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:
co-authored by
Claude Opus 5
parent
804e5377aa
commit
8c1557a79c
@@ -138,14 +138,14 @@ class ApplyRotaryEmb(CustomOp):
|
||||
is_neox_style: bool = True,
|
||||
enable_fp32_compute: bool = False,
|
||||
) -> torch.Tensor:
|
||||
"""
|
||||
Args:
|
||||
x: [batch_size (optional), seq_len, num_heads, head_size]
|
||||
cos: [seq_len, head_size // 2]
|
||||
sin: [seq_len, head_size // 2]
|
||||
is_neox_style: Whether to use the Neox-style or GPT-J-style.
|
||||
enable_fp32_compute: Temporarily convert x, cos, sin to FP32 dtype
|
||||
for higher accuracy.
|
||||
"""Args:
|
||||
x: [batch_size (optional), seq_len, num_heads, head_size]
|
||||
cos: [seq_len, head_size // 2]
|
||||
sin: [seq_len, head_size // 2]
|
||||
is_neox_style: Whether to use the Neox-style or GPT-J-style.
|
||||
enable_fp32_compute: Temporarily convert x, cos, sin to FP32 dtype
|
||||
for higher accuracy.
|
||||
|
||||
"""
|
||||
origin_dtype = x.dtype
|
||||
if enable_fp32_compute:
|
||||
|
||||
Reference in New Issue
Block a user