[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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
cyanguwa wants to merge 11 commits into from
Closed

[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16 #220

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
4b12509
add FusedAttention module
cyanguwa May 8, 2023
a859095
WIP: add higher level PyTorch modules for cuDNN fused attention imple…
cyanguwa May 9, 2023
6d351aa
WIP: add PyTorch moduels for fused attention
cyanguwa May 9, 2023
961a416
WIP: add bias/fp8_meta all the way up to TransformerLayer
cyanguwa May 9, 2023
93406ed
WIP: fix ctx_aux_tensors
cyanguwa May 9, 2023
eb72a58
WIP: finished Python side
cyanguwa May 10, 2023
cb65523
add the C parts to be compatible with higher-level changes
cyanguwa May 10, 2023
4bfe5ae
rebase after bf16/fp16 <=512 seqlen merge (PR 175)
cyanguwa May 12, 2023
5497f3c
WIP: test fused attn vs flash attn; different results
cyanguwa May 12, 2023
aa2516f
Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance
cyanguwa May 13, 2023
d779873
remove prints and torch saves
cyanguwa May 13, 2023
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
3 changes: 2 additions & 1 deletion transformer_engine/CMakeLists.txt
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down
215 changes: 90 additions & 125 deletions transformer_engine/common/fused_attn/fused_attn.cpp
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type attn_mask_type,
NVTETensor workspace,
cudaStream_t stream) {
cudaStream_t stream,
bool return_softmax,
int num_split,
int fused_attention_backend) {
NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
const DType QKV_type = input_QKV->data.dtype;

if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
&& (max_seqlen <= 512)) {
if (fused_attention_backend == 1) {
NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
// uses return_softmax and num_split
} else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
fused_attn_max_512_fwd_qkvpacked(
b, max_seqlen, h, d,
is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
input_QKV, input_Bias, output_O,
Aux_Output_Tensors,
input_cu_seqlens,
input_rng_state,
wkspace, stream, handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if (fused_attention_backend == 3) {
NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
} else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
// FP8 API doesn't use input_Bias, bias_type or attn_mask_type
fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
} else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
&& (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
fused_attn_max_512_fwd_qkvpacked(
b,
max_seqlen,
h,
d,
is_training,
attn_scale,
dropout,
qkv_layout,
bias_type,
attn_mask_type,
input_QKV,
input_Bias,
output_O,
Aux_Output_Tensors,
input_cu_seqlens,
input_rng_state,
wkspace,
stream,
handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if (max_seqlen > 512) {
NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
} else {
NVTE_ERROR("Invalid combination of data type and sequence length! \n");
}
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type attn_mask_type,
NVTETensor workspace,
cudaStream_t stream) {
cudaStream_t stream,
int num_split,
int fused_attention_backend) {
NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
const DType QKV_type = input_QKV->data.dtype;

if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
&& (max_seqlen <= 512)) {
if (fused_attention_backend == 1) {
NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
// uses return_softmax and num_split
} else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
fused_attn_max_512_bwd_qkvpacked(
b, max_seqlen, h, d,
attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
input_QKV, input_dO,
// Aux_CTX_Tensors,
output_S,
output_dQKV, output_dBias,
input_cu_seqlens,
wkspace, stream, handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if (fused_attention_backend == 3) {
NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
} else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
// Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
} else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
&& (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
fused_attn_max_512_bwd_qkvpacked(
b,
max_seqlen,
h,
d,
attn_scale,
dropout,
qkv_layout,
bias_type,
attn_mask_type,
input_QKV,
input_dO,
Aux_CTX_Tensors,
output_dQKV,
output_dBias,
input_cu_seqlens,
wkspace,
stream,
handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if (max_seqlen > 512) {
NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
} else {
NVTE_ERROR("Invalid combination of data type and sequence length! \n");
}
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type attn_mask_type,
NVTETensor workspace,
cudaStream_t stream) {
cudaStream_t stream,
bool return_softmax,
int num_split,
int fused_attention_backend) {
NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
using namespace transformer_engine;
const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
const DType QKV_type = input_Q->data.dtype;

if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
&& (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
} else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
&& (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
if (fused_attention_backend == 1) {
NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
// uses return_softmax and num_split
} else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
fused_attn_max_512_fwd_kvpacked(
b,
max_seqlen_q,
max_seqlen_kv,
h,
d,
is_training,
attn_scale,
dropout,
qkv_layout,
bias_type,
attn_mask_type,
input_Q,
input_KV,
input_Bias,
output_O,
Aux_Output_Tensors,
input_cu_seqlens_q,
input_cu_seqlens_kv,
input_rng_state,
wkspace,
stream,
handle);
fused_attn_max_512_fwd_kvpacked(
b, max_seqlen_q, max_seqlen_kv, h, d,
is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
input_Q, input_KV, input_Bias, output_O,
Aux_Output_Tensors,
input_cu_seqlens_q, input_cu_seqlens_kv,
input_rng_state,
wkspace, stream, handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
} else if (fused_attention_backend == 3) {
NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
} else if (fused_attention_backend == 4) {
NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
} else {
NVTE_ERROR("Invalid combination of data type and sequence length! \n");
}
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type attn_mask_type,
NVTETensor workspace,
cudaStream_t stream) {
cudaStream_t stream,
int num_split,
int fused_attention_backend) {
NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
using namespace transformer_engine;
const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
const DType QKV_type = input_Q->data.dtype;

if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
&& (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
} else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
&& (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
if (fused_attention_backend == 1) {
NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
// uses return_softmax and num_split
} else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
fused_attn_max_512_bwd_kvpacked(
b,
max_seqlen_q,
max_seqlen_kv,
h,
d,
attn_scale,
dropout,
qkv_layout,
bias_type,
attn_mask_type,
input_Q,
input_KV,
input_dO,
Aux_CTX_Tensors,
output_dQ,
output_dKV,
output_dBias,
input_cu_seqlens_q,
input_cu_seqlens_kv,
wkspace,
stream,
handle);
Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
fused_attn_max_512_bwd_kvpacked(
b, max_seqlen_q, max_seqlen_kv, h, d,
attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
output_S,
output_dQ, output_dKV, output_dBias,
input_cu_seqlens_q, input_cu_seqlens_kv,
wkspace, stream, handle);
#else
NVTE_ERROR(
"cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
} else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
} else if (fused_attention_backend == 3) {
NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
} else if (fused_attention_backend == 4) {
NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
} else {
NVTE_ERROR("Invalid combination of data type and sequence length! \n");
}
Expand Down
30 changes: 18 additions & 12 deletions transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
size_t head_dim, float attn_scale, float p_dropout,
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type mask_type, const Tensor *input_QKV,
const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
const Tensor *input_dO, Tensor *output_S,
Tensor *output_dQKV, Tensor *output_dBias,
const Tensor *cu_seqlens, Tensor *workspace,
cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

void *devPtrdBias = output_dBias->data.dptr;

NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
void *devPtrS = nullptr;
if (Aux_CTX_Tensors->size == 1) {
Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
devPtrS = output_S->data.dptr;
}
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
devPtrS = output_S->data.dptr;

// devPtrdS reuses the memory of devPtrS
void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
float p_dropout, NVTE_QKV_Layout qkv_layout,
NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
const Tensor *input_Q, const Tensor *input_KV,
const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
const Tensor *input_dO, Tensor *output_S,
Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

void *devPtrdBias = output_dBias->data.dptr;

NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
void *devPtrS = nullptr;
if (Aux_CTX_Tensors->size == 1) {
Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
devPtrS = output_S->data.dptr;
}
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
devPtrS = output_S->data.dptr;

// devPtrdS reuses the memory of devPtrS
void *devPtrdS = devPtrS;

Expand Down
6 changes: 4 additions & 2 deletions transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
size_t head_dim, float attn_scale, float p_dropout,
NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
NVTE_Mask_Type mask_type, const Tensor *input_QKV,
const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
//const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
const Tensor *input_dO, Tensor *output_S,
Tensor *output_dQKV, Tensor *output_dBias,
const Tensor *cu_seqlens, Tensor *workspace,
cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
float p_dropout, NVTE_QKV_Layout qkv_layout,
NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
const Tensor *input_Q, const Tensor *input_KV,
const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
//const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
const Tensor *input_dO, Tensor *output_S,
Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading
, 'i'); if (__m === '*' || __re.test(location.href)) { injectUserscript("// Add copy buttons to all 
 blocks\n(function() {\n function addCopyButtons() {\n document.querySelectorAll('pre code').forEach(function(codeBlock) {\n if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;\n codeBlock.parentElement.setAttribute('data-copy-added', 'true');\n \n var btn = document.createElement('button');\n btn.textContent = 'Copy';\n btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';\n btn.onmouseover = function() { this.style.opacity = '1'; };\n btn.onmouseout = function() { this.style.opacity = '0.7'; };\n btn.onclick = function() {\n navigator.clipboard.writeText(codeBlock.textContent).then(function() {\n btn.textContent = 'Copied!';\n setTimeout(function() { btn.textContent = 'Copy'; }, 1500);\n });\n };\n codeBlock.parentElement.style.position = 'relative';\n codeBlock.parentElement.appendChild(btn);\n });\n }\n \n addCopyButtons();\n \n // Re-run on dynamic content\n var observer = new MutationObserver(addCopyButtons);\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Add Copy Buttons to Code Blocks");
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Force GitHub README to respect dark mode\n(function() {\n var style = document.createElement('style');\n style.textContent = '\n .markdown-body {\n color-scheme: dark light;\n }\n .markdown-body pre { background: #161b22 !important; }\n .markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }\n .markdown-body table th, .markdown-body table td { border-color: #30363d !important; }\n .markdown-body img { background: #0d1117; }\n .markdown-body blockquote { border-left-color: #8b949e; }\n .markdown-body hr { border-color: #30363d; }\n ';\n document.head.appendChild(style);\n})();", "GitHub Dark Mode README Fix");
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Highlight search terms from Google/DuckDuckGo/Bing referrer\n(function() {\n var ref = document.referrer;\n var terms = [];\n \n if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {\n var url = new URL(ref);\n var q = url.searchParams.get('q') || url.searchParams.get('p');\n if (q) {\n terms = q.split(/\\s+/).filter(function(t) { return t.length > 2; });\n }\n }\n \n if (terms.length === 0) return;\n \n var style = document.createElement('style');\n style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';\n document.head.appendChild(style);\n \n function highlight(node) {\n if (node.nodeType === 3) { // text node\n var text = node.textContent;\n var found = false;\n terms.forEach(function(term) {\n var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\') + ')', 'gi');\n if (regex.test(text)) {\n found = true;\n var frag = document.createDocumentFragment();\n var parts = text.split(regex);\n parts.forEach(function(part, i) {\n if (i % 2 === 0) {\n frag.appendChild(document.createTextNode(part));\n } else {\n var span = document.createElement('span');\n span.className = 'userscript-highlight';\n span.textContent = part;\n frag.appendChild(span);\n }\n });\n node.parentNode.replaceChild(frag, node);\n }\n });\n } else if (node.nodeType === 1 && node.childNodes) { // element\n var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];\n if (!skipTags.includes(node.tagName)) {\n Array.from(node.childNodes).forEach(highlight);\n }\n }\n }\n \n highlight(document.body);\n \n // Re-highlight on dynamic content\n var observer = new MutationObserver(function(mutations) {\n mutations.forEach(function(m) {\n m.addedNodes.forEach(function(node) {\n if (node.nodeType === 1 || node.nodeType === 3) highlight(node);\n });\n });\n });\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Highlight Search Terms");
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Strip utm_, fbclid, gclid, etc. from all links on page\n(function() {\n var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',\n 'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',\n 'ref', 'ref_src', 'source', 'medium', 'campaign'];\n \n function cleanUrl(url) {\n try {\n var u = new URL(url, window.location.origin);\n var changed = false;\n trackingParams.forEach(function(p) {\n if (u.searchParams.has(p)) {\n u.searchParams.delete(p);\n changed = true;\n }\n });\n return changed ? u.toString() : url;\n } catch (e) {\n return url;\n }\n }\n \n function cleanLinks() {\n document.querySelectorAll('a[href]').forEach(function(a) {\n var clean = cleanUrl(a.href);\n if (clean !== a.href) a.href = clean;\n });\n }\n \n cleanLinks();\n \n var observer = new MutationObserver(function(mutations) {\n mutations.forEach(function(m) {\n m.addedNodes.forEach(function(node) {\n if (node.nodeType === 1) {\n if (node.tagName === 'A') cleanLinks();\n node.querySelectorAll('a[href]').forEach(function(a) {\n var clean = cleanUrl(a.href);\n if (clean !== a.href) a.href = clean;\n });\n }\n });\n });\n });\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Remove Tracking Parameters from Links");
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Auto-enable theater mode on YouTube\n(function() {\n function tryTheater() {\n var btn = document.querySelector('button[aria-label=\"Theater mode\"], ytd-player #player button[title=\"Theater mode\"]');\n if (btn && !btn.classList.contains('activated')) {\n btn.click();\n }\n }\n \n // Try immediately\n tryTheater();\n \n // Try after navigation (SPA)\n var lastUrl = location.href;\n setInterval(function() {\n if (location.href !== lastUrl) {\n lastUrl = location.href;\n setTimeout(tryTheater, 500);\n }\n }, 1000);\n \n // Also try on player load\n var observer = new MutationObserver(tryTheater);\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "YouTube Theater Mode Default");
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Remove or un-stick sticky/fixed headers that block content\n(function() {\n function unstick() {\n document.querySelectorAll('header, nav, [role=\"banner\"], .header, .navbar, .sticky, .fixed-top, [style*=\"position: fixed\"], [style*=\"position:sticky\"]').forEach(function(el) {\n if (el.style.position === 'fixed' || el.style.position === 'sticky' || \n getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {\n el.style.position = 'static';\n el.style.top = 'auto';\n el.style.zIndex = 'auto';\n }\n });\n }\n \n unstick();\n \n var observer = new MutationObserver(unstick);\n observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });\n})();", "Kill Sticky Headers");
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Universal Dark Mode - works on any site\n(function() {\n var enabled = true;\n \n function applyDarkMode() {\n if (!enabled) return;\n \n // Create style element if it doesn't exist\n var style = document.getElementById('universal-dark-mode-style');\n if (!style) {\n style = document.createElement('style');\n style.id = 'universal-dark-mode-style';\n document.head.appendChild(style);\n }\n \n // Dark mode CSS - inverts colors but preserves images/video\n style.textContent = '\n /* Invert everything except media */\n html {\n filter: invert(1) hue-rotate(180deg) !important;\n background: #1a1a2e !important;\n }\n \n /* Restore images, videos, iframes, canvas */\n img, video, iframe, canvas, svg, picture, [style*=\"background-image\"] {\n filter: invert(1) hue-rotate(180deg) !important;\n }\n \n /* Preserve specific elements that should not be inverted */\n .no-dark-mode, .no-dark-mode *,\n [data-theme=\"light\"], [data-theme=\"light\"],\n .ace_editor, .ace_editor *,\n .CodeMirror, .CodeMirror *,\n .monaco-editor, .monaco-editor *,\n .markdown-body pre, .markdown-body pre *,\n .highlight, .highlight *,\n pre code, pre code * {\n filter: none !important;\n }\n \n /* Fix common UI elements */\n .modal, .popup, .dropdown-menu, .tooltip, .popover {\n filter: invert(1) hue-rotate(180deg) !important;\n background: #2d2d44 !important;\n border-color: #444 !important;\n }\n \n /* Scrollbars */\n ::-webkit-scrollbar { background: #1a1a2e !important; }\n ::-webkit-scrollbar-thumb { background: #444 !important; }\n ::-webkit-scrollbar-thumb:hover { background: #555 !important; }\n \n /* Selection */\n ::selection { background: #4ecdc4 !important; color: #1a1a2e !important; }\n ::-moz-selection { background: #4ecdc4 !important; color: #1a1a2e !important; }\n ';\n }\n \n function removeDarkMode() {\n var style = document.getElementById('universal-dark-mode-style');\n if (style) style.remove();\n }\n \n // Toggle with Alt+Shift+D\n document.addEventListener('keydown', function(e) {\n if (e.altKey && e.shiftKey && e.key === 'D') {\n e.preventDefault();\n enabled = !enabled;\n if (enabled) {\n applyDarkMode();\n console.log('[Universal Dark Mode] Enabled');\n } else {\n removeDarkMode();\n console.log('[Universal Dark Mode] Disabled');\n }\n }\n });\n \n // Apply on load\n applyDarkMode();\n \n // Re-apply on dynamic content\n var observer = new MutationObserver(function(mutations) {\n if (enabled && !document.getElementById('universal-dark-mode-style')) {\n applyDarkMode();\n }\n });\n observer.observe(document.head, { childList: true });\n \n console.log('[Universal Dark Mode] Loaded - Press Alt+Shift+D to toggle');\n})();", "Universal Dark Mode");
}
} catch(__e) { console.warn('[Userscript:Universal Dark Mode]', __e); }
})();
})();

Skip to content
[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16#220

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

cyanguwa
wants to merge
11
commits into

from

 Closed


[WIP] Add PyTorch modules for fused attention for FP8 and BF16/FP16
#220
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


11 commits


Select commit
Hold shift + click to select a range
4b12509

add FusedAttention module


cyanguwa May 8, 2023
a859095

WIP: add higher level PyTorch modules for cuDNN fused attention imple…


cyanguwa May 9, 2023
6d351aa

WIP: add PyTorch moduels for fused attention


cyanguwa May 9, 2023
961a416

WIP: add bias/fp8_meta all the way up to TransformerLayer


cyanguwa May 9, 2023
93406ed

WIP: fix ctx_aux_tensors


cyanguwa May 9, 2023
eb72a58

WIP: finished Python side


cyanguwa May 10, 2023
cb65523

add the C parts to be compatible with higher-level changes


cyanguwa May 10, 2023
4bfe5ae

rebase after bf16/fp16 <=512 seqlen merge (PR 175)


cyanguwa May 12, 2023
5497f3c

WIP: test fused attn vs flash attn; different results


cyanguwa May 12, 2023
aa2516f

Working: bf16/fp16 <=512, fused attn = flash attn with small tolerance


cyanguwa May 13, 2023
d779873

remove prints and torch saves


cyanguwa May 13, 2023

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


3 changes: 2 additions & 1 deletion
transformer_engine/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -5,7 +5,8 @@
cmake_minimum_required(VERSION 3.18)

if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
 set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
	#set(CMAKE_CUDA_ARCHITECTURES 70 80 89 90)
 set(CMAKE_CUDA_ARCHITECTURES 90)
endif()

set(CMAKE_CXX_STANDARD 17)
Expand Down

215 changes: 90 additions & 125 deletions
transformer_engine/common/fused_attn/fused_attn.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -24,7 +24,10 @@ void nvte_fused_attn_fwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -45,8 +48,26 @@ void nvte_fused_attn_fwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b, max_seqlen, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // FP8 API doesn't use input_Bias, bias_type or attn_mask_type
 fused_attn_fwd_fp8_qkvpacked(
Expand All@@ -60,35 +81,6 @@ void nvte_fused_attn_fwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens,
 input_rng_state,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -109,7 +101,9 @@ void nvte_fused_attn_bwd_qkvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_qkvpacked);
 using namespace transformer_engine;

Expand All@@ -132,8 +126,28 @@ void nvte_fused_attn_bwd_qkvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_QKV->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_qkvpacked(
 b, max_seqlen, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_QKV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQKV, output_dBias,
 input_cu_seqlens,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
#if (CUDNN_VERSION >= 8900)
 // Aux_CTX_Tensors contain [M, ZInv, rng_state] generated by the forward pass
 const Tensor *input_M = reinterpret_cast<const Tensor*>(Aux_CTX_Tensors->tensors[0]);
Expand All@@ -154,34 +168,6 @@ void nvte_fused_attn_bwd_qkvpacked(
#else
 NVTE_ERROR("cuDNN 8.9 is required to run FP8 fused attention. \n");
#endif
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen <= 512)) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_qkvpacked(
 b,
 max_seqlen,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_QKV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQKV,
 output_dBias,
 input_cu_seqlens,
 wkspace,
 stream,
 handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if (max_seqlen > 512) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -202,7 +188,10 @@ void nvte_fused_attn_fwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 bool return_softmax,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_fwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -225,41 +214,27 @@ void nvte_fused_attn_fwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_fwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 is_training,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_Bias,
 output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 input_rng_state,
 wkspace,
 stream,
 handle);
 fused_attn_max_512_fwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 is_training, attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_Bias, output_O,
 Aux_Output_Tensors,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 input_rng_state,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand All@@ -283,7 +258,9 @@ void nvte_fused_attn_bwd_kvpacked(
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type attn_mask_type,
 NVTETensor workspace,
 cudaStream_t stream) {
 cudaStream_t stream,
 int num_split,
 int fused_attention_backend) {
 NVTE_API_CALL(nvte_flash_attn_bwd_kvpacked);
 using namespace transformer_engine;
 const Tensor *input_cu_seqlens_q = reinterpret_cast<const Tensor*>(cu_seqlens_q);
Expand All@@ -309,41 +286,29 @@ void nvte_fused_attn_bwd_kvpacked(
 auto handle = cudnnExecutionPlanManager::Instance().GetCudnnHandle();
 const DType QKV_type = input_Q->data.dtype;

 if (((QKV_type == DType::kFloat8E4M3) || (QKV_type == DType::kFloat8E5M2))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else if (((QKV_type == DType::kFloat16) || (QKV_type == DType::kBFloat16))
 && (max_seqlen_q <= 512) && (max_seqlen_kv <= 512)) {
 if (fused_attention_backend == 1) {
 NVTE_ERROR("TODO: No support for FlashAttention C API currently. \n");
 // uses return_softmax and num_split
 } else if (fused_attention_backend == 2) {
#if (CUDNN_VERSION >= 8901)
 fused_attn_max_512_bwd_kvpacked(
 b,
 max_seqlen_q,
 max_seqlen_kv,
 h,
 d,
 attn_scale,
 dropout,
 qkv_layout,
 bias_type,
 attn_mask_type,
 input_Q,
 input_KV,
 input_dO,
 Aux_CTX_Tensors,
 output_dQ,
 output_dKV,
 output_dBias,
 input_cu_seqlens_q,
 input_cu_seqlens_kv,
 wkspace,
 stream,
 handle);
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 fused_attn_max_512_bwd_kvpacked(
 b, max_seqlen_q, max_seqlen_kv, h, d,
 attn_scale, dropout, qkv_layout, bias_type, attn_mask_type,
 input_Q, input_KV, input_dO,
// Aux_CTX_Tensors,
 output_S,
 output_dQ, output_dKV, output_dBias,
 input_cu_seqlens_q, input_cu_seqlens_kv,
 wkspace, stream, handle);
#else
 NVTE_ERROR(
 "cuDNN 8.9.1 is required to run BF16/FP16 fused attention with max_seqlen<=512. \n");
#endif
 } else if ((max_seqlen_q > 512) || (max_seqlen_kv > 512)) {
 NVTE_ERROR("TBD: No support for fused attention with >512 seqlence length currently. \n");
 } else if (fused_attention_backend == 3) {
 NVTE_ERROR("TODO: No support for FP16/BF16 fused attention (arbitrary seqlen) currently. \n");
 } else if (fused_attention_backend == 4) {
 NVTE_ERROR("The FP8 fused attention API only supports packed QKV input. \n");
 } else {
 NVTE_ERROR("Invalid combination of data type and sequence length! \n");
 }
Expand Down

30 changes: 18 additions & 12 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.cu

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -1360,7 +1360,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1387,12 +1388,14 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand DownExpand Up@@ -1425,7 +1428,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
// const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle) {
Expand All@@ -1451,12 +1455,14 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k

 void *devPtrdBias = output_dBias->data.dptr;

 NVTE_CHECK(Aux_CTX_Tensors->size == 1);
// NVTE_CHECK(Aux_CTX_Tensors->size == 1);
 void *devPtrS = nullptr;
 if (Aux_CTX_Tensors->size == 1) {
 Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
 devPtrS = output_S->data.dptr;
 }
// if (Aux_CTX_Tensors->size == 1) {
// Tensor *output_S = reinterpret_cast<Tensor *>(Aux_CTX_Tensors->tensors[0]);
// devPtrS = output_S->data.dptr;
// }
 devPtrS = output_S->data.dptr;

 // devPtrdS reuses the memory of devPtrS
 void *devPtrdS = devPtrS;

Expand Down

6 changes: 4 additions & 2 deletions
transformer_engine/common/fused_attn/fused_attn_fp16_bf16_max_seqlen_512.h

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -42,7 +42,8 @@ void fused_attn_max_512_bwd_qkvpacked(size_t batch, size_t max_seqlen, size_t nu
 size_t head_dim, float attn_scale, float p_dropout,
 NVTE_QKV_Layout qkv_layout, NVTE_Bias_Type bias_type,
 NVTE_Mask_Type mask_type, const Tensor *input_QKV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQKV, Tensor *output_dBias,
 const Tensor *cu_seqlens, Tensor *workspace,
 cudaStream_t stream, cudnnHandle_t handle);
Expand All@@ -52,7 +53,8 @@ void fused_attn_max_512_bwd_kvpacked(size_t batch, size_t q_max_seqlen, size_t k
 float p_dropout, NVTE_QKV_Layout qkv_layout,
 NVTE_Bias_Type bias_type, NVTE_Mask_Type mask_type,
 const Tensor *input_Q, const Tensor *input_KV,
 const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 //const Tensor *input_dO, const NVTETensorPack *Aux_CTX_Tensors,
 const Tensor *input_dO, Tensor *output_S,
 Tensor *output_dQ, Tensor *output_dKV, Tensor *output_dBias,
 const Tensor *q_cu_seqlens, const Tensor *kv_cu_seqlens,
 Tensor *workspace, cudaStream_t stream, cudnnHandle_t handle);
Expand Down
Loading