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

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

.github/workflows/ruff.yml

@@ -25,10 +25,10 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install ruff==0.1.5 codespell==2.3.0 tomli==2.0.1 isort==5.13.2
+        pip install -r requirements-lint.txt
     - name: Analysing the code with ruff
       run: |
-        ruff .
+        ruff check .
     - name: Spelling check with codespell
       run: |
         codespell --toml pyproject.toml

benchmarks/kernels/graph_machete_bench.py

@@ -45,8 +45,7 @@ if __name__ == "__main__":
     rows = int(math.ceil(len(results) / 2))
     fig, axs = plt.subplots(rows, 2, figsize=(12, 5 * rows))
     axs = axs.flatten()
-    axs_idx = 0
-    for shape, data in results.items():
+    for axs_idx, (shape, data) in enumerate(results.items()):
         plt.sca(axs[axs_idx])
         df = pd.DataFrame(data)
         sns.lineplot(data=df,
@@ -59,6 +58,5 @@ if __name__ == "__main__":
                      palette="Dark2")
         plt.title(f"Shape: {shape}")
         plt.ylabel("time (median, s)")
-        axs_idx += 1
     plt.tight_layout()
     plt.savefig("graph_machete_bench.pdf")

format.sh

@@ -159,7 +159,7 @@ echo 'vLLM codespell: Done'
 
 # Lint specified files
 lint() {
-    ruff "$@"
+    ruff check "$@"
 }
 
 # Lint files that differ from main branch. Ignores dirs that are not slated
@@ -175,7 +175,7 @@ lint_changed() {
 
     if ! git diff --diff-filter=ACM --quiet --exit-code "$MERGEBASE" -- '*.py' '*.pyi' &>/dev/null; then
         git diff --name-only --diff-filter=ACM "$MERGEBASE" -- '*.py' '*.pyi' | xargs \
-             ruff
+             ruff check
     fi
 
 }

pyproject.toml

@@ -42,6 +42,8 @@ ignore = [
     "E731",
     # Loop control variable not used within loop body
     "B007",
+    # f-string format
+    "UP032",
 ]
 
 [tool.mypy]

requirements-lint.txt

@@ -2,7 +2,7 @@
 yapf==0.32.0
 toml==0.10.2
 tomli==2.0.1
-ruff==0.1.5
+ruff==0.6.5
 codespell==2.3.0
 isort==5.13.2
 clang-format==18.1.5

tests/conftest.py

@@ -158,10 +158,7 @@ def should_do_global_cleanup_after_test(request) -> bool:
     to initialize torch.
     """
 
-    if request.node.get_closest_marker("skip_global_cleanup"):
-        return False
-
-    return True
+    return not request.node.get_closest_marker("skip_global_cleanup")
 
 
 @pytest.fixture(autouse=True)

tests/lora/conftest.py

@@ -65,10 +65,7 @@ def should_do_global_cleanup_after_test(request) -> bool:
     to initialize torch.
     """
 
-    if request.node.get_closest_marker("skip_global_cleanup"):
-        return False
-
-    return True
+    return not request.node.get_closest_marker("skip_global_cleanup")
 
 
 @pytest.fixture(autouse=True)

tests/multimodal/test_base.py

@@ -5,7 +5,7 @@ from vllm.multimodal.base import MultiModalInputs, NestedTensors
 
 def assert_nested_tensors_equal(expected: NestedTensors,
                                 actual: NestedTensors):
-    assert type(expected) == type(actual)
+    assert type(expected) == type(actual)  # noqa: E721
     if isinstance(expected, torch.Tensor):
         assert torch.equal(expected, actual)
     else:

tests/test_cache_block_hashing.py

@@ -66,8 +66,7 @@ def test_auto_prefix_caching(model: str, block_size: int, max_num_seqs: int,
 
             hashes.append([])
             prompts = [prefix + prompt for prompt in sample_prompts]
-            seq_id = 0
-            for prompt in prompts:
+            for seq_id, prompt in enumerate(prompts):
                 hashes[-1].append([])
                 prompt_token_ids = tokenizer.encode(prompt)
                 seq = Sequence(seq_id,
@@ -83,8 +82,6 @@ def test_auto_prefix_caching(model: str, block_size: int, max_num_seqs: int,
                 for idx in range(num_blocks):
                     hashes[-1][-1].append(seq.hash_of_block(idx))
 
-                seq_id += 1
-
     # Check that hashes made with two prefixes with different first blocks are
     # different everywhere.
     for hash0, hash1 in zip(flatten_2d(hashes[0]), flatten_2d(hashes[1])):

tests/test_logger.py

@@ -111,7 +111,7 @@ def test_an_error_is_raised_when_custom_logging_config_file_does_not_exist():
     configuration occurs."""
     with pytest.raises(RuntimeError) as ex_info:
         _configure_vllm_root_logger()
-    assert ex_info.type == RuntimeError
+    assert ex_info.type == RuntimeError  # noqa: E721
     assert "File does not exist" in str(ex_info)
 
 
@@ -152,7 +152,7 @@ def test_an_error_is_raised_when_custom_logging_config_is_unexpected_json(
                    logging_config_file.name):
             with pytest.raises(ValueError) as ex_info:
                 _configure_vllm_root_logger()
-            assert ex_info.type == ValueError
+            assert ex_info.type == ValueError  # noqa: E721
             assert "Invalid logging config. Expected Dict, got" in str(ex_info)
 
 

tests/worker/test_encoder_decoder_model_runner.py

@@ -453,8 +453,7 @@ def test_prepare_decode(batch_size):
     # each sequence) in the decode phase
 
     expected_selected_token_indices = []
-    selected_token_start_idx = 0
-    for seq_len in seq_lens:
+    for selected_token_start_idx, seq_len in enumerate(seq_lens):
         # Compute the index offset of the final token in each
         # sequence's decoded outputs; since a single token is
         # decoded per iteration per sequence, then the length
@@ -463,7 +462,6 @@ def test_prepare_decode(batch_size):
         # generated tokens is 0 (i.e. the expected sampling index
         # for a given sequence is just `selected_token_start_idx`)
         expected_selected_token_indices.append(selected_token_start_idx)
-        selected_token_start_idx += 1
 
     sampling_metadata = model_input.sampling_metadata
     actual = sampling_metadata.selected_token_indices

tests/worker/test_model_runner.py

@@ -241,10 +241,8 @@ def test_prepare_decode_cuda_graph(batch_size):
 
     # Verify Sampling
     expected_selected_token_indices = []
-    selected_token_start_idx = 0
-    for _ in context_lens:
+    for selected_token_start_idx, _ in enumerate(context_lens):
         expected_selected_token_indices.append(selected_token_start_idx)
-        selected_token_start_idx += 1
     sampling_metadata = SamplingMetadata.prepare(
         seq_group_metadata_list,
         seq_lens,

vllm/adapter_commons/utils.py

@@ -42,7 +42,7 @@ def list_adapters(registered_adapters: Dict[int, Any]) -> Dict[int, Any]:
 
 def get_adapter(adapter_id: int,
                 registered_adapters: Dict[int, Any]) -> Optional[Any]:
-    return registered_adapters.get(adapter_id, None)
+    return registered_adapters.get(adapter_id)
 
 
 ## worker functions

vllm/attention/backends/utils.py

@@ -33,10 +33,8 @@ def is_block_tables_empty(block_tables: Union[None, Dict]):
     """
     if block_tables is None:
         return True
-    if isinstance(block_tables, dict) and all(
-            value is None for value in block_tables.values()):
-        return True
-    return False
+    return (isinstance(block_tables, dict)
+            and all(value is None for value in block_tables.values()))
 
 
 def compute_slot_mapping_start_idx(is_prompt: bool, query_len: int,

vllm/core/block/prefix_caching_block.py

@@ -417,9 +417,7 @@ class PrefixCachingBlockAllocator(BlockAllocator):
 
     def is_block_cached(self, block: Block) -> bool:
         assert block.content_hash is not None
-        if block.content_hash in self._cached_blocks:
-            return True
-        return False
+        return block.content_hash in self._cached_blocks
 
     def promote_to_immutable_block(self, block: Block) -> BlockId:
         """Once a mutable block is full, it can be promoted to an immutable

vllm/core/block_manager_v2.py

@@ -399,9 +399,7 @@ class BlockSpaceManagerV2(BlockSpaceManager):
         """
         alloc_status = self._can_swap(seq_group, Device.CPU,
                                       SequenceStatus.RUNNING)
-        if alloc_status == AllocStatus.OK:
-            return True
-        return False
+        return alloc_status == AllocStatus.OK
 
     def swap_out(self, seq_group: SequenceGroup) -> List[Tuple[int, int]]:
         """Returns the block id mapping (from GPU to CPU) generated by

vllm/engine/async_llm_engine.py

@@ -826,7 +826,7 @@ class AsyncLLMEngine:
             request_id: The unique id of the request.
             lora_request: LoRA request to use for generation, if any.
             trace_headers: OpenTelemetry trace headers.
-            prompt_adapter_request: Prompt Adapter request to use 
+            prompt_adapter_request: Prompt Adapter request to use
                                             for generation, if any.
 
         Yields:
@@ -1042,7 +1042,7 @@ class AsyncLLMEngine:
     async def start_profile(self) -> None:
         # using type instead of isinstance to check to avoid capturing
         # inherited classes
-        if type(self.engine.model_executor) == GPUExecutorAsync:
+        if type(self.engine.model_executor) == GPUExecutorAsync:  # noqa: E721
             self.engine.model_executor.start_profile()
         else:
             self.engine.model_executor._run_workers("start_profile")
@@ -1050,7 +1050,7 @@ class AsyncLLMEngine:
     async def stop_profile(self) -> None:
         # using type instead of isinstance to check to avoid capturing
         # inherited classes
-        if type(self.engine.model_executor) == GPUExecutorAsync:
+        if type(self.engine.model_executor) == GPUExecutorAsync:  # noqa: E721
             self.engine.model_executor.stop_profile()
         else:
             self.engine.model_executor._run_workers("stop_profile")

vllm/engine/llm_engine.py

@@ -144,7 +144,7 @@ class LLMEngine:
             decoding.
         executor_class: The model executor class for managing distributed
             execution.
-        prompt_adapter_config (Optional): The configuration related to serving 
+        prompt_adapter_config (Optional): The configuration related to serving
             prompt adapters.
         log_stats: Whether to log statistics.
         usage_context: Specified entry point, used for usage info collection.
@@ -1605,7 +1605,7 @@ class LLMEngine:
     def start_profile(self) -> None:
         # using type instead of isinstance to check to avoid capturing
         # inherited classes (MultiprocessingGPUExecutor)
-        if type(self.model_executor) == GPUExecutor:
+        if type(self.model_executor) == GPUExecutor:  # noqa: E721
             self.model_executor.start_profile()
         else:
             self.model_executor._run_workers("start_profile")
@@ -1613,7 +1613,7 @@ class LLMEngine:
     def stop_profile(self) -> None:
         # using type instead of isinstance to check to avoid capturing
         # inherited classes (MultiprocessingGPUExecutor)
-        if type(self.model_executor) == GPUExecutor:
+        if type(self.model_executor) == GPUExecutor:  # noqa: E721
             self.model_executor.stop_profile()
         else:
             self.model_executor._run_workers("stop_profile")

vllm/model_executor/guided_decoding/outlines_logits_processors.py

@@ -67,9 +67,9 @@ class BaseLogitsProcessor:
         instruction = self._guide.get_next_instruction(
             state=self._fsm_state[seq_id])
 
-        if type(instruction) == Generate:
+        if type(instruction) == Generate:  # noqa: E721
             allowed_tokens = instruction.tokens
-        elif type(instruction) == Write:
+        elif type(instruction) == Write:  # noqa: E721
             # TODO: support fast forward tokens
             allowed_tokens = [instruction.tokens[0]]
         else:

vllm/model_executor/layers/quantization/awq_marlin.py

@@ -110,9 +110,9 @@ class AWQMarlinConfig(QuantizationConfig):
     def is_awq_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)
-        has_zp = quant_config.get("zero_point", None)
+        num_bits = quant_config.get("bits")
+        group_size = quant_config.get("group_size")
+        has_zp = quant_config.get("zero_point")
 
         if quant_method != "awq":
             return False