[deps] goodbye python 3.9 (#9677)

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: hiyouga <16256802+hiyouga@users.noreply.github.com>
Co-authored-by: hiyouga <hiyouga@buaa.edu.cn>

src/llamafactory/data/converter.py

@@ -15,7 +15,7 @@ import json
 import os
 from abc import abstractmethod
 from dataclasses import dataclass
-from typing import TYPE_CHECKING, Any, Optional, Union
+from typing import TYPE_CHECKING, Any, Union
 
 from ..extras import logging
 from .data_utils import Role
@@ -40,7 +40,7 @@ class DatasetConverter:
     dataset_attr: "DatasetAttr"
     data_args: "DataArguments"
 
-    def _find_medias(self, medias: Union["MediaType", list["MediaType"], None]) -> Optional[list["MediaType"]]:
+    def _find_medias(self, medias: Union["MediaType", list["MediaType"], None]) -> list["MediaType"] | None:
         r"""Optionally concatenate media path to media dir when loading from local disk."""
         if medias is None:
             return None

src/llamafactory/data/formatter.py

@@ -16,7 +16,6 @@ import json
 import re
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
-from typing import Optional, Union
 
 from typing_extensions import override
 
@@ -27,14 +26,14 @@ from .tool_utils import FunctionCall, get_tool_utils
 @dataclass
 class Formatter(ABC):
     slots: SLOTS = field(default_factory=list)
-    tool_format: Optional[str] = None
+    tool_format: str | None = None
 
     @abstractmethod
     def apply(self, **kwargs) -> SLOTS:
         r"""Forms a list of slots according to the inputs to encode."""
         ...
 
-    def extract(self, content: str) -> Union[str, list["FunctionCall"]]:
+    def extract(self, content: str) -> str | list["FunctionCall"]:
         r"""Extract a list of tuples from the response message if using tools.
 
         Each tuple consists of function name and function arguments.
@@ -156,5 +155,5 @@ class ToolFormatter(Formatter):
             raise RuntimeError(f"Invalid JSON format in tool description: {str([content])}.")  # flat string
 
     @override
-    def extract(self, content: str) -> Union[str, list["FunctionCall"]]:
+    def extract(self, content: str) -> str | list["FunctionCall"]:
         return self.tool_utils.tool_extractor(content)

src/llamafactory/data/loader.py

@@ -162,13 +162,13 @@ def _load_single_dataset(
 
 
 def _get_merged_dataset(
-    dataset_names: Optional[list[str]],
+    dataset_names: list[str] | None,
     model_args: "ModelArguments",
     data_args: "DataArguments",
     training_args: "Seq2SeqTrainingArguments",
     stage: Literal["pt", "sft", "rm", "ppo", "kto"],
     return_dict: bool = False,
-) -> Optional[Union["Dataset", "IterableDataset", dict[str, "Dataset"]]]:
+) -> Union["Dataset", "IterableDataset", dict[str, "Dataset"]] | None:
     r"""Return the merged datasets in the standard format."""
     if dataset_names is None:
         return None
@@ -227,7 +227,7 @@ def _get_dataset_processor(
 
 
 def _get_preprocessed_dataset(
-    dataset: Optional[Union["Dataset", "IterableDataset"]],
+    dataset: Union["Dataset", "IterableDataset"] | None,
     data_args: "DataArguments",
     training_args: "Seq2SeqTrainingArguments",
     stage: Literal["pt", "sft", "rm", "ppo", "kto"],
@@ -235,7 +235,7 @@ def _get_preprocessed_dataset(
     tokenizer: "PreTrainedTokenizer",
     processor: Optional["ProcessorMixin"] = None,
     is_eval: bool = False,
-) -> Optional[Union["Dataset", "IterableDataset"]]:
+) -> Union["Dataset", "IterableDataset"] | None:
     r"""Preprocesses the dataset, including format checking and tokenization."""
     if dataset is None:
         return None

src/llamafactory/data/mm_plugin.py

@@ -22,7 +22,7 @@ import re
 from copy import deepcopy
 from dataclasses import dataclass
 from io import BytesIO
-from typing import TYPE_CHECKING, BinaryIO, Literal, Optional, TypedDict, Union
+from typing import TYPE_CHECKING, BinaryIO, Literal, NotRequired, Optional, TypedDict, Union
 
 import numpy as np
 import torch
@@ -32,7 +32,7 @@ from transformers.models.mllama.processing_mllama import (
     convert_sparse_cross_attention_mask_to_dense,
     get_cross_attention_token_mask,
 )
-from typing_extensions import NotRequired, override
+from typing_extensions import override
 
 from ..extras.constants import AUDIO_PLACEHOLDER, IGNORE_INDEX, IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
 from ..extras.packages import is_pillow_available, is_pyav_available, is_transformers_version_greater_than
@@ -63,8 +63,8 @@ if TYPE_CHECKING:
     from transformers.video_processing_utils import BaseVideoProcessor
 
     class EncodedImage(TypedDict):
-        path: Optional[str]
-        bytes: Optional[bytes]
+        path: str | None
+        bytes: bytes | None
 
     ImageInput = Union[str, bytes, EncodedImage, BinaryIO, ImageObject]
     VideoInput = Union[str, BinaryIO, list[list[ImageInput]]]
@@ -144,9 +144,9 @@ def _check_video_is_nested_images(video: "VideoInput") -> bool:
 
 @dataclass
 class MMPluginMixin:
-    image_token: Optional[str]
-    video_token: Optional[str]
-    audio_token: Optional[str]
+    image_token: str | None
+    video_token: str | None
+    audio_token: str | None
     expand_mm_tokens: bool = True
 
     def _validate_input(
@@ -328,7 +328,7 @@ class MMPluginMixin:
         videos: list["VideoInput"],
         audios: list["AudioInput"],
         processor: "MMProcessor",
-        imglens: Optional[list[int]] = None,
+        imglens: list[int] | None = None,
     ) -> dict[str, "torch.Tensor"]:
         r"""Process visual inputs.
 
@@ -426,13 +426,13 @@ class BasePlugin(MMPluginMixin):
     def process_token_ids(
         self,
         input_ids: list[int],
-        labels: Optional[list[int]],
+        labels: list[int] | None,
         images: list["ImageInput"],
         videos: list["VideoInput"],
         audios: list["AudioInput"],
         tokenizer: "PreTrainedTokenizer",
         processor: Optional["MMProcessor"],
-    ) -> tuple[list[int], Optional[list[int]]]:
+    ) -> tuple[list[int], list[int] | None]:
         r"""Pre-process token ids after tokenization for VLMs."""
         self._validate_input(processor, images, videos, audios)
         return input_ids, labels
@@ -1305,13 +1305,13 @@ class PaliGemmaPlugin(BasePlugin):
     def process_token_ids(
         self,
         input_ids: list[int],
-        labels: Optional[list[int]],
+        labels: list[int] | None,
         images: list["ImageInput"],
         videos: list["VideoInput"],
         audios: list["AudioInput"],
         tokenizer: "PreTrainedTokenizer",
         processor: Optional["MMProcessor"],
-    ) -> tuple[list[int], Optional[list[int]]]:
+    ) -> tuple[list[int], list[int] | None]:
         self._validate_input(processor, images, videos, audios)
         num_images = len(images)
         image_seqlen = processor.image_seq_length if self.expand_mm_tokens else 0  # skip mm token
@@ -2126,9 +2126,9 @@ def register_mm_plugin(name: str, plugin_class: type["BasePlugin"]) -> None:
 
 def get_mm_plugin(
     name: str,
-    image_token: Optional[str] = None,
-    video_token: Optional[str] = None,
-    audio_token: Optional[str] = None,
+    image_token: str | None = None,
+    video_token: str | None = None,
+    audio_token: str | None = None,
     **kwargs,
 ) -> "BasePlugin":
     r"""Get plugin for multimodal inputs."""

src/llamafactory/data/parser.py

@@ -15,7 +15,7 @@
 import json
 import os
 from dataclasses import dataclass
-from typing import Any, Literal, Optional, Union
+from typing import Any, Literal
 
 from huggingface_hub import hf_hub_download
 
@@ -33,40 +33,40 @@ class DatasetAttr:
     formatting: Literal["alpaca", "sharegpt", "openai"] = "alpaca"
     ranking: bool = False
     # extra configs
-    subset: Optional[str] = None
+    subset: str | None = None
     split: str = "train"
-    folder: Optional[str] = None
-    num_samples: Optional[int] = None
+    folder: str | None = None
+    num_samples: int | None = None
     # common columns
-    system: Optional[str] = None
-    tools: Optional[str] = None
-    images: Optional[str] = None
-    videos: Optional[str] = None
-    audios: Optional[str] = None
+    system: str | None = None
+    tools: str | None = None
+    images: str | None = None
+    videos: str | None = None
+    audios: str | None = None
     # dpo columns
-    chosen: Optional[str] = None
-    rejected: Optional[str] = None
-    kto_tag: Optional[str] = None
+    chosen: str | None = None
+    rejected: str | None = None
+    kto_tag: str | None = None
     # alpaca columns
-    prompt: Optional[str] = "instruction"
-    query: Optional[str] = "input"
-    response: Optional[str] = "output"
-    history: Optional[str] = None
+    prompt: str | None = "instruction"
+    query: str | None = "input"
+    response: str | None = "output"
+    history: str | None = None
     # sharegpt columns
-    messages: Optional[str] = "conversations"
+    messages: str | None = "conversations"
     # sharegpt tags
-    role_tag: Optional[str] = "from"
-    content_tag: Optional[str] = "value"
-    user_tag: Optional[str] = "human"
-    assistant_tag: Optional[str] = "gpt"
-    observation_tag: Optional[str] = "observation"
-    function_tag: Optional[str] = "function_call"
-    system_tag: Optional[str] = "system"
+    role_tag: str | None = "from"
+    content_tag: str | None = "value"
+    user_tag: str | None = "human"
+    assistant_tag: str | None = "gpt"
+    observation_tag: str | None = "observation"
+    function_tag: str | None = "function_call"
+    system_tag: str | None = "system"
 
     def __repr__(self) -> str:
         return self.dataset_name
 
-    def set_attr(self, key: str, obj: dict[str, Any], default: Optional[Any] = None) -> None:
+    def set_attr(self, key: str, obj: dict[str, Any], default: Any | None = None) -> None:
         setattr(self, key, obj.get(key, default))
 
     def join(self, attr: dict[str, Any]) -> None:
@@ -90,7 +90,7 @@ class DatasetAttr:
                 self.set_attr(tag, attr["tags"])
 
 
-def get_dataset_list(dataset_names: Optional[list[str]], dataset_dir: Union[str, dict]) -> list["DatasetAttr"]:
+def get_dataset_list(dataset_names: list[str] | None, dataset_dir: str | dict) -> list["DatasetAttr"]:
     r"""Get the attributes of the datasets."""
     if dataset_names is None:
         dataset_names = []