Skip to content

Allow BatchDecodeWithPagedKVCacheWrapper for GQA ratio 16 and 32 #2895

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

Closed
bkryu wants to merge 2 commits into from
Closed

Allow BatchDecodeWithPagedKVCacheWrapper for GQA ratio 16 and 32 #2895

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cb8f3d0
First commit
bkryu Mar 25, 2026
afacf80
Remove group size 64
bkryu Mar 25, 2026
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
9 changes: 9 additions & 0 deletions include/flashinfer/utils.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -150,6 +150,15 @@
} else if (group_size == 8) { \
constexpr size_t GROUP_SIZE = 8; \
__VA_ARGS__ \
} else if (group_size == 16) { \
constexpr size_t GROUP_SIZE = 16; \
__VA_ARGS__ \
} else if (group_size == 32) { \
constexpr size_t GROUP_SIZE = 32; \
__VA_ARGS__ \
} else if (group_size == 64) { \
constexpr size_t GROUP_SIZE = 64; \
__VA_ARGS__ \
Comment thread
coderabbitai[bot] marked this conversation as resolved.
Outdated
} else { \
std::ostringstream err_msg; \
err_msg << "Unsupported group_size: " << group_size; \
Expand Down
8 changes: 4 additions & 4 deletions tests/attention/test_batch_decode_kernels.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,7 @@ def warmup_jit():
@pytest.mark.parametrize("batch_size", [12, 17, 128])
@pytest.mark.parametrize("kv_len", [54, 97, 512, 2048, 16384])
@pytest.mark.parametrize("page_size", [1, 8, 16])
@pytest.mark.parametrize("num_kv_heads", [4])
@pytest.mark.parametrize("num_kv_heads", [2, 4])
@pytest.mark.parametrize("num_qo_heads", [4, 32])
@pytest.mark.parametrize("head_dim", [128, 256])
@pytest.mark.parametrize("kv_layout", ["NHD"])
Expand Down Expand Up @@ -199,7 +199,7 @@ def test_batch_decode_with_paged_kv_cache(
@pytest.mark.parametrize("batch_size", [12, 17, 128])
@pytest.mark.parametrize("kv_len", [54, 97, 512, 2048, 16384])
@pytest.mark.parametrize("page_size", [1, 8, 16])
@pytest.mark.parametrize("num_kv_heads", [4])
@pytest.mark.parametrize("num_kv_heads", [2, 4])
@pytest.mark.parametrize("num_qo_heads", [4, 32])
@pytest.mark.parametrize("head_dim", [128, 256])
@pytest.mark.parametrize("kv_layout", ["NHD"])
Expand Down Expand Up @@ -358,7 +358,7 @@ def test_batch_decode_with_paged_kv_cache_with_fast_plan(
@pytest.mark.parametrize("batch_size", [12, 17, 128])
@pytest.mark.parametrize("kv_len", [54, 97, 512, 2048, 16384])
@pytest.mark.parametrize("page_size", [1, 8, 16])
@pytest.mark.parametrize("num_kv_heads", [4])
@pytest.mark.parametrize("num_kv_heads", [2, 4])
@pytest.mark.parametrize("num_qo_heads", [4, 32])
@pytest.mark.parametrize("head_dim", [128, 256])
@pytest.mark.parametrize("kv_layout", ["NHD"])
Expand Down Expand Up @@ -496,7 +496,7 @@ def test_batch_decode_with_tuple_paged_kv_cache(
@pytest.mark.parametrize("batch_size", [12, 17, 128])
@pytest.mark.parametrize("kv_len", [54, 2048, 16384])
@pytest.mark.parametrize("page_size", [1, 8, 16])
@pytest.mark.parametrize("num_kv_heads", [4])
@pytest.mark.parametrize("num_kv_heads", [2, 4])
@pytest.mark.parametrize("num_qo_heads", [4, 32])
@pytest.mark.parametrize("head_dim", [128, 256])
@pytest.mark.parametrize("kv_layout", ["NHD"])
Expand Down
Loading