Skip to content
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

[Common/PyTorch] Fix FP8 fused attention input args #592

Merged
merged 3 commits into from
Jan 8, 2024
Merged

[Common/PyTorch] Fix FP8 fused attention input args #592

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ae82c41
fix FP8 dims
cyanguwa Jan 6, 2024
6a2deab
Merge branch 'NVIDIA:main' into fused_attn/fp8_dims
cyanguwa Jan 6, 2024
e262b56
Merge branch 'main' into fused_attn/fp8_dims
cyanguwa Jan 6, 2024
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
2 changes: 1 addition & 1 deletion tests/pytorch/test_fused_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -842,7 +842,7 @@ def _run_transformer_layer(

@pytest.mark.skipif(_cudnn_version() < (8,9,3), reason="cuDNN 8.9.3+ is required.")
@pytest.mark.skipif(not fp8_available, reason=reason_for_no_fp8)
@pytest.mark.skipif(get_device_compute_capability != (9, 0), reason="FP8 tests require Hopper.")
@pytest.mark.skipif(get_device_compute_capability() != (9, 0), reason="FP8 tests require Hopper.")
@pytest.mark.parametrize("dtype", param_types_fp8)
@pytest.mark.parametrize("model", model_configs_fp8.keys())
def test_dpa_fp8(dtype, model):
Expand Down
12 changes: 4 additions & 8 deletions transformer_engine/common/fused_attn/fused_attn_fp8.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1862,8 +1862,7 @@ void fused_attn_fp8_bwd_impl(int64_t b, int64_t h, int64_t s_q, int64_t s_kv, in
#if (CUDNN_VERSION >= 8900)
// fused attention FWD FP8 with packed QKV
void fused_attn_fp8_fwd_qkvpacked(
size_t b, size_t max_seqlen,
size_t h, size_t d,
size_t b, size_t h, size_t max_seqlen, size_t d,
bool is_training, float attn_scale,
float p_dropout, NVTE_QKV_Layout qkv_layout,
const Tensor *input_QKV,
Expand Down Expand Up @@ -1960,8 +1959,7 @@ void fused_attn_fp8_fwd_qkvpacked(
}
// fused attention BWD FP8 with packed QKV
void fused_attn_fp8_bwd_qkvpacked(
size_t b, size_t max_seqlen,
size_t h, size_t d,
size_t b, size_t h, size_t max_seqlen, size_t d,
float attn_scale, float p_dropout, NVTE_QKV_Layout qkv_layout,
const Tensor *input_QKV,
const Tensor *input_O,
Expand Down Expand Up @@ -2055,8 +2053,7 @@ void fused_attn_fp8_bwd_qkvpacked(
}
// fused attention FWD FP8 with separate Q, K, V
void fused_attn_fp8_fwd(
size_t b, size_t max_seqlen_q, size_t max_seqlen_kv,
size_t h, size_t d,
size_t b, size_t h, size_t max_seqlen_q, size_t max_seqlen_kv, size_t d,
bool is_training, float attn_scale,
float p_dropout, NVTE_QKV_Layout qkv_layout,
const Tensor *input_Q,
Expand Down Expand Up @@ -2156,8 +2153,7 @@ void fused_attn_fp8_fwd(
}
// fused attention BWD FP8 with separate Q, K, V
void fused_attn_fp8_bwd(
size_t b, size_t max_seqlen_q, size_t max_seqlen_kv,
size_t h, size_t d,
size_t b, size_t h, size_t max_seqlen_q, size_t max_seqlen_kv, size_t d,
float attn_scale, float p_dropout, NVTE_QKV_Layout qkv_layout,
const Tensor *input_Q,
const Tensor *input_K,
Expand Down