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
+9 -11
View File
@@ -415,8 +415,7 @@ def postprocess_mamba_fused_kernel(
# the existing 2D-grid contract.
TEMPORAL_TILES: tl.constexpr = 1,
):
"""
Fused GPU kernel for postprocess_mamba that computes decisions AND performs
"""Fused GPU kernel for postprocess_mamba that computes decisions AND performs
mamba state copies without any CPU-GPU synchronization.
Grid: (num_reqs, num_states [, TEMPORAL_TILES])
@@ -769,8 +768,7 @@ class MambaCopyBuffers:
@dataclasses.dataclass
class MambaSpecDecodeGPUContext:
"""
Context for GPU-side Mamba state copy operations during the
"""Context for GPU-side Mamba state copy operations during the
fused postprocess path.
Only used when speculative decoding is enabled on a hybrid model
@@ -927,8 +925,7 @@ class MambaSpecDecodeGPUContext:
mamba_state_copy_funcs: MambaStateCopyFuncsByType,
block_tables: list[torch.Tensor],
) -> None:
"""
Extract and cache memory layout metadata from Mamba state tensors.
"""Extract and cache memory layout metadata from Mamba state tensors.
This method populates the pre-allocated metadata tensors with information
needed by `postprocess_mamba_fused_kernel` to perform state copies entirely
@@ -960,6 +957,7 @@ class MambaSpecDecodeGPUContext:
block_tables: per-mamba-group persistent block-table tensors, in
the same order as `mamba_group_ids`. Their `data_ptr()` /
`stride(0)` are captured once for the kernel to index into.
"""
if self.is_initialized:
return
@@ -1096,7 +1094,7 @@ class MambaSpecDecodeGPUContext:
seq_lens: torch.Tensor,
num_reqs: int,
) -> torch.Tensor:
"""compute every Mamba group's aligned physical state IDs in one launch."""
"""Compute every Mamba group's aligned physical state IDs in one launch."""
assert self.is_initialized
assert seq_lens.is_cuda
assert 0 <= num_reqs <= seq_lens.shape[0]
@@ -1137,8 +1135,7 @@ class MambaSpecDecodeGPUContext:
num_computed_tokens_gpu: torch.Tensor,
num_draft_tokens_gpu: torch.Tensor,
) -> None:
"""
Run the fused postprocess_mamba kernel on GPU.
"""Run the fused postprocess_mamba kernel on GPU.
This computes decisions and performs mamba state copies entirely on GPU,
eliminating the CPU-GPU sync that was previously needed.
@@ -1150,6 +1147,7 @@ class MambaSpecDecodeGPUContext:
num_scheduled_tokens_gpu: [num_reqs] scheduled token counts
num_computed_tokens_gpu: [num_reqs] computed token counts
num_draft_tokens_gpu: [num_reqs] draft token counts
"""
if num_reqs == 0 or not self.is_initialized:
return
@@ -1205,6 +1203,7 @@ class MambaSpecDecodeGPUContext:
token_bias_gpu: [max_reqs] accepted-token bias (num_accepted - 1).
idx_mapping: optional [num_reqs] batch_idx -> req_state_idx.
None means V1 batch order already equals request state order.
"""
if num_reqs == 0 or not self.is_initialized:
return
@@ -1442,8 +1441,7 @@ def preprocess_mamba(
copy_bufs: MambaCopyBuffers,
align_ctx: MambaSpecDecodeGPUContext | None = None,
):
"""
Copy the mamba state of previous step to the last
"""Copy the mamba state of previous step to the last
(1 + num_speculative_blocks) block.
"""
fused = _resolve_fused_precopy(align_ctx)