Add pydocstyle to the ruff rules (#52136)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Harry Mellor <hej.mellor@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-16 16:44:29 +00:00
committed by GitHub
co-authored by Claude Opus 5
parent 804e5377aa
commit 8c1557a79c
2002 changed files with 15024 additions and 14269 deletions
+5 -10
View File
@@ -1,7 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""
Unit-test DeepGEMM FP8 and FP4 kernels (no DeepEP).
"""Unit-test DeepGEMM FP8 and FP4 kernels (no DeepEP).
Compare DeepGEMM path against the Triton fallback inside vLLM's fused_experts.
"""
@@ -79,8 +78,7 @@ def make_block_quant_fp8_weights(
k: int,
block_size: list[int],
):
"""
Generate (w1, w2) expert weights and their per-block scale tensors
"""Generate (w1, w2) expert weights and their per-block scale tensors
in FP8 block-quantized format.
w1 shape: (E, 2N, K)
@@ -121,8 +119,7 @@ def make_block_quant_fp8_weights(
def run_single_case(m, n, k, topk, num_experts, block_size):
"""
Run one (M,N,K) configuration on a single GPU and assert DeepGEMM ==
"""Run one (M,N,K) configuration on a single GPU and assert DeepGEMM ==
Triton baseline within tolerance.
"""
tokens_bf16 = (
@@ -248,8 +245,7 @@ def make_mxfp4_weights(
n: int,
k: int,
):
"""
Generate (w1, w2) expert weights in MXFP4 packed format with float32 scales,
"""Generate (w1, w2) expert weights in MXFP4 packed format with float32 scales,
plus BF16 reference weights for validation.
w1 shape: (E, 2N, K//2) uint8 — packed FP4
@@ -310,8 +306,7 @@ def _bf16_moe_reference(x, w1, w2, topk_weights, topk_ids):
def run_single_fp4_case(m, n, k, topk, num_experts):
"""
Run one (M,N,K) configuration with FP4 weights on DeepGEMM and assert
"""Run one (M,N,K) configuration with FP4 weights on DeepGEMM and assert
DeepGEMM FP4 == BF16 reference within tolerance.
"""
tokens_bf16 = torch.randn(m, k, device="cuda", dtype=torch.bfloat16) * (k**-0.5)