evaluation/py/eval_utils_zsre.py

"""
Contains evaluation utilities for pytorch-based rewriting methods.
To use, simply call `compute_rewrite_quality_zsre` with the
appropriate arguments, which returns a dictionary containing them.


Script from memit ROME implementation

MIT License

Copyright (c) 2022 Kevin Meng

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
"""

import typing
from itertools import chain

import numpy as np
import torch
from sklearn.feature_extraction.text import TfidfVectorizer
from transformers import AutoModelForCausalLM, AutoTokenizer


def compute_rewrite_quality_zsre(
    model: AutoModelForCausalLM,
    tok: AutoTokenizer,
    record: typing.Dict,
    vec: TfidfVectorizer,
) -> typing.Dict:
    """
    Given a rewritten model, computes generalization and specificity metrics for
    the desired rewrite (passed in via the CounterFact dataset record). Returns a
    dictionary containing those metrics.

    :param model: Rewritten model
    :param tok: Tokenizer
    :param record: CounterFact dataset record
    :param vec: ???
    :return: Dictionary containing rewriting metrics
    """

    # First, unpack rewrite evaluation record.
    subject, target_new, target_true = (
        record["requested_rewrite"][x] for x in ["subject", "target_new", "target_true"]
    )
    rewrite_prompts = [record["requested_rewrite"]["prompt"].format(subject)]
    paraphrase_prompts = record["paraphrase_prompts"]
    neighborhood_prompts = record["neighborhood_prompts"]

    # Form a list of lists of prefixes to test.
    prob_prompts = [
        rewrite_prompts,
        paraphrase_prompts,
    ]
    # Flatten all the evaluated prefixes into one list.
    target_tok = tok(" " + target_new["str"], add_special_tokens=False)["input_ids"]
    inp_prompts_og = list(chain(*prob_prompts))
    inp_prompts = [
        el + tok.decode(target_tok[:i])
        for el in inp_prompts_og
        for i in range(len(target_tok))
    ]
    inp_targets = [
        tok.decode(target_tok[i])
        for _ in range(len(inp_prompts_og))
        for i in range(len(target_tok))
    ]

    stuff_probs = test_batch_prediction_acc(model, tok, inp_prompts, inp_targets)

    # Predict for neighborhood prompts (dictionary format).
    neighborhood_correct = test_batch_prediction_acc(
        model,
        tok,
        [
            el["prompt"].format(record["requested_rewrite"])
            for el in neighborhood_prompts
        ],
        [el["target"] for el in neighborhood_prompts],
    )

    probs = stuff_probs + neighborhood_correct

    # Unflatten the results again into a list of lists.
    cutoffs = [0] + np.cumsum(
        [l * len(target_tok) for l in map(len, prob_prompts)]
    ).tolist()
    ret_probs = [probs[cutoffs[i - 1] : cutoffs[i]] for i in range(1, len(cutoffs))]
    # Structure the restuls as a dictionary.
    ret = {
        f"{key}_correct": ret_probs[i]
        for i, key in enumerate(
            [
                "rewrite_prompts",
                "paraphrase_prompts",
            ]
        )
    }
    ret["neighborhood_prompts_correct"] = neighborhood_correct

    return ret


def test_batch_prediction_acc(model, tok, prompts: typing.List[str], target):
    prompt_tok = tok(
        prompts,
        padding=True,
        return_tensors="pt",
    ).to("cuda")

    with torch.no_grad():
        logits = model(**prompt_tok).logits
        last_non_masked = prompt_tok["attention_mask"].sum(1) - 1

        # account for weird tokenizers (like that of gemma) which pads in front instead of back!
        if tok.name_or_path.startswith('google/gemma'):
            last_non_masked = torch.from_numpy(np.array([prompt_tok['attention_mask'].shape[1]-1]*last_non_masked.shape[0], dtype=int)).cuda()

        to_gather = last_non_masked.unsqueeze(1).repeat(1, logits.size(-1)).unsqueeze(1)
        gathered = torch.gather(logits, 1, to_gather).squeeze(1)
        ans = torch.argmax(gathered, dim=1)

        correct_id = tok(target, padding=True, return_tensors="pt", add_special_tokens=False).to("cuda")[
            "input_ids"
        ]
        # Temporary hack to deal with foreign characters.
        correct_id = correct_id[:, 0].squeeze()

        return (ans == correct_id).detach().cpu().numpy().tolist()