src/llamafactory/data/processors/supervised.py

# Copyright 2024 the LlamaFactory team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from collections import defaultdict
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Tuple

from ...extras.constants import IGNORE_INDEX
from ...extras.logging import get_logger
from .processor_utils import get_paligemma_token_type_ids, get_pixel_values, greedy_knapsack


if TYPE_CHECKING:
    from transformers import PreTrainedTokenizer, ProcessorMixin

    from ...hparams import DataArguments
    from ..template import Template


logger = get_logger(__name__)


def _encode_supervised_example(
    prompt: Sequence[Dict[str, str]],
    response: Sequence[Dict[str, str]],
    system: Optional[str],
    tools: Optional[str],
    template: "Template",
    tokenizer: "PreTrainedTokenizer",
    processor: Optional["ProcessorMixin"],
    data_args: "DataArguments",
) -> Tuple[List[int], List[int]]:
    if processor is not None and not hasattr(processor, "image_seq_length"):  # llava-like models
        prompt[0]["content"] = template.image_token + prompt[0]["content"]

    messages = prompt + response
    input_ids, labels = [], []

    if processor is not None and hasattr(processor, "image_seq_length"):  # paligemma models
        image_token_id = tokenizer.convert_tokens_to_ids(template.image_token)
        input_ids += [image_token_id] * getattr(processor, "image_seq_length")
        labels += [IGNORE_INDEX] * getattr(processor, "image_seq_length")

    encoded_pairs = template.encode_multiturn(
        tokenizer, messages, system, tools, data_args.cutoff_len, data_args.reserved_label_len
    )
    for turn_idx, (source_ids, target_ids) in enumerate(encoded_pairs):
        if data_args.train_on_prompt:
            source_mask = source_ids
        elif turn_idx != 0 and template.efficient_eos:
            source_mask = [tokenizer.eos_token_id] + [IGNORE_INDEX] * (len(source_ids) - 1)
        else:
            source_mask = [IGNORE_INDEX] * len(source_ids)

        input_ids += source_ids + target_ids
        labels += source_mask + target_ids

    if template.efficient_eos:
        input_ids += [tokenizer.eos_token_id]
        labels += [tokenizer.eos_token_id]

    return input_ids, labels


def preprocess_supervised_dataset(
    examples: Dict[str, List[Any]],
    template: "Template",
    tokenizer: "PreTrainedTokenizer",
    processor: Optional["ProcessorMixin"],
    data_args: "DataArguments",
) -> Dict[str, List[List[int]]]:
    # build inputs with format `<bos> X Y <eos>` and labels with format `<ignore> ... <ignore> Y <eos>`
    # for multiturn examples, we only mask the prompt part in each prompt-response pair.
    model_inputs = {"input_ids": [], "attention_mask": [], "labels": []}
    if processor is not None:
        model_inputs["pixel_values"] = []
        if hasattr(processor, "image_seq_length"):  # paligemma models
            model_inputs["token_type_ids"] = []

    for i in range(len(examples["prompt"])):
        if len(examples["prompt"][i]) % 2 != 1 or len(examples["response"][i]) != 1:
            logger.warning("Dropped invalid example: {}".format(examples["prompt"][i] + examples["response"][i]))
            continue

        input_ids, labels = _encode_supervised_example(
            prompt=examples["prompt"][i],
            response=examples["response"][i],
            system=examples["system"][i],
            tools=examples["tools"][i],
            template=template,
            tokenizer=tokenizer,
            processor=processor,
            data_args=data_args,
        )
        model_inputs["input_ids"].append(input_ids)
        model_inputs["attention_mask"].append([1] * len(input_ids))
        model_inputs["labels"].append(labels)
        if processor is not None:
            model_inputs["pixel_values"].append(get_pixel_values(examples["images"][i], processor))
            if hasattr(processor, "image_seq_length"):  # paligemma models
                model_inputs["token_type_ids"].append(get_paligemma_token_type_ids(len(input_ids), processor))

    return model_inputs


def preprocess_packed_supervised_dataset(
    examples: Dict[str, List[Any]],
    template: "Template",
    tokenizer: "PreTrainedTokenizer",
    data_args: "DataArguments",
) -> Dict[str, List[List[int]]]:
    # build inputs with format `<bos> X1 Y1 <eos> <bos> X2 Y2 <eos>`
    # and labels with format `<ignore> ... <ignore> Y1 <eos> <ignore> ... <ignore> Y2 <eos>`
    valid_num = 0
    batch_input_ids, batch_labels = [], []
    lengths = []
    length2indexes = defaultdict(list)
    for i in range(len(examples["prompt"])):
        if len(examples["prompt"][i]) % 2 != 1 or len(examples["response"][i]) != 1:
            logger.warning("Dropped invalid example: {}".format(examples["prompt"][i] + examples["response"][i]))
            continue

        input_ids, labels = _encode_supervised_example(
            prompt=examples["prompt"][i],
            response=examples["response"][i],
            system=examples["system"][i],
            tools=examples["tools"][i],
            template=template,
            tokenizer=tokenizer,
            processor=None,
            data_args=data_args,
        )
        length = len(input_ids)
        if length > data_args.cutoff_len:
            logger.warning("Dropped lengthy example with length {} > {}.".format(length, data_args.cutoff_len))
        else:
            lengths.append(length)
            length2indexes[length].append(valid_num)
            batch_input_ids.append(input_ids)
            batch_labels.append(labels)
            valid_num += 1

    model_inputs = {"input_ids": [], "attention_mask": [], "labels": []}
    knapsacks = greedy_knapsack(lengths, data_args.cutoff_len)
    for knapsack in knapsacks:
        packed_input_ids, packed_labels = [], []
        for length in knapsack:
            index = length2indexes[length].pop()
            packed_input_ids += batch_input_ids[index]
            packed_labels += batch_labels[index]

        if len(packed_input_ids) < data_args.cutoff_len:
            pad_length = data_args.cutoff_len - len(packed_input_ids)
            packed_input_ids += [tokenizer.pad_token_id] * pad_length
            packed_labels += [IGNORE_INDEX] * pad_length

        if len(packed_input_ids) != data_args.cutoff_len:
            raise ValueError("The length of packed example should be identical to the cutoff length.")

        model_inputs["input_ids"].append(packed_input_ids)
        model_inputs["attention_mask"].append([1] * data_args.cutoff_len)
        model_inputs["labels"].append(packed_labels)

    return model_inputs


def print_supervised_dataset_example(example: Dict[str, List[int]], tokenizer: "PreTrainedTokenizer") -> None:
    valid_labels = list(filter(lambda x: x != IGNORE_INDEX, example["labels"]))
    print("input_ids:\n{}".format(example["input_ids"]))
    print("inputs:\n{}".format(tokenizer.decode(example["input_ids"], skip_special_tokens=False)))
    print("label_ids:\n{}".format(example["labels"]))
    print("labels:\n{}".format(tokenizer.decode(valid_labels, skip_special_tokens=False)))