ilql_sentiments.py

import json
import os
import sys
from typing import Dict, List

from datasets import load_dataset
from transformers import pipeline

import trlx
from trlx.data.default_configs import TRLConfig, default_ilql_config


def get_positive_score(scores):
    "Extract value associated with a positive sentiment from pipeline's output"
    return dict(map(lambda x: tuple(x.values()), scores))["POSITIVE"]


def main(hparams={}):
    # Merge sweep config with default config if given
    config = TRLConfig.update(default_ilql_config().to_dict(), hparams)

    sentiment_fn = pipeline(
        "sentiment-analysis",
        "lvwerra/distilbert-imdb",
        top_k=2,
        truncation=True,
        batch_size=256,
        device=0 if int(os.environ.get("LOCAL_RANK", 0)) == 0 else -1,
    )

    def metric_fn(samples: List[str], **kwargs) -> Dict[str, List[float]]:
        sentiments = list(map(get_positive_score, sentiment_fn(samples)))
        return {"sentiments": sentiments}

    imdb = load_dataset("imdb", split="train+test")

    trlx.train(
        samples=imdb["text"],
        rewards=imdb["label"],
        eval_prompts=["I don't know much about Hungarian underground"] * 256,
        metric_fn=metric_fn,
        config=config,
    )


if __name__ == "__main__":
    hparams = {} if len(sys.argv) == 1 else json.loads(sys.argv[1])
    main(hparams)