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
@@ -118,8 +118,8 @@ def test_per_token_group_quant_fp8_packed(
num_tokens, hidden_dim, group_size, poisoned_scales
):
"""Test the packed DeepGEMM quantization kernel against the Triton
reference (row-major, UE8M0 scales)."""
reference (row-major, UE8M0 scales).
"""
device = "cuda"
torch.manual_seed(42)
@@ -206,7 +206,6 @@ def test_per_token_group_quant_fp8_packed_all_zero():
For all-zero input, eps/fp8_max < 1e-10, so the inner fmax clamps back to
1e-10, giving exp2(ceil(log2(1e-10))) = exp2(-33) => UE8M0 byte 0x5E (94).
"""
device = "cuda"
num_tokens, hidden_dim, group_size = 4, 7168, 128
x = torch.zeros((num_tokens, hidden_dim), device=device, dtype=torch.bfloat16)
@@ -258,8 +257,8 @@ def test_per_token_group_quant_fp8_packed_all_zero():
def test_per_token_group_quant_fp8_packed_mantissa_rounds_up():
"""Inputs whose absmax/max_8bit produces a non-power-of-2 force the
mantissa-rounding-up branch (exp_byte += 1). Locks down this behavior
before optimization."""
before optimization.
"""
device = "cuda"
num_tokens, hidden_dim, group_size = 4, 7168, 128
@@ -331,8 +330,8 @@ def test_per_token_group_quant_fp8_packed_zero_fills_padded_output_q(
):
"""When output_q is allocated with shape (tma_aligned_mn, k) instead of
(mn, k), the kernel must overwrite the padded mn rows with zeros so
callers can use ``torch.empty`` instead of ``torch.zeros``."""
callers can use ``torch.empty`` instead of ``torch.zeros``.
"""
device = "cuda"
group_size = 128
torch.manual_seed(42)
@@ -396,7 +395,6 @@ def test_per_token_group_quant_fp8_packed_large_mn():
This is a differential test that compares fp8 output against Triton output
reference when token size sits just above the gridDim.y 2^16 - 1 limit.
"""
device = "cuda"
group_size = 128
# hidden 2048 -> 2048/128 = 16 groups per row -> kx=16, ry=1: one grid row per mn