Hugging Face's logo

Kernels:
flashrt
/
flashrt-nvfp4

Kernel card Files
xet
 Community
liangsu9988 commited on
Commit
28b0e6c
·
verified ·
1 Parent(s): b9f8d14

Uploaded using `kernel-builder`.

Browse files
Files changed (16) hide show
  1. build/torch211-cxx11-cu128-x86_64-linux/__init__.py +13 -1
  2. build/torch211-cxx11-cu128-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} +1 -1
  3. build/torch211-cxx11-cu128-x86_64-linux/_ops.py +3 -3
  4. build/torch211-cxx11-cu128-x86_64-linux/metadata.json +1 -1
  5. build/torch211-cxx11-cu130-x86_64-linux/__init__.py +13 -1
  6. build/torch211-cxx11-cu130-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} +1 -1
  7. build/torch211-cxx11-cu130-x86_64-linux/_ops.py +3 -3
  8. build/torch211-cxx11-cu130-x86_64-linux/metadata.json +1 -1
  9. build/torch212-cxx11-cu130-x86_64-linux/__init__.py +13 -1
  10. build/torch212-cxx11-cu130-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} +1 -1
  11. build/torch212-cxx11-cu130-x86_64-linux/_ops.py +3 -3
  12. build/torch212-cxx11-cu130-x86_64-linux/metadata.json +1 -1
  13. build/torch212-cxx11-cu132-x86_64-linux/__init__.py +13 -1
  14. build/torch212-cxx11-cu132-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} +1 -1
  15. build/torch212-cxx11-cu132-x86_64-linux/_ops.py +3 -3
  16. build/torch212-cxx11-cu132-x86_64-linux/metadata.json +1 -1
build/torch211-cxx11-cu128-x86_64-linux/__init__.py CHANGED
@@ -6,7 +6,19 @@ from typing import Optional
6
 
7
  import torch
8
 
9
- from ._ops import ops
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
 
12
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
 
6
 
7
  import torch
8
 
9
+ from ._ops import add_op_namespace_prefix, ops
10
+
11
+
12
+ @torch.library.register_fake(add_op_namespace_prefix("nvfp4_sf_linear_to_swizzled"))
13
+ def _nvfp4_sf_linear_to_swizzled_fake(
14
+ scales: torch.Tensor,
15
+ out: torch.Tensor,
16
+ D: int,
17
+ is_sfb: bool = False,
18
+ ) -> None:
19
+ if scales.dim() != 2:
20
+ raise RuntimeError("scales must have shape (rows, D / 16)")
21
+ return None
22
 
23
 
24
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
build/torch211-cxx11-cu128-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:695400c280dd6f770e837da25ff054f0caee876890f88aff1ab9fa300aa1daf8
3
  size 95200
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a66f1b697da2b7d3b7496360c28732c4c754e84468baa48c3a03a517c51c67bb
3
  size 95200
build/torch211-cxx11-cu128-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flashrt_nvfp4_cuda_e9a1fe0
3
- ops = torch.ops._flashrt_nvfp4_cuda_e9a1fe0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flashrt_nvfp4_cuda_e9a1fe0::{op_name}"
 
1
  import torch
2
+ from . import _flashrt_nvfp4_cuda_c4d802d
3
+ ops = torch.ops._flashrt_nvfp4_cuda_c4d802d
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flashrt_nvfp4_cuda_c4d802d::{op_name}"
build/torch211-cxx11-cu128-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flashrt-nvfp4",
3
- "id": "_flashrt_nvfp4_cuda_e9a1fe0",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
 
1
  {
2
  "name": "flashrt-nvfp4",
3
+ "id": "_flashrt_nvfp4_cuda_c4d802d",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
build/torch211-cxx11-cu130-x86_64-linux/__init__.py CHANGED
@@ -6,7 +6,19 @@ from typing import Optional
6
 
7
  import torch
8
 
9
- from ._ops import ops
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
 
12
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
 
6
 
7
  import torch
8
 
9
+ from ._ops import add_op_namespace_prefix, ops
10
+
11
+
12
+ @torch.library.register_fake(add_op_namespace_prefix("nvfp4_sf_linear_to_swizzled"))
13
+ def _nvfp4_sf_linear_to_swizzled_fake(
14
+ scales: torch.Tensor,
15
+ out: torch.Tensor,
16
+ D: int,
17
+ is_sfb: bool = False,
18
+ ) -> None:
19
+ if scales.dim() != 2:
20
+ raise RuntimeError("scales must have shape (rows, D / 16)")
21
+ return None
22
 
23
 
24
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
build/torch211-cxx11-cu130-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a29dd302ad9728a9bea7899650db2a9671cd63298138658f830c6c7776243229
3
  size 100344
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d8e6c0b823b0bebe75d49f7a9bd4806378107c013e043df41ca3d587c55e913
3
  size 100344
build/torch211-cxx11-cu130-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flashrt_nvfp4_cuda_e9a1fe0
3
- ops = torch.ops._flashrt_nvfp4_cuda_e9a1fe0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flashrt_nvfp4_cuda_e9a1fe0::{op_name}"
 
1
  import torch
2
+ from . import _flashrt_nvfp4_cuda_c4d802d
3
+ ops = torch.ops._flashrt_nvfp4_cuda_c4d802d
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flashrt_nvfp4_cuda_c4d802d::{op_name}"
build/torch211-cxx11-cu130-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flashrt-nvfp4",
3
- "id": "_flashrt_nvfp4_cuda_e9a1fe0",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
 
1
  {
2
  "name": "flashrt-nvfp4",
3
+ "id": "_flashrt_nvfp4_cuda_c4d802d",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
build/torch212-cxx11-cu130-x86_64-linux/__init__.py CHANGED
@@ -6,7 +6,19 @@ from typing import Optional
6
 
7
  import torch
8
 
9
- from ._ops import ops
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
 
12
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
 
6
 
7
  import torch
8
 
9
+ from ._ops import add_op_namespace_prefix, ops
10
+
11
+
12
+ @torch.library.register_fake(add_op_namespace_prefix("nvfp4_sf_linear_to_swizzled"))
13
+ def _nvfp4_sf_linear_to_swizzled_fake(
14
+ scales: torch.Tensor,
15
+ out: torch.Tensor,
16
+ D: int,
17
+ is_sfb: bool = False,
18
+ ) -> None:
19
+ if scales.dim() != 2:
20
+ raise RuntimeError("scales must have shape (rows, D / 16)")
21
+ return None
22
 
23
 
24
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
build/torch212-cxx11-cu130-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:26f4c11dc535b3cce288d30dd0df4a6262031286fe39db48149fd25fa8165457
3
  size 111104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b0a24f8bbe4406e273fb631b33cabc777ef02ee23b23f1be618b3441bdc2ec9
3
  size 111104
build/torch212-cxx11-cu130-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flashrt_nvfp4_cuda_e9a1fe0
3
- ops = torch.ops._flashrt_nvfp4_cuda_e9a1fe0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flashrt_nvfp4_cuda_e9a1fe0::{op_name}"
 
1
  import torch
2
+ from . import _flashrt_nvfp4_cuda_c4d802d
3
+ ops = torch.ops._flashrt_nvfp4_cuda_c4d802d
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flashrt_nvfp4_cuda_c4d802d::{op_name}"
build/torch212-cxx11-cu130-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flashrt-nvfp4",
3
- "id": "_flashrt_nvfp4_cuda_e9a1fe0",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
 
1
  {
2
  "name": "flashrt-nvfp4",
3
+ "id": "_flashrt_nvfp4_cuda_c4d802d",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
build/torch212-cxx11-cu132-x86_64-linux/__init__.py CHANGED
@@ -6,7 +6,19 @@ from typing import Optional
6
 
7
  import torch
8
 
9
- from ._ops import ops
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
 
12
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
 
6
 
7
  import torch
8
 
9
+ from ._ops import add_op_namespace_prefix, ops
10
+
11
+
12
+ @torch.library.register_fake(add_op_namespace_prefix("nvfp4_sf_linear_to_swizzled"))
13
+ def _nvfp4_sf_linear_to_swizzled_fake(
14
+ scales: torch.Tensor,
15
+ out: torch.Tensor,
16
+ D: int,
17
+ is_sfb: bool = False,
18
+ ) -> None:
19
+ if scales.dim() != 2:
20
+ raise RuntimeError("scales must have shape (rows, D / 16)")
21
+ return None
22
 
23
 
24
  def nvfp4_sf_swizzled_bytes(rows: int, D: int) -> int:
build/torch212-cxx11-cu132-x86_64-linux/{_flashrt_nvfp4_cuda_e9a1fe0.abi3.so → _flashrt_nvfp4_cuda_c4d802d.abi3.so} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:23ef46609656ff3387039e83a410b131ab72c2da5c3099df19acea47c6e38d45
3
  size 111104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:91f6297b7871d47faf300ef1308b9a340666ab3afaa6b46b9a2d4d314b77255f
3
  size 111104
build/torch212-cxx11-cu132-x86_64-linux/_ops.py CHANGED
@@ -1,9 +1,9 @@
1
  import torch
2
- from . import _flashrt_nvfp4_cuda_e9a1fe0
3
- ops = torch.ops._flashrt_nvfp4_cuda_e9a1fe0
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
- return f"_flashrt_nvfp4_cuda_e9a1fe0::{op_name}"
 
1
  import torch
2
+ from . import _flashrt_nvfp4_cuda_c4d802d
3
+ ops = torch.ops._flashrt_nvfp4_cuda_c4d802d
4
 
5
  def add_op_namespace_prefix(op_name: str):
6
  """
7
  Prefix op by namespace.
8
  """
9
+ return f"_flashrt_nvfp4_cuda_c4d802d::{op_name}"
build/torch212-cxx11-cu132-x86_64-linux/metadata.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "name": "flashrt-nvfp4",
3
- "id": "_flashrt_nvfp4_cuda_e9a1fe0",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],
 
1
  {
2
  "name": "flashrt-nvfp4",
3
+ "id": "_flashrt_nvfp4_cuda_c4d802d",
4
  "version": 1,
5
  "license": "Apache-2.0",
6
  "python-depends": [],