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
@@ -60,7 +60,8 @@ def test_checkpoint_lm_head_can_override_tied_config(monkeypatch) -> None:
def test_gemma4_kv_shared_layer_loads_plain_q_proj() -> None:
"""KV-shared layers have q_proj instead of a packed qkv_proj; their
redundant K/V tensors in original checkpoints have no parameter and are
skipped rather than failing the load."""
skipped rather than failing the load.
"""
model = torch.nn.Module()
model.config = SimpleNamespace(num_experts=0)
model.start_layer, model.end_layer = 0, 2
@@ -92,7 +93,8 @@ def test_gemma4_kv_shared_layer_loads_plain_q_proj() -> None:
@pytest.mark.cpu_test
def test_gemma3n_kv_shared_layer_mapper() -> None:
"""Only non-shared layers pack q/k/v into qkv_proj; KV-shared layers keep
q_proj and drop the redundant K/V tensors original checkpoints ship."""
q_proj and drop the redundant K/V tensors original checkpoints ship.
"""
config = SimpleNamespace(num_hidden_layers=4, num_kv_shared_layers=2)
mapper = Gemma3nTextModel.hf_to_vllm_mapper | _kv_sharing_weights_mapper(config)
weights = [
@@ -164,8 +164,7 @@ def test_chunked_prefill_with_parallel_sampling(
max_tokens: int,
conv_state_layout: str,
) -> None:
"""
Tests chunked prefill in conjunction with n > 1.
"""Tests chunked prefill in conjunction with n > 1.
In this case, prefill is populated with decoding tokens and
we test that it doesn't fail.
@@ -199,8 +198,7 @@ def test_mamba_cache_cg_padding(
max_tokens: int,
conv_state_layout: str,
) -> None:
"""
This test is for verifying that mamba cache is padded to CG captured
"""This test is for verifying that mamba cache is padded to CG captured
batch size. If it's not, a torch RuntimeError will be raised because
tensor dimensions aren't compatible.
"""
@@ -236,8 +234,7 @@ def test_fail_upon_inc_requests_and_finished_requests_lt_available_blocks(
example_prompts,
model: str,
) -> None:
"""
This test is for verifying that the hybrid inner state management doesn't
"""This test is for verifying that the hybrid inner state management doesn't
collapse in case where the number of incoming requests and
finished_requests_ids is larger than the maximum mamba block capacity.
@@ -263,8 +260,7 @@ def test_state_cleanup(
example_prompts,
model: str,
) -> None:
"""
This test is for verifying that the Hybrid state is cleaned up between
"""This test is for verifying that the Hybrid state is cleaned up between
steps.
If it's not cleaned, an error would be expected.
@@ -274,7 +274,8 @@ def test_mistral_function_calling(vllm_runner, model: str, dtype: str) -> None:
def test_mistral_function_call_nested_json():
"""Ensure that the function-name regex captures the entire outermost
JSON block, including nested braces."""
JSON block, including nested braces.
"""
# Create a minimal stub tokenizer that provides the few attributes the
# parser accesses (`version` and `get_vocab`).
@@ -1,8 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""
Tests for max_tokens_per_doc and max_tokens_per_query.
"""
"""Tests for max_tokens_per_doc and max_tokens_per_query."""
import json
import os
@@ -29,7 +29,8 @@ from vllm.v1.worker.gpu.states import RequestState
@torch.inference_mode
def test_splade_pooler_matches_reference_formula(B, T, H, V):
"""Ensure SPLADESparsePooler forward() matches the mathematical formula:
log1p(relu(logits)) -> max over sequence length (after masking)."""
log1p(relu(logits)) -> max over sequence length (after masking).
"""
torch.manual_seed(0)
# Prepare [B] sequences of shape [T, H]