[CI/Build] Update Ruff version (#8469)

Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py

-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, cast
 
 import torch
 from pydantic import BaseModel
@@ -79,8 +79,8 @@ class CompressedTensorsConfig(QuantizationConfig):
     @classmethod
     def from_config(cls, config: Dict[str, Any]) -> "CompressedTensorsConfig":
         target_scheme_map: Dict[str, Any] = dict()
-        ignore: List[str] = config.get("ignore", None)
-        quant_format: str = config.get("format", None)
+        ignore = cast(List[str], config.get("ignore"))
+        quant_format = cast(str, config.get("format"))
 
         # The quant_config has multiple config_groups, each containing
         # an input_activations key with details about how the activations are
@@ -200,7 +200,7 @@ class CompressedTensorsConfig(QuantizationConfig):
         is_per_tensor_or_channel_weight = (weight_quant.strategy in [
             QuantizationStrategy.TENSOR, QuantizationStrategy.CHANNEL
         ])
-        if not (is_symmetric_weight and is_static_weight
+        if not (is_symmetric_weight and is_static_weight  # noqa: SIM103
                 and is_per_tensor_or_channel_weight):
             return False
 
@@ -333,7 +333,7 @@ class CompressedTensorsLinearMethod(LinearMethodBase):
                        output_size: int, params_dtype: torch.dtype,
                        **extra_weight_attrs):
         """
-        Use the CompressedTensorsScheme associated with each layer to create 
+        Use the CompressedTensorsScheme associated with each layer to create
         the necessary parameters for the layer. See LinearMethodBase for param
         details
         """
@@ -352,8 +352,8 @@ class CompressedTensorsLinearMethod(LinearMethodBase):
               x: torch.Tensor,
               bias: Optional[torch.Tensor] = None):
         """
-        Use the output of create_weights and the CompressedTensorsScheme 
-        associated with the layer to apply the forward pass with the 
+        Use the output of create_weights and the CompressedTensorsScheme
+        associated with the layer to apply the forward pass with the
         layer input.  See LinearMethodBase for param details
 
         """

vllm/model_executor/layers/quantization/gptq_marlin.py

@@ -132,10 +132,10 @@ class GPTQMarlinConfig(QuantizationConfig):
     def is_gptq_marlin_compatible(cls, quant_config: Dict[str, Any]):
         # Extract data from quant config.
         quant_method = quant_config.get("quant_method", "").lower()
-        num_bits = quant_config.get("bits", None)
-        group_size = quant_config.get("group_size", None)
-        sym = quant_config.get("sym", None)
-        desc_act = quant_config.get("desc_act", None)
+        num_bits = quant_config.get("bits")
+        group_size = quant_config.get("group_size")
+        sym = quant_config.get("sym")
+        desc_act = quant_config.get("desc_act")
 
         if quant_method != "gptq":
             return False

vllm/model_executor/model_loader/tensorizer.py

@@ -408,9 +408,7 @@ def is_vllm_tensorized(tensorizer_config: "TensorizerConfig") -> bool:
             "inferred as vLLM models, so setting vllm_tensorized=True is "
             "only necessary for models serialized prior to this change.")
         return True
-    if (".vllm_tensorized_marker" in deserializer):
-        return True
-    return False
+    return ".vllm_tensorized_marker" in deserializer
 
 
 def serialize_vllm_model(

vllm/model_executor/models/minicpmv.py

@@ -884,7 +884,7 @@ class MiniCPMV(MiniCPMVBaseModel):
             version = str(config.version).split(".")
             version = tuple([int(x) for x in version])
         # Dispatch class based on version
-        instance_class = _SUPPORT_VERSION.get(version, None)
+        instance_class = _SUPPORT_VERSION.get(version)
         if instance_class is None:
             raise ValueError(
                 "Currently, MiniCPMV only supports versions 2.0, 2.5, and 2.6")

vllm/spec_decode/draft_model_runner.py

@@ -183,10 +183,7 @@ class TP1DraftModelRunner(ModelRunner):
             return False
 
         # TODO: Add soft-tuning prompt adapter support
-        if self.prompt_adapter_config:
-            return False
-
-        return True
+        return not self.prompt_adapter_config
 
     @torch.inference_mode()
     def execute_model(

vllm/spec_decode/metrics.py

@@ -104,13 +104,10 @@ class AsyncMetricsCollector:
         if self._rank != 0:
             return False
 
-        if (now - self._last_metrics_collect_time <
-                self._rejsample_metrics_collect_interval_s):
-            return False
-        return True
+        return now - self._last_metrics_collect_time >= self._rejsample_metrics_collect_interval_s  # noqa: E501
 
     def _copy_rejsample_metrics_async(self) -> torch.cuda.Event:
-        """Copy rejection/typical-acceptance sampling metrics 
+        """Copy rejection/typical-acceptance sampling metrics
         (number of accepted tokens, etc) to CPU asynchronously.
 
         Returns a CUDA event recording when the copy is complete.

vllm/triton_utils/libentry.py

@@ -35,8 +35,8 @@ class LibEntry(triton.KernelInterface):
         dns_key = [
             arg.dtype if hasattr(
                 arg, "data_ptr") else type(arg) if not isinstance(arg, int)
-            else "i32" if -(2**31) <= arg and arg <= 2**31 -
-            1 else "u64" if 2**63 <= arg and arg <= 2**64 - 1 else "i64"
+            else "i32" if arg >= -(2**31) and arg <= 2**31 -
+            1 else "u64" if arg >= 2**63 and arg <= 2**64 - 1 else "i64"
             for arg in dns_args
         ]
         # const args passed by position