Hugging Face's logo

Kernels:
kernels-community
/
flash-attn2
Trusted publisher

Kernel card Files
xet
 Community
kernels-bot commited on
Commit
c5b9b7c
·
verified ·
1 Parent(s): 4d869f3

Uploaded using `kernel-builder`.

Browse files
Files changed (12) hide show
  1. build/torch210-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} +2 -2
  2. build/torch210-cxx11-xpu20253-x86_64-linux/_ops.py +3 -3
  3. build/torch210-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py +1 -8
  4. build/torch210-cxx11-xpu20253-x86_64-linux/metadata.json +1 -1
  5. build/torch211-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} +2 -2
  6. build/torch211-cxx11-xpu20253-x86_64-linux/_ops.py +3 -3
  7. build/torch211-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py +1 -8
  8. build/torch211-cxx11-xpu20253-x86_64-linux/metadata.json +1 -1
  9. build/torch212-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} +2 -2
  10. build/torch212-cxx11-xpu20253-x86_64-linux/_ops.py +3 -3
  11. build/torch212-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py +1 -8
  12. build/torch212-cxx11-xpu20253-x86_64-linux/metadata.json +1 -1
build/torch210-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4aebc32fcc23d3e214267b829b34593617c300865c7ee84388b62820bcc96c59
3
- size 37243424
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15a9d3044a02aff0b2f3c240b62c7eb54124878c480717102384dd89090e77ae
3
+ size 17713168
build/torch210-cxx11-xpu20253-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flash_attn2_xpu_14f6010
3
- ops = torch.ops._flash_attn2_xpu_14f6010
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flash_attn2_xpu_14f6010::{op_name}"
 
1
  import torch
2
+ from . import _flash_attn2_xpu_85c21a0
3
+ ops = torch.ops._flash_attn2_xpu_85c21a0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flash_attn2_xpu_85c21a0::{op_name}"
build/torch210-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py CHANGED
@@ -36,14 +36,7 @@ def _get_block_size_n(device, head_dim, is_dropout, is_causal):
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
- if head_dim <= 96:
40
- return 64
41
- elif head_dim <= 128:
42
- return 32
43
- elif head_dim <= 256:
44
- return 64
45
- else:
46
- return 32
47
 
48
  # This should match the block sizes in the CUDA kernel
49
  major, minor = torch.cuda.get_device_capability(device)
 
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
+ return 64
 
 
 
 
 
 
 
40
 
41
  # This should match the block sizes in the CUDA kernel
42
  major, minor = torch.cuda.get_device_capability(device)
build/torch210-cxx11-xpu20253-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flash-attn2",
3
- "id": "_flash_attn2_xpu_14f6010",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],
 
1
  {
2
  "name": "flash-attn2",
3
+ "id": "_flash_attn2_xpu_85c21a0",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],
build/torch211-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:58e4af382503705b7154ce52ea48013f45f9bd1558f4469b96e17a25878b6ab4
3
- size 37243424
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54c1c44285ae6adbd98d2eeaaa1f475dd74170c5d5676091ac26bf01091a0da5
3
+ size 17713168
build/torch211-cxx11-xpu20253-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flash_attn2_xpu_14f6010
3
- ops = torch.ops._flash_attn2_xpu_14f6010
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flash_attn2_xpu_14f6010::{op_name}"
 
1
  import torch
2
+ from . import _flash_attn2_xpu_85c21a0
3
+ ops = torch.ops._flash_attn2_xpu_85c21a0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flash_attn2_xpu_85c21a0::{op_name}"
build/torch211-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py CHANGED
@@ -36,14 +36,7 @@ def _get_block_size_n(device, head_dim, is_dropout, is_causal):
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
- if head_dim <= 96:
40
- return 64
41
- elif head_dim <= 128:
42
- return 32
43
- elif head_dim <= 256:
44
- return 64
45
- else:
46
- return 32
47
 
48
  # This should match the block sizes in the CUDA kernel
49
  major, minor = torch.cuda.get_device_capability(device)
 
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
+ return 64
 
 
 
 
 
 
 
40
 
41
  # This should match the block sizes in the CUDA kernel
42
  major, minor = torch.cuda.get_device_capability(device)
build/torch211-cxx11-xpu20253-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flash-attn2",
3
- "id": "_flash_attn2_xpu_14f6010",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],
 
1
  {
2
  "name": "flash-attn2",
3
+ "id": "_flash_attn2_xpu_85c21a0",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],
build/torch212-cxx11-xpu20253-x86_64-linux/{_flash_attn2_xpu_14f6010.abi3.so → _flash_attn2_xpu_85c21a0.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:66c1acbc81d29095817f87ce78459026c256d33b40b54ca3416e5a24d7df9b8b
3
- size 37198312
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d32013ee9eba7775b2ce44c94d6059798f6c56cf7e9b32b5c7f7684aaedd99ff
3
+ size 17672144
build/torch212-cxx11-xpu20253-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flash_attn2_xpu_14f6010
3
- ops = torch.ops._flash_attn2_xpu_14f6010
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flash_attn2_xpu_14f6010::{op_name}"
 
1
  import torch
2
+ from . import _flash_attn2_xpu_85c21a0
3
+ ops = torch.ops._flash_attn2_xpu_85c21a0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flash_attn2_xpu_85c21a0::{op_name}"
build/torch212-cxx11-xpu20253-x86_64-linux/flash_attn_interface.py CHANGED
@@ -36,14 +36,7 @@ def _get_block_size_n(device, head_dim, is_dropout, is_causal):
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
- if head_dim <= 96:
40
- return 64
41
- elif head_dim <= 128:
42
- return 32
43
- elif head_dim <= 256:
44
- return 64
45
- else:
46
- return 32
47
 
48
  # This should match the block sizes in the CUDA kernel
49
  major, minor = torch.cuda.get_device_capability(device)
 
36
  assert head_dim <= 256
37
 
38
  if device.type == "xpu":
39
+ return 64
 
 
 
 
 
 
 
40
 
41
  # This should match the block sizes in the CUDA kernel
42
  major, minor = torch.cuda.get_device_capability(device)
build/torch212-cxx11-xpu20253-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flash-attn2",
3
- "id": "_flash_attn2_xpu_14f6010",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],
 
1
  {
2
  "name": "flash-attn2",
3
+ "id": "_flash_attn2_xpu_85c21a0",
4
  "version": 1,
5
  "license": "BSD-3-Clause",
6
  "python-depends": [],