Unverified Commit 87441885 authored by Charlene Yang's avatar Charlene Yang Committed by GitHub
Browse files

Minor fixes for attention (#1504)



* minor fixes for attention
Signed-off-by: default avatarCharlene Yang <charleney@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



---------
Signed-off-by: default avatarCharlene Yang <charleney@nvidia.com>
Co-authored-by: default avatarpre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
parent 229dd045
...@@ -153,7 +153,7 @@ NVTE_Fused_Attn_Backend nvte_get_fused_attn_backend( ...@@ -153,7 +153,7 @@ NVTE_Fused_Attn_Backend nvte_get_fused_attn_backend(
// TODO(cyang): replace with cudnn-frontend check_support for cleaner logic and better error messaging // TODO(cyang): replace with cudnn-frontend check_support for cleaner logic and better error messaging
// special conditions for blackwell // special conditions for blackwell
// TODO: enable THD max_t in f16_arbitrary_seqlen when support becomes available in 9.7 // TODO: enable THD max_t in f16_arbitrary_seqlen when support becomes available in 9.7
!(sm_arch_ == 100 && (head_dim_qk > 128 || head_dim_v > 128)) && !(sm_arch_ >= 100 && (head_dim_qk > 128 || head_dim_v > 128)) &&
// architecture // architecture
((cudnn_runtime_version >= 8903 && sm_arch_ >= 80) || ((cudnn_runtime_version >= 8903 && sm_arch_ >= 80) ||
(cudnn_runtime_version < 8903 && (sm_arch_ == 80 || sm_arch_ == 90))) && (cudnn_runtime_version < 8903 && (sm_arch_ == 80 || sm_arch_ == 90))) &&
...@@ -238,12 +238,12 @@ NVTE_Fused_Attn_Backend nvte_get_fused_attn_backend( ...@@ -238,12 +238,12 @@ NVTE_Fused_Attn_Backend nvte_get_fused_attn_backend(
((window_size_left >= 0 || window_size_left == -1) && window_size_right == 0 && ((window_size_left >= 0 || window_size_left == -1) && window_size_right == 0 &&
((attn_mask_type == NVTE_Mask_Type::NVTE_CAUSAL_BOTTOM_RIGHT_MASK && ((attn_mask_type == NVTE_Mask_Type::NVTE_CAUSAL_BOTTOM_RIGHT_MASK &&
// TODO(cyang): fix bug for BRCM + cross-attention on sm100 // TODO(cyang): fix bug for BRCM + cross-attention on sm100
(sm_arch_ < 100 || (sm_arch_ == 100 && ((max_seqlen_q == max_seqlen_kv && (sm_arch_ < 100 || (sm_arch_ >= 100 && ((max_seqlen_q == max_seqlen_kv &&
cudnn_runtime_version <= 90700) || cudnn_runtime_version <= 90700) ||
cudnn_runtime_version > 90700)))) || cudnn_runtime_version > 90700)))) ||
attn_mask_type == NVTE_Mask_Type::NVTE_PADDING_CAUSAL_MASK || attn_mask_type == NVTE_Mask_Type::NVTE_PADDING_CAUSAL_MASK ||
(attn_mask_type == NVTE_Mask_Type::NVTE_PADDING_CAUSAL_BOTTOM_RIGHT_MASK && (attn_mask_type == NVTE_Mask_Type::NVTE_PADDING_CAUSAL_BOTTOM_RIGHT_MASK &&
(sm_arch_ < 100 || (sm_arch_ == 100 && ((max_seqlen_q == max_seqlen_kv && (sm_arch_ < 100 || (sm_arch_ >= 100 && ((max_seqlen_q == max_seqlen_kv &&
cudnn_runtime_version <= 90700) || cudnn_runtime_version <= 90700) ||
cudnn_runtime_version > 90700))))) && cudnn_runtime_version > 90700))))) &&
max_seqlen_q <= max_seqlen_kv && bias_type == NVTE_Bias_Type::NVTE_NO_BIAS && max_seqlen_q <= max_seqlen_kv && bias_type == NVTE_Bias_Type::NVTE_NO_BIAS &&
......
...@@ -118,7 +118,7 @@ _flash_attn_is_installed = False ...@@ -118,7 +118,7 @@ _flash_attn_is_installed = False
_flash_attn_version = PkgVersion("0") _flash_attn_version = PkgVersion("0")
_flash_attn_version_required = PkgVersion("2.1.1") _flash_attn_version_required = PkgVersion("2.1.1")
_flash_attn_version_required_blackwell = PkgVersion("2.7.3") _flash_attn_version_required_blackwell = PkgVersion("2.7.3")
_flash_attn_max_version = PkgVersion("2.7.3") _flash_attn_max_version = PkgVersion("2.7.4.post1")
_flash_attn_2_plus = False _flash_attn_2_plus = False
_flash_attn_2_1_plus = False _flash_attn_2_1_plus = False
_flash_attn_2_3_plus = False _flash_attn_2_3_plus = False
...@@ -507,13 +507,16 @@ def get_attention_backend( ...@@ -507,13 +507,16 @@ def get_attention_backend(
if use_flash_attention and ( if use_flash_attention and (
head_dim_qk > 256 head_dim_qk > 256
or head_dim_qk % 8 != 0 or head_dim_qk % 8 != 0
or (head_dim_qk > 192 and device_compute_capability not in ((8, 0), (9, 0))) or (
head_dim_qk > 192
and device_compute_capability not in ((8, 0), (9, 0), (10, 0), (12, 0))
)
): ):
if _flash_attn_is_installed: if _flash_attn_is_installed:
logger.debug( logger.debug(
"Disabling FlashAttention due to unsupported head_dim_qk and head_dim_v. " "Disabling FlashAttention due to unsupported head_dim_qk and head_dim_v. "
"Supported: head_dim_qk = head_dim_v, head_dim_qk %%8 = 0, " "Supported: head_dim_qk = head_dim_v, head_dim_qk %%8 = 0, "
"head_dim_qk <= 256 (>192 requires sm80/90). " "head_dim_qk <= 256 (>192 requires sm80/90/100+). "
"Found: head_dim_qk = %s, head_dim_v = %s, on sm%s.", "Found: head_dim_qk = %s, head_dim_v = %s, on sm%s.",
head_dim_qk, head_dim_qk,
head_dim_v, head_dim_v,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment