suggestion: import的时候，改为相对引用

[yuanzhoulvpi2017]

从推特过来的，你们这个项目，觉得非常棒🎉，期待更多的代码更新。
给个建议：megatron-lm最让我讨厌的一点，就是导入的时候，是绝对引用。
你们这个包，还在发展中，我看在import的时候，是绝对引用，希望改为相对引用。这样方便更多的开发者在这个上面进行二次开发。

当然这个也是不痛不痒的一个点。

[Sleepyhead01]

Looks like a good first issue. Could you clarify if there are any specific guidelines for switching from absolute to relative imports? Else, I can explore and open a PR with the changes. Tx

[yuanzhoulvpi2017]

may be get some idea from transformers code

# copy code from https://github.com/huggingface/transformers/blob/a5bb52847139bf6ad7489ac62a5fb6d0fa3d2ec6/src/transformers/models/paligemma/modeling_paligemma.py#L24
from ...cache_utils import Cache, HybridCache, StaticCache
from ...generation import GenerationMixin
from ...modeling_utils import PreTrainedModel
from ...utils import (
    ModelOutput,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    is_flash_attn_2_available,
    logging,
    replace_return_docstrings,
)
from .configuration_paligemma import PaliGemmaConfig


if is_flash_attn_2_available():
    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa

from ..auto import AutoModel, AutoModelForCausalLM

[Sleepyhead01]

Hi, I’ve submitted a PR after replacing absolute imports with relative imports across the codebase. Looking forward to your feedback! Tx