process_data.py

from datasets import load_dataset

dataset_name = "2nji/makebelieve-480"

# Load the dataset
dataset = load_dataset(dataset_name)

# Shuffle the dataset and slice it
dataset = dataset['train'].shuffle(seed=42).select(range(480))


# Define a function to transform the data
def transform_conversation(example):
    conversation_text = example['text']
    segments = conversation_text.split('###')

    reformatted_segments = []

    # Iterate over pairs of segments
    for i in range(1, len(segments) - 1, 2):
        human_text = segments[i].strip().replace('Human:', '').strip()

        # Check if there is a corresponding assistant segment before processing
        if i + 1 < len(segments):
            assistant_text = segments[i+1].strip().replace('Assistant:', '').strip()

            # Apply the new template
            reformatted_segments.append(f'<s>[INST] {human_text} [/INST] {assistant_text} </s>')
        else:
            # Handle the case where there is no corresponding assistant segment
            reformatted_segments.append(f'<s>[INST] {human_text} [/INST] </s>')

    return {'text': ''.join(reformatted_segments)}


# Apply the transformation
transformed_dataset = dataset.map(transform_conversation)

# Upload the dataset to the Hub
# Don't forget to replace the token with your own
transformed_dataset.push_to_hub(dataset_name, token="...")