Skip to content

[Triton] [GFX12] Add gluon pa_decode_sparse from AITER #1290

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
k50112113 wants to merge 1 commit into
base: main
Choose a base branch
from
+13 −1
Open

[Triton] [GFX12] Add gluon pa_decode_sparse from AITER #1290

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 13 additions & 1 deletion atom/model_ops/v4_kernels/paged_decode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,7 @@
import triton.language as tl
from aiter.ops.triton.utils.device_info import get_num_sms
from atom.model_ops.sparse_attn_v4 import _sparse_attn_ragged_torch
from aiter.ops.triton.attention.pa_decode_sparse import pa_decode_sparse

LOG2E = 1.4426950408889634 # log2(e); folded into qk_scale so softmax can use exp2.
_MAX_KV_SPLITS = 64 # Hard cap on kv_splits (see _kv_splits_heuristic).
Expand Down Expand Up @@ -916,7 +917,18 @@ def sparse_attn_v4_paged_decode(
When ``kv_scales`` is provided, ``unified_kv`` must be fp8 (e4m3fnuz) and
will be dequantized in-kernel using 1xGROUP_SIZE (default 64) block scales.
"""
if os.environ.get("ATOM_USE_TRITON_ATTN", "1") == "1":
if os.environ.get("ATOM_USE_AITER_TRITON_ATTN", "0") == "1":
return pa_decode_sparse(
q,
unified_kv,
kv_indices,
kv_indptr,
attn_sink,
softmax_scale,
has_invalid=False,
kv_scales=kv_scales,
)
elif os.environ.get("ATOM_USE_TRITON_ATTN", "1") == "1":
return _sparse_attn_v4_paged_decode_triton(
q,
unified_kv,
Expand Down
Loading