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
+14 -11
View File
@@ -275,14 +275,14 @@ class MiMoVisionAttention(nn.Module):
max_seqlen: torch.Tensor,
full_attn: bool = True,
) -> torch.Tensor:
"""
Args:
x: [seq_len, batch=1, embed_dim] (seq-first convention)
cu_seqlens: cumulative sequence lengths [num_seqs+1], int32
rotary_pos_emb_cos: [seq_len, qk_channels // 2]
rotary_pos_emb_sin: [seq_len, qk_channels // 2]
max_seqlen: maximum sequence length
full_attn: if True, full attention; if False, window attention
"""Args:
x: [seq_len, batch=1, embed_dim] (seq-first convention)
cu_seqlens: cumulative sequence lengths [num_seqs+1], int32
rotary_pos_emb_cos: [seq_len, qk_channels // 2]
rotary_pos_emb_sin: [seq_len, qk_channels // 2]
max_seqlen: maximum sequence length
full_attn: if True, full attention; if False, window attention
"""
# [seq_len, 1, embed_dim] -> QKV projection
qkv, _ = self.qkv(x) # [seq_len, 1, q_size + kv_size + kv_size]
@@ -531,6 +531,7 @@ class MiMoVisionTransformer(nn.Module):
Returns:
cos: [total_tokens, qk_channels // 2]
sin: [total_tokens, qk_channels // 2]
"""
cos_list, sin_list = [], []
for i in range(grid_thw.size(0)):
@@ -576,12 +577,12 @@ class MiMoVisionTransformer(nn.Module):
return torch.cat(cos_list, dim=0), torch.cat(sin_list, dim=0)
def forward(self, x: torch.Tensor, grid_thw: torch.Tensor) -> torch.Tensor:
"""
Args:
"""Args:
x: [total_tokens, C] pre-flattened patches
grid_thw: [num_images, 3] tensor of (t, h, w) for each image/video
Returns:
[merged_tokens, out_hidden_size]
"""
# Ensure grid_thw is a tensor
if not isinstance(grid_thw, torch.Tensor):
@@ -1556,8 +1557,10 @@ class MiMoV2OmniForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsQ
batch. **NOTE**: If mrope is enabled (default setting for
Qwen2.5-VL opensource models), the shape will be `(3, seq_len)`,
otherwise it will be `(seq_len,).
"""
intermediate_tensors: Intermediate tensors from prior forward pass.
inputs_embeds: Optional tensor of input embeddings.
"""
if intermediate_tensors is not None:
inputs_embeds = None