22 lines
705 B
Python
22 lines
705 B
Python
"""CUDA attention kernel wrappers with torch fallback.
|
|
|
|
Public API:
|
|
- ``attn_decode`` — single-query decode attention
|
|
- ``attn_prefill`` — multi-query prefill attention
|
|
- ``attn_paged_decode`` — paged decode attention (direct page-table access)
|
|
|
|
Each wrapper dispatches to its compiled CUDA kernel (``astrai.extension.attn_*``)
|
|
when available, otherwise falls back to ``torch.nn.functional.scaled_dot_product_attention``.
|
|
"""
|
|
|
|
from astrai.extension.loader import KERNEL_NAMES, is_available
|
|
from astrai.extension.ops import attn_decode, attn_paged_decode, attn_prefill
|
|
|
|
__all__ = [
|
|
"attn_decode",
|
|
"attn_paged_decode",
|
|
"attn_prefill",
|
|
"is_available",
|
|
"KERNEL_NAMES",
|
|
]
|