Skip to content

[WIP] Add cudnn fused multi-head attention for JAX#105

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
zlsh80826 wants to merge 7 commits into from
Closed

[WIP] Add cudnn fused multi-head attention for JAX #105

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
aeabddc
Refine test_layer for different transpose_bs settings
zlsh80826 Mar 12, 2023
e1be3a7
Move query scale into core_attention
zlsh80826 Mar 11, 2023
075102e
Add cudnn fused multi-head attention
zlsh80826 Mar 16, 2023
f2397af
Add sharding for JAX FMHA
zlsh80826 Mar 16, 2023
35f4ae1
Add MHA fall back warning
zlsh80826 Mar 16, 2023
466c06f
Use CMake FetchContent_Declare instead of a submodule
zlsh80826 Mar 16, 2023
fd7406d
Fix fmha.h license format
zlsh80826 Mar 16, 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
536 changes: 536 additions & 0 deletions tests/jax/test_fmha.py
View file
Open in desktop

Large diffs are not rendered by default.

52 changes: 33 additions & 19 deletions tests/jax/test_layer.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_DROPOUT_RATE: 0.0,
_KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
_KEY_OF_FUSE_MLP_WI: True
}, {
_KEY_OF_TRANSPOSE_BS: False,
_KEY_OF_SCALE_ATTN_LOGITS: True,
_KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
_KEY_OF_DROPOUT_RATE: 0.0,
_KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
_KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
batch, seqlen = data_shape[:2]
if transpose_batch_sequence:
data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
sequence_dim = 0 if transpose_batch_sequence else 1

data_rng, init_rng, apply_rng = generate_test_rngs()
inputs = (jax.random.normal(data_rng, data_shape, dtype),)

batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
ref_masks = (1 - padded_mask,)
test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
else:
te_layer_attrs[k] = v
ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
sequence_dim = 0
layer_cls = partial(TransformerLayer,
hidden_dropout_dims=(sequence_dim,),
layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
del data_rng, init_rng, apply_rng

def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
batch, seqlen = data_shape[:2]
if transpose_batch_sequence:
data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
sequence_dim = 0 if transpose_batch_sequence else 1

data_rng, init_rng, apply_rng = generate_test_rngs()
inputs = (jax.random.normal(data_rng, data_shape, dtype),)

batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
ref_masks = (1 - padded_mask,)
test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
else:
te_layer_attrs[k] = v
ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
sequence_dim = 0
layer_cls = partial(TransformerLayer,
hidden_dropout_dims=(sequence_dim,),
layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
data_rng, init_rng, apply_rng = generate_test_rngs()

batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
batch, seqlen = data_shape[:2]
if transpose_batch_sequence:
data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
sequence_dim = 0 if transpose_batch_sequence else 1

data_rng, init_rng, apply_rng = generate_test_rngs()
inputs = (jax.random.normal(data_rng, data_shape,
dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
else:
te_layer_attrs[k] = v
ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
sequence_dim = 0
layer_cls = partial(TransformerLayer,
hidden_dropout_dims=(sequence_dim,),
layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
del data_rng, init_rng, apply_rng

def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
data_rng, init_rng, apply_rng = generate_test_rngs()

batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
batch, seqlen = data_shape[:2]
if transpose_batch_sequence:
data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
sequence_dim = 0 if transpose_batch_sequence else 1

data_rng, init_rng, apply_rng = generate_test_rngs()
inputs = (jax.random.normal(data_rng, data_shape,
dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
else:
te_layer_attrs[k] = v
ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
sequence_dim = 0
layer_cls = partial(TransformerLayer,
hidden_dropout_dims=(sequence_dim,),
layer_type=TransformerLayerType.DECODER,
Expand Down
18 changes: 16 additions & 2 deletions transformer_engine/common/CMakeLists.txt
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
rmsnorm/rmsnorm_fwd_cuda_kernel.cu
util/cast.cu
fused_softmax/scaled_masked_softmax.cu
fused_softmax/scaled_upper_triang_masked_softmax.cu)
fused_softmax/scaled_upper_triang_masked_softmax.cu
fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
fused_softmax/scaled_upper_triang_masked_softmax.cu
PROPERTIES
COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
cudnn-frontend
GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
PROPERTIES
INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading
, 'i'); if (__m === '*' || __re.test(location.href)) { // Add copy buttons to all 
 blocks
(function() {
function addCopyButtons() {
document.querySelectorAll('pre code').forEach(function(codeBlock) {
if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;
codeBlock.parentElement.setAttribute('data-copy-added', 'true');
var btn = document.createElement('button');
btn.textContent = 'Copy';
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;';
btn.onmouseover = function() { this.style.opacity = '1'; };
btn.onmouseout = function() { this.style.opacity = '0.7'; };
btn.onclick = function() {
navigator.clipboard.writeText(codeBlock.textContent).then(function() {
btn.textContent = 'Copied!';
setTimeout(function() { btn.textContent = 'Copy'; }, 1500);
});
};
codeBlock.parentElement.style.position = 'relative';
codeBlock.parentElement.appendChild(btn);
});
}
addCopyButtons();
// Re-run on dynamic content
var observer = new MutationObserver(addCopyButtons);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Force GitHub README to respect dark mode
(function() {
var style = document.createElement('style');
style.textContent = '
.markdown-body {
color-scheme: dark light;
}
.markdown-body pre { background: #161b22 !important; }
.markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }
.markdown-body table th, .markdown-body table td { border-color: #30363d !important; }
.markdown-body img { background: #0d1117; }
.markdown-body blockquote { border-left-color: #8b949e; }
.markdown-body hr { border-color: #30363d; }
';
document.head.appendChild(style);
})();
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Highlight search terms from Google/DuckDuckGo/Bing referrer
(function() {
var ref = document.referrer;
var terms = [];
if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {
var url = new URL(ref);
var q = url.searchParams.get('q') || url.searchParams.get('p');
if (q) {
terms = q.split(/\s+/).filter(function(t) { return t.length > 2; });
}
}
if (terms.length === 0) return;
var style = document.createElement('style');
style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';
document.head.appendChild(style);
function highlight(node) {
if (node.nodeType === 3) { // text node
var text = node.textContent;
var found = false;
terms.forEach(function(term) {
var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\]\\]/g, '\\') + ')', 'gi');
if (regex.test(text)) {
found = true;
var frag = document.createDocumentFragment();
var parts = text.split(regex);
parts.forEach(function(part, i) {
if (i % 2 === 0) {
frag.appendChild(document.createTextNode(part));
} else {
var span = document.createElement('span');
span.className = 'userscript-highlight';
span.textContent = part;
frag.appendChild(span);
}
});
node.parentNode.replaceChild(frag, node);
}
});
} else if (node.nodeType === 1 && node.childNodes) { // element
var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];
if (!skipTags.includes(node.tagName)) {
Array.from(node.childNodes).forEach(highlight);
}
}
}
highlight(document.body);
// Re-highlight on dynamic content
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1 || node.nodeType === 3) highlight(node);
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Strip utm_, fbclid, gclid, etc. from all links on page
(function() {
var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',
'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',
'ref', 'ref_src', 'source', 'medium', 'campaign'];
function cleanUrl(url) {
try {
var u = new URL(url, window.location.origin);
var changed = false;
trackingParams.forEach(function(p) {
if (u.searchParams.has(p)) {
u.searchParams.delete(p);
changed = true;
}
});
return changed ? u.toString() : url;
} catch (e) {
return url;
}
}
function cleanLinks() {
document.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
cleanLinks();
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1) {
if (node.tagName === 'A') cleanLinks();
node.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Auto-enable theater mode on YouTube
(function() {
function tryTheater() {
var btn = document.querySelector('button[aria-label="Theater mode"], ytd-player #player button[title="Theater mode"]');
if (btn && !btn.classList.contains('activated')) {
btn.click();
}
}
// Try immediately
tryTheater();
// Try after navigation (SPA)
var lastUrl = location.href;
setInterval(function() {
if (location.href !== lastUrl) {
lastUrl = location.href;
setTimeout(tryTheater, 500);
}
}, 1000);
// Also try on player load
var observer = new MutationObserver(tryTheater);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Remove or un-stick sticky/fixed headers that block content
(function() {
function unstick() {
document.querySelectorAll('header, nav, [role="banner"], .header, .navbar, .sticky, .fixed-top, [style*="position: fixed"], [style*="position:sticky"]').forEach(function(el) {
if (el.style.position === 'fixed' || el.style.position === 'sticky' || getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {
el.style.position = 'static';
el.style.top = 'auto';
el.style.zIndex = 'auto';
}
});
}
unstick();
var observer = new MutationObserver(unstick);
observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });
})();
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Universal Dark Mode - works on any site
(function() {
var enabled = true;
function applyDarkMode() {
if (!enabled) return;
// Create style element if it doesn't exist
var style = document.getElementById('universal-dark-mode-style');
if (!style) {
style = document.createElement('style');
style.id = 'universal-dark-mode-style';
document.head.appendChild(style);
}
// Dark mode CSS - inverts colors but preserves images/video
style.textContent = '
/* Invert everything except media */
html {
filter: invert(1) hue-rotate(180deg) !important;
background: #1a1a2e !important;
}
/* Restore images, videos, iframes, canvas */
img, video, iframe, canvas, svg, picture, [style*="background-image"] {
filter: invert(1) hue-rotate(180deg) !important;
}
/* Preserve specific elements that should not be inverted */
.no-dark-mode, .no-dark-mode *,
[data-theme="light"], [data-theme="light"],
.ace_editor, .ace_editor *,
.CodeMirror, .CodeMirror *,
.monaco-editor, .monaco-editor *,
.markdown-body pre, .markdown-body pre *,
.highlight, .highlight *,
pre code, pre code * {
filter: none !important;
}
/* Fix common UI elements */
.modal, .popup, .dropdown-menu, .tooltip, .popover {
filter: invert(1) hue-rotate(180deg) !important;
background: #2d2d44 !important;
border-color: #444 !important;
}
/* Scrollbars */
::-webkit-scrollbar { background: #1a1a2e !important; }
::-webkit-scrollbar-thumb { background: #444 !important; }
::-webkit-scrollbar-thumb:hover { background: #555 !important; }
/* Selection */
::selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
::-moz-selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
';
}
function removeDarkMode() {
var style = document.getElementById('universal-dark-mode-style');
if (style) style.remove();
}
// Toggle with Alt+Shift+D
document.addEventListener('keydown', function(e) {
if (e.altKey && e.shiftKey && e.key === 'D') {
e.preventDefault();
enabled = !enabled;
if (enabled) {
applyDarkMode();
console.log('[Universal Dark Mode] Enabled');
} else {
removeDarkMode();
console.log('[Universal Dark Mode] Disabled');
}
}
});
// Apply on load
applyDarkMode();
// Re-apply on dynamic content
var observer = new MutationObserver(function(mutations) {
if (enabled && !document.getElementById('universal-dark-mode-style')) {
applyDarkMode();
}
});
observer.observe(document.head, { childList: true });
console.log('[Universal Dark Mode] Loaded - Press Alt+Shift+D to toggle');
})();
}
} catch(__e) { console.warn('[Userscript:Universal Dark Mode]', __e); }
})();
})();
[WIP] Add cudnn fused multi-head attention for JAX by zlsh80826 · Pull Request #105 · NVIDIA/TransformerEngine · GitHub
Skip to content
[WIP] Add cudnn fused multi-head attention for JAX#105

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

zlsh80826
wants to merge
7
commits into

from

 Closed


[WIP] Add cudnn fused multi-head attention for JAX
#105
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


7 commits


Select commit
Hold shift + click to select a range
aeabddc

Refine test_layer for different transpose_bs settings


zlsh80826 Mar 12, 2023
e1be3a7

Move query scale into core_attention


zlsh80826 Mar 11, 2023
075102e

Add cudnn fused multi-head attention


zlsh80826 Mar 16, 2023
f2397af

Add sharding for JAX FMHA


zlsh80826 Mar 16, 2023
35f4ae1

Add MHA fall back warning


zlsh80826 Mar 16, 2023
466c06f

Use CMake FetchContent_Declare instead of a submodule


zlsh80826 Mar 16, 2023
fd7406d

Fix fmha.h license format


zlsh80826 Mar 16, 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


536 changes: 536 additions & 0 deletions
tests/jax/test_fmha.py

View file


Open in desktop


Large diffs are not rendered by default.


52 changes: 33 additions & 19 deletions
tests/jax/test_layer.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -54,7 +54,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 err_msg=f"{key=} is not close")


DATA_SHAPE = [(128, 32, 512), (512, 32, 512)] # (seqlen, batch, emb_dim)
DATA_SHAPE = [(32, 128, 1024), (32, 512, 1024)] # (batch, seqlen, emb_dim)
DTYPE = [jnp.float32, jnp.bfloat16]
FP8_FORMATS = [Format.E4M3, Format.HYBRID]

Expand All@@ -67,6 +67,7 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
_KEY_OF_FUSE_MLP_WI = "fuse_mlp_wi"
_KEY_OF_LAYERNORM_TYPE = 'layernorm_type'
_KEY_OF_TRANSPOSE_BS = 'transpose_batch_sequence'
_KEY_OF_SCALE_ATTN_LOGITS = "scale_attn_logits"

BASE_ATTRS = {_KEY_OF_TRANSPOSE_BS: True}

Expand DownExpand Up@@ -95,6 +96,13 @@ def compare_frozen_dict(ref_fd, test_fd, rtol=1e-05, atol=1e-08):
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}, {
 _KEY_OF_TRANSPOSE_BS: False,
 _KEY_OF_SCALE_ATTN_LOGITS: True,
 _KEY_OF_LAYERNORM_TYPE: 'rmsnorm',
 _KEY_OF_DROPOUT_RATE: 0.0,
 _KEY_OF_MLP_ACTIVATIONS: (('gelu', 'linear')),
 _KEY_OF_FUSE_MLP_WI: True
}]

ATTRS = [{**BASE_ATTRS, **attr} for attr in ATTRS]
Expand DownExpand Up@@ -125,12 +133,15 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -145,7 +156,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand All@@ -167,12 +177,15 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape, dtype),)

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]

 padded_mask = jnp.zeros((batch, 1, seqlen, seqlen), dtype=jnp.uint8)
 ref_masks = (1 - padded_mask,)
 test_masks = (None, padded_mask) # The second arg of Transformer is encoded tokens.
Expand All@@ -187,7 +200,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefEncoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.ENCODER,
Expand DownExpand Up@@ -331,11 +343,13 @@ def sync_params(ref, target, attrs):
 return ref, flax.core.frozen_dict.FrozenDict(unfreeze_target)

 def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -354,7 +368,6 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand All@@ -375,11 +388,13 @@ def forward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 del data_rng, init_rng, apply_rng

 def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-08):
 data_rng, init_rng, apply_rng = generate_test_rngs()

 batch = data_shape[1] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[0]
 seqlen = data_shape[0] if _KEY_OF_TRANSPOSE_BS in attrs else data_shape[1]
 transpose_batch_sequence = _KEY_OF_TRANSPOSE_BS in attrs and attrs[_KEY_OF_TRANSPOSE_BS]
 batch, seqlen = data_shape[:2]
 if transpose_batch_sequence:
 data_shape = (data_shape[1], data_shape[0], *data_shape[2:])
 sequence_dim = 0 if transpose_batch_sequence else 1

 data_rng, init_rng, apply_rng = generate_test_rngs()
 inputs = (jax.random.normal(data_rng, data_shape,
 dtype), jax.random.normal(data_rng, data_shape, dtype))

Expand All@@ -398,7 +413,6 @@ def forward_backward_runner(self, data_shape, dtype, attrs, rtol=1e-05, atol=1e-
 else:
 te_layer_attrs[k] = v
 ref_layer_cls = partial(RefDecoderLayer, dtype=dtype, **attrs)
 sequence_dim = 0
 layer_cls = partial(TransformerLayer,
 hidden_dropout_dims=(sequence_dim,),
 layer_type=TransformerLayerType.DECODER,
Expand Down

18 changes: 16 additions & 2 deletions
transformer_engine/common/CMakeLists.txt

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -17,11 +17,12 @@ add_library(transformer_engine SHARED
 rmsnorm/rmsnorm_fwd_cuda_kernel.cu
 util/cast.cu
 fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu)
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 fused_mha/cudnn_fmha.cu)

target_include_directories(transformer_engine PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}/include")

list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt)
list(APPEND transformer_engine_LINKER_LIBS CUDA::cublas CUDA::cudart CUDA::nvToolsExt cudnn)
target_link_libraries(transformer_engine PUBLIC ${transformer_engine_LINKER_LIBS})

target_include_directories(transformer_engine PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
Expand All@@ -30,5 +31,18 @@ set_source_files_properties(fused_softmax/scaled_masked_softmax.cu
 fused_softmax/scaled_upper_triang_masked_softmax.cu
 PROPERTIES
 COMPILE_OPTIONS "--use_fast_math")

include(FetchContent)
FetchContent_Declare(
 cudnn-frontend
 GIT_REPOSITORY https://github.com/NVIDIA/cudnn-frontend.git
 GIT_TAG 8f488bd41229aa0a3d5f7c0168f59e4d69c618ee # release v0.8
)
FetchContent_Populate(cudnn-frontend)

set_source_files_properties(fused_mha/cudnn_fmha.cu
 PROPERTIES
 INCLUDE_DIRECTORIES "${cudnn-frontend_SOURCE_DIR}/include")

set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
Loading