prompt-dataset en een preferentie-dataset voor het beloningsmodel.

Reinforcement Learning from Human Feedback (RLHF)

Prompt-dataset

  • Vragen voor het model
  • Te vinden op Hugging Face Datasets
prompt_data = load_dataset("center-for-humans-and-machines/rlhf-hackathon-prompts", 
                           split="train")
prompt_data['prompt'][0]
'How important is climate change?'
  • Misschien moet je de prompt extraheren
  • Zoek naar markeringen zoals: Input=, {{Text}}:, ###Human:
Reinforcement Learning from Human Feedback (RLHF)

De preferentie-dataset verkennen

from datasets import load_dataset
preference_data = load_dataset("trl-internal-testing/hh-rlhf-helpful-base-trl-style", 
                               split="train")

Een screenshot van het Hugging Face-platform met de hh-rlhf-dataset.

Reinforcement Learning from Human Feedback (RLHF)

De preferentie-dataset verwerken

def extract_prompt(text):
    # Extract the prompt as the first element in the list
    prompt = text[0]["content"]
    return prompt
# Apply the extraction function to the dataset
preference_data_with_prompt = preference_data.map(
    lambda example: {**example, 'prompt': extract_prompt(example['chosen'])}
)
  • De extractie van prompts verschilt per dataset
Reinforcement Learning from Human Feedback (RLHF)

Definitieve preferentie-dataset

sample = preference_data_with_prompt.select(range(1))
sample['prompt']
'What vitamins are essential for the body to function?'
sample['chosen']
[ { "content": "What vitamins are essential for the body to function?", "role": 
   "user" }, { "content": "There are some very important vitamins that ensure the 
   proper functioning of the body, including Vitamins A, C, D, E, and K along ...}]
Reinforcement Learning from Human Feedback (RLHF)

Laten we oefenen!

Reinforcement Learning from Human Feedback (RLHF)

Preparing Video For Download...