Update pre-commit hooks (vllm-project#12475)

Signed-off-by: Harry Mellor <[email protected]>

Author: hmellor

.pre-commit-config.yaml

@@ -3,18 +3,18 @@ default_stages:
   - manual # Run in CI
 repos:
 - repo: https://github.com/google/yapf
-  rev: v0.32.0
+  rev: v0.43.0
   hooks:
   - id: yapf
     args: [--in-place, --verbose]
     additional_dependencies: [toml] # TODO: Remove when yapf is upgraded
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.6.5
+  rev: v0.9.3
   hooks:
   - id: ruff
     args: [--output-format, github]
 - repo: https://github.com/codespell-project/codespell
-  rev: v2.3.0
+  rev: v2.4.0
   hooks:
   - id: codespell
     exclude: 'benchmarks/sonnet.txt|(build|tests/(lora/data|models/fixtures|prompts))/.*'
@@ -23,7 +23,7 @@ repos:
   hooks:
   - id: isort
 - repo: https://github.com/pre-commit/mirrors-clang-format
-  rev: v18.1.5
+  rev: v19.1.7
   hooks:
   - id: clang-format
     exclude: 'csrc/(moe/topk_softmax_kernels.cu|quantization/gguf/(ggml-common.h|dequantize.cuh|vecdotq.cuh|mmq.cuh|mmvq.cuh))'
@@ -35,7 +35,7 @@ repos:
   - id: pymarkdown
     files: docs/.*
 - repo: https://github.com/rhysd/actionlint
-  rev: v1.7.6
+  rev: v1.7.7
   hooks:
   - id: actionlint
 - repo: local

benchmarks/benchmark_serving.py

@@ -926,8 +926,8 @@ def main(args: argparse.Namespace):
                     )
 
         # Traffic
-        result_json["request_rate"] = (
-            args.request_rate if args.request_rate < float("inf") else "inf")
+        result_json["request_rate"] = (args.request_rate if args.request_rate
+                                       < float("inf") else "inf")
         result_json["burstiness"] = args.burstiness
         result_json["max_concurrency"] = args.max_concurrency
 

csrc/custom_all_reduce.cuh

@@ -38,9 +38,13 @@ struct Signal {
   alignas(128) FlagType peer_counter[2][kMaxBlocks][8];
 };
 
-struct __align__(16) RankData { const void* __restrict__ ptrs[8]; };
+struct __align__(16) RankData {
+  const void* __restrict__ ptrs[8];
+};
 
-struct __align__(16) RankSignals { Signal* signals[8]; };
+struct __align__(16) RankSignals {
+  Signal* signals[8];
+};
 
 // like std::array, but aligned
 template <typename T, int sz>

csrc/moe/marlin_kernels/marlin_moe_kernel.h

@@ -138,8 +138,8 @@ __device__ inline FragB dequant<vllm::kU4B8.id()>(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
   // We want signed int4 outputs, hence we fuse the `-8` symmetric zero point
   // directly into `SUB` and `ADD`.
   const int SUB = 0x64086408;
@@ -182,8 +182,8 @@ __device__ inline FragB dequant<vllm::kU4.id()>(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
 
   const int SUB = 0x64006400;
   const int MUL = 0x2c002c00;

csrc/quantization/gptq_marlin/gptq_marlin.cu

@@ -173,8 +173,8 @@ dequant<half, vllm::kU4B8.id()>(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
   // We want signed int4 outputs, hence we fuse the `-8` symmetric zero point
   // directly into `SUB` and `ADD`.
   const int SUB = 0x64086408;
@@ -197,9 +197,9 @@ dequant<nv_bfloat16, vllm::kU4B8.id()>(int q) {
 
   // Guarantee that the `(a & b) | c` operations are LOP3s.
 
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, MASK, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, MASK, EX);
   q >>= 4;
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, MASK, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, MASK, EX);
 
   typename ScalarType<nv_bfloat16>::FragB frag_b;
   static constexpr uint32_t MUL = 0x3F803F80;
@@ -221,8 +221,8 @@ dequant<half, vllm::kU4.id()>(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
 
   const int SUB = 0x64006400;
   const int MUL = 0x2c002c00;
@@ -244,9 +244,9 @@ dequant<nv_bfloat16, vllm::kU4.id()>(int q) {
 
   // Guarantee that the `(a & b) | c` operations are LOP3s.
 
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, MASK, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, MASK, EX);
   q >>= 4;
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, MASK, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, MASK, EX);
 
   typename ScalarType<nv_bfloat16>::FragB frag_b;
   static constexpr uint32_t MUL = 0x3F803F80;

csrc/quantization/marlin/dense/marlin_cuda_kernel.cu

@@ -96,8 +96,8 @@ __device__ inline FragB dequant(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
   // We want signed int4 outputs, hence we fuse the `-8` symmetric zero point
   // directly into `SUB` and `ADD`.
   const int SUB = 0x64086408;

csrc/quantization/marlin/qqq/marlin_qqq_gemm_kernel.cu

@@ -141,8 +141,8 @@ __device__ inline FragB dequant_per_group(int q, FragS_GROUP& frag_s, int i) {
   static constexpr uint32_t HI = 0x00f000f0;
   static constexpr uint32_t EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  uint32_t t0 = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  uint32_t t1 = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  uint32_t t0 = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  uint32_t t1 = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
   // We want signed int4 outputs, hence we fuse the `-8` symmetric zero point
   // directly into `SUB` and `ADD`.
   static constexpr uint32_t SUB = 0x64086408;

csrc/quantization/marlin/sparse/common/mma.h

@@ -127,8 +127,8 @@ __device__ inline FragB dequant_4bit(int q) {
   const int HI = 0x00f000f0;
   const int EX = 0x64006400;
   // Guarantee that the `(a & b) | c` operations are LOP3s.
-  int lo = lop3<(0xf0 & 0xcc) | 0xaa>(q, LO, EX);
-  int hi = lop3<(0xf0 & 0xcc) | 0xaa>(q, HI, EX);
+  int lo = lop3 < (0xf0 & 0xcc) | 0xaa > (q, LO, EX);
+  int hi = lop3 < (0xf0 & 0xcc) | 0xaa > (q, HI, EX);
   // We want signed int4 outputs, hence we fuse the `-8` symmetric zero point
   // directly into `SUB` and `ADD`.
   const int SUB = 0x64086408;

csrc/rocm/attention.cu

@@ -907,7 +907,9 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_reduce_kernel(
     const scalar_t* __restrict__ tmp_out,  // [num_seqs, num_heads,
                                            // max_num_partitions, head_size]
     const int* __restrict__ context_lens,  // [num_seqs]
-    const int max_num_partitions){UNREACHABLE_CODE}
+    const int max_num_partitions) {
+  UNREACHABLE_CODE
+}
 
 #endif  // defined(__HIP__MI300_MI250__) TODO: Add NAVI support
 

setup.py

@@ -417,7 +417,7 @@ def get_rocm_version():
 
         if (get_rocm_core_version(ctypes.byref(major), ctypes.byref(minor),
                                   ctypes.byref(patch)) == 0):
-            return "%d.%d.%d" % (major.value, minor.value, patch.value)
+            return f"{major.value}.{minor.value}.{patch.value}"
         return None
     except Exception:
         return None