[CI/Build] drop support for Python 3.8 EOL (#8464)

Signed-off-by: Aaron Pham <contact@aarnphm.xyz>

vllm/model_executor/models/qwen2_vl.py

-# coding=utf-8
 # Adapted from
 # https://github.com/huggingface/transformers/blob/19e6e80e10118f855137b90740936c0b11ac397f/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py
 # Copyright 2024 The Qwen team.
@@ -246,9 +245,8 @@ class Qwen2VisionAttention(nn.Module):
         q, k, v = dist_utils.split_tensor_along_last_dim(x, 3)
         batch_size = q.shape[1]
 
-        q, k, v = [
-            rearrange(x, "s b ... -> b s ...").contiguous() for x in (q, k, v)
-        ]
+        q, k, v = (rearrange(x, "s b ... -> b s ...").contiguous()
+                   for x in (q, k, v))
         if rotary_pos_emb is not None:
             q = apply_rotary_pos_emb_vision(q, rotary_pos_emb)
             k = apply_rotary_pos_emb_vision(k, rotary_pos_emb)
@@ -258,7 +256,7 @@ class Qwen2VisionAttention(nn.Module):
             #   flash_attn_varlen_func)
             from flash_attn import flash_attn_varlen_func
 
-            q, k, v = [rearrange(x, "b s ... -> (b s) ...") for x in [q, k, v]]
+            q, k, v = (rearrange(x, "b s ... -> (b s) ...") for x in [q, k, v])
 
             max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max().item()
             output = flash_attn_varlen_func(q,
@@ -276,7 +274,7 @@ class Qwen2VisionAttention(nn.Module):
                                       b=batch_size)
         elif self.attn_backend == _Backend.TORCH_SDPA:
             seq_length = q.size(1)
-            q, k, v = [rearrange(x, "b s h d -> b h s d") for x in [q, k, v]]
+            q, k, v = (rearrange(x, "b s h d -> b h s d") for x in [q, k, v])
             attention_mask = torch.zeros([1, seq_length, seq_length],
                                          device=q.device,
                                          dtype=torch.bool)

vllm/model_executor/models/solar.py

-# coding=utf-8
 # Adapted from
 # https://github.com/huggingface/transformers/blob/v4.28.0/src/transformers/models/llama/modeling_llama.py
 # Copyright 2023 The vLLM team.

vllm/model_executor/models/stablelm.py

-# coding=utf-8
 # Copyright 2023 Stability AI, EleutherAI, and The HuggingFace Inc. team.
 # All rights reserved.
 #

vllm/model_executor/models/starcoder2.py

-# coding=utf-8
 # Copyright 2024 BigCode and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX

vllm/model_executor/models/xverse.py

-# coding=utf-8
 # Adapted from
 # https://huggingface.co/xverse/XVERSE-7B/blob/main/modeling_xverse.py
 # Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.

vllm/multimodal/base.py

-import sys
 from abc import ABC, abstractmethod
 from collections import UserDict, defaultdict
 from typing import (TYPE_CHECKING, Any, Callable, Dict, List, Mapping,
@@ -34,14 +33,9 @@ A dictionary containing nested tensors which have been batched via
 :meth:`MultiModalInputs.batch`.
 """
 
-if sys.version_info < (3, 9):
-    # UserDict cannot be subscripted
-    class _MultiModalInputsBase(UserDict):
-        pass
-else:
 
-    class _MultiModalInputsBase(UserDict[str, NestedTensors]):
-        pass
+class _MultiModalInputsBase(UserDict[str, NestedTensors]):
+    pass
 
 
 class MultiModalInputs(_MultiModalInputsBase):
@@ -262,18 +256,23 @@ class MultiModalPlugin(ABC):
                 logger.warning(
                     "Model class %s already has an input mapper "
                     "registered to %s. It is overwritten by the new one.",
-                    model_cls, self)
+                    model_cls,
+                    self,
+                )
 
-            self._input_mappers[model_cls] = mapper \
-                or self._default_input_mapper
+            self._input_mappers[model_cls] = (mapper
+                                              or self._default_input_mapper)
 
             return model_cls
 
         return wrapper
 
-    def map_input(self, model_config: "ModelConfig",
-                  data: MultiModalData[object],
-                  mm_processor_kwargs: Dict[str, Any]) -> MultiModalInputs:
+    def map_input(
+        self,
+        model_config: "ModelConfig",
+        data: MultiModalData[object],
+        mm_processor_kwargs: Dict[str, Any],
+    ) -> MultiModalInputs:
         """
         Transform the data into a dictionary of model inputs using the
         input mapper registered for that model.
@@ -348,13 +347,15 @@ class MultiModalPlugin(ABC):
                 logger.warning(
                     "Model class %s already calculates maximum number of "
                     "tokens in %s. It is overwritten by the new one.",
-                    model_cls, self)
+                    model_cls,
+                    self,
+                )
 
             if isinstance(max_mm_tokens, int):
                 self._validate_max_multimodal_tokens(max_mm_tokens)
 
-            self._max_mm_tokens[model_cls] = max_mm_tokens \
-                or self._default_max_multimodal_tokens
+            self._max_mm_tokens[model_cls] = (
+                max_mm_tokens or self._default_max_multimodal_tokens)
 
             return model_cls
 
@@ -482,8 +483,10 @@ class MultiModalPlaceholderMap:
         placeholder_maps: Dict[str, MultiModalPlaceholderMap] = defaultdict(
             MultiModalPlaceholderMap)
 
-        for modality, placeholders in seq_group.multi_modal_placeholders.items(
-        ):
+        for (
+                modality,
+                placeholders,
+        ) in seq_group.multi_modal_placeholders.items():
             mm_items = mm_data.pop(modality)
             if not isinstance(mm_items, list):
                 mm_items = [mm_items]
@@ -499,8 +502,11 @@ class MultiModalPlaceholderMap:
         return mm_data, placeholder_maps
 
     def append_items_from_seq_group(
-            self, positions: range, multi_modal_items: List[_T],
-            multi_modal_placeholders: List[PlaceholderRange]) -> List[_T]:
+        self,
+        positions: range,
+        multi_modal_items: List[_T],
+        multi_modal_placeholders: List[PlaceholderRange],
+    ) -> List[_T]:
         """
         Adds the multi-modal items that intersect ```positions`` to this
         placeholder map and returns the intersecting items.
@@ -515,20 +521,26 @@ class MultiModalPlaceholderMap:
                                              multi_modal_items):
             placeholder = range(
                 placeholder_dict["offset"],
-                placeholder_dict["offset"] + placeholder_dict["length"])
-            intersection = range(max(positions.start, placeholder.start),
-                                 min(positions.stop, placeholder.stop))
+                placeholder_dict["offset"] + placeholder_dict["length"],
+            )
+            intersection = range(
+                max(positions.start, placeholder.start),
+                min(positions.stop, placeholder.stop),
+            )
 
             if not intersection:
                 # Skip this multi-modal item.
                 continue
 
-            token_embedding_range = range(intersection.start - positions.start,
-                                          intersection.stop - positions.start)
+            token_embedding_range = range(
+                intersection.start - positions.start,
+                intersection.stop - positions.start,
+            )
 
             multimodal_embedding_range = range(
                 intersection.start - placeholder.start + self.src_len,
-                intersection.stop - placeholder.start + self.src_len)
+                intersection.stop - placeholder.start + self.src_len,
+            )
 
             intersecting_items.append(mm_item)
             self.dest_ranges.append(token_embedding_range)

vllm/prompt_adapter/utils.py

@@ -37,9 +37,8 @@ def load_peft_weights(model_id: str,
             Additional arguments to pass to the `hf_hub_download` method when 
             loading from the HuggingFace Hub.
     """
-    path = (os.path.join(model_id, hf_hub_download_kwargs["subfolder"])
-            if hf_hub_download_kwargs.get("subfolder", None) is not None else
-            model_id)
+    path = (os.path.join(model_id, hf_hub_download_kwargs["subfolder"]) if
+            hf_hub_download_kwargs.get("subfolder") is not None else model_id)
 
     if device is None:
         device = infer_device()
@@ -51,19 +50,19 @@ def load_peft_weights(model_id: str,
         filename = os.path.join(path, WEIGHTS_NAME)
         use_safetensors = False
     else:
-        token = hf_hub_download_kwargs.get("token", None)
+        token = hf_hub_download_kwargs.get("token")
         if token is None:
-            token = hf_hub_download_kwargs.get("use_auth_token", None)
+            token = hf_hub_download_kwargs.get("use_auth_token")
 
         hub_filename = (os.path.join(hf_hub_download_kwargs["subfolder"],
                                      SAFETENSORS_WEIGHTS_NAME)
-                        if hf_hub_download_kwargs.get("subfolder", None)
-                        is not None else SAFETENSORS_WEIGHTS_NAME)
+                        if hf_hub_download_kwargs.get("subfolder") is not None
+                        else SAFETENSORS_WEIGHTS_NAME)
         has_remote_safetensors_file = file_exists(
             repo_id=model_id,
             filename=hub_filename,
-            revision=hf_hub_download_kwargs.get("revision", None),
-            repo_type=hf_hub_download_kwargs.get("repo_type", None),
+            revision=hf_hub_download_kwargs.get("revision"),
+            repo_type=hf_hub_download_kwargs.get("repo_type"),
             token=token,
         )
         use_safetensors = has_remote_safetensors_file

vllm/transformers_utils/config.py

@@ -308,7 +308,7 @@ def load_params_config(model, revision) -> PretrainedConfig:
         config_path = Path(
             hf_hub_download(model, config_file_name, revision=revision))
 
-    with open(config_path, "r") as file:
+    with open(config_path) as file:
         config_dict = json.load(file)
 
     config_mapping = {

vllm/transformers_utils/configs/chatglm.py

-# coding=utf-8
 # Adapted from
 # https://github.com/THUDM/ChatGLM2-6B
 from transformers import PretrainedConfig

vllm/transformers_utils/configs/exaone.py

-# coding=utf-8
 # Copied from
 # https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/blob/main/configuration_exaone.py
 # Copyright 2021 The LG AI Research EXAONE Lab. All rights reserved.

vllm/transformers_utils/configs/jais.py

-# coding=utf-8
 # Copyright 2023 The OpenAI Team Authors and HuggingFace Inc. team.
 # Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
 # Copyright 2023 Cerebras Systems.

vllm/transformers_utils/configs/mpt.py

-# coding=utf-8
 # Copied from
 # https://huggingface.co/mosaicml/mpt-7b/blob/main/configuration_mpt.py
 """A HuggingFace-style model configuration."""
@@ -117,10 +116,10 @@ class MPTConfig(PretrainedConfig):
                                                      init_config_defaults)
         if self.d_model % self.n_heads != 0:
             raise ValueError('d_model must be divisible by n_heads')
-        if any((
+        if any(
                 prob < 0 or prob > 1 for prob in
-            [self.attn_config['attn_pdrop'], self.resid_pdrop, self.emb_pdrop]
-        )):
+            [self.attn_config['attn_pdrop'], self.resid_pdrop, self.emb_pdrop
+             ]):
             raise ValueError(
                 "self.attn_config['attn_pdrop'], resid_pdrop, emb_pdrop are "
                 "probabilities and must be between 0 and 1")

vllm/transformers_utils/configs/nemotron.py

-# coding=utf-8
 # Copyright 2024 HuggingFace Inc. team. All rights reserved.
 # Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.
 #
@@ -144,7 +143,7 @@ class NemotronConfig(PretrainedConfig):
         self.intermediate_size = intermediate_size
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
-        head_dim = head_dim or kwargs.get("kv_channels", None)
+        head_dim = head_dim or kwargs.get("kv_channels")
         self.head_dim = head_dim if head_dim is not None else (
             hidden_size // num_attention_heads)
 
@@ -160,8 +159,8 @@ class NemotronConfig(PretrainedConfig):
         self.rope_theta = rope_theta
         self.rope_scaling = rope_scaling
         # for backward compatibility
-        partial_rotary_factor = kwargs.get("rope_percent", None) or kwargs.get(
-            "rope_percentage", None) or partial_rotary_factor
+        partial_rotary_factor = kwargs.get("rope_percent") or kwargs.get(
+            "rope_percentage") or partial_rotary_factor
         self.partial_rotary_factor = partial_rotary_factor
         self._rope_scaling_validation()
         self.attention_bias = attention_bias

vllm/transformers_utils/configs/solar.py

-# coding=utf-8
 # Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX

vllm/utils.py

@@ -1153,7 +1153,7 @@ class SortedHelpFormatter(argparse.HelpFormatter):
 
     def add_arguments(self, actions):
         actions = sorted(actions, key=lambda x: x.option_strings)
-        super(SortedHelpFormatter, self).add_arguments(actions)
+        super().add_arguments(actions)
 
 
 class FlexibleArgumentParser(argparse.ArgumentParser):
@@ -1279,7 +1279,7 @@ class FlexibleArgumentParser(argparse.ArgumentParser):
 
         config: Dict[str, Union[int, str]] = {}
         try:
-            with open(file_path, 'r') as config_file:
+            with open(file_path) as config_file:
                 config = yaml.safe_load(config_file)
         except Exception as ex:
             logger.error(