Instructions to use replicate/paged-attention with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Kernels
How to use replicate/paged-attention with Kernels:
# !pip install kernels from kernels import get_kernel # a version (or an explicit revision) is required; see the "Files and versions" tab for the available ones kernel = get_kernel("replicate/paged-attention", version=1) - Notebooks
- Google Colab
- Kaggle
Build uploaded using `kernels`.
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- build/torch210-cxx11-cu126-x86_64-linux/__init__.py +21 -0
- build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/_custom_ops.py +0 -0
- build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/_ops.py +3 -3
- build/{torch28-cxx11-cu129-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu126-x86_64-linux/_paged_attention_61fdee0.abi3.so} +2 -2
- build/torch210-cxx11-cu126-x86_64-linux/metadata.json +1 -0
- build/torch210-cxx11-cu126-x86_64-linux/paged_attention/__init__.py +26 -0
- build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/platforms.py +0 -0
- build/torch210-cxx11-cu128-x86_64-linux/__init__.py +21 -0
- build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/_custom_ops.py +0 -0
- build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/_ops.py +3 -3
- build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu128-x86_64-linux/_paged_attention_61fdee0.abi3.so} +2 -2
- build/torch210-cxx11-cu128-x86_64-linux/metadata.json +1 -0
- build/torch210-cxx11-cu128-x86_64-linux/paged_attention/__init__.py +26 -0
- build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/platforms.py +0 -0
- build/torch210-cxx11-cu130-x86_64-linux/__init__.py +21 -0
- build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/_custom_ops.py +0 -0
- build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/_ops.py +3 -3
- build/{torch29-cxx11-cu126-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu130-x86_64-linux/_paged_attention_61fdee0.abi3.so} +2 -2
- build/torch210-cxx11-cu130-x86_64-linux/metadata.json +1 -0
- build/torch210-cxx11-cu130-x86_64-linux/paged_attention/__init__.py +26 -0
- build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/platforms.py +0 -0
- build/torch210-cxx11-rocm70-x86_64-linux/__init__.py +21 -0
- build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/_custom_ops.py +0 -0
- build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/_ops.py +3 -3
- build/{torch28-cxx11-cu128-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-rocm70-x86_64-linux/_paged_attention_61fdee0.abi3.so} +2 -2
- build/torch210-cxx11-rocm70-x86_64-linux/metadata.json +1 -0
- build/torch210-cxx11-rocm70-x86_64-linux/paged_attention/__init__.py +26 -0
- build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/platforms.py +0 -0
- build/torch210-cxx11-rocm71-x86_64-linux/__init__.py +21 -0
- build/{torch28-cxx11-rocm64-x86_64-linux/paged_attention → torch210-cxx11-rocm71-x86_64-linux}/_custom_ops.py +0 -0
- build/torch210-cxx11-rocm71-x86_64-linux/_ops.py +9 -0
- build/torch210-cxx11-rocm71-x86_64-linux/_paged_attention_61fdee0.abi3.so +3 -0
- build/torch210-cxx11-rocm71-x86_64-linux/metadata.json +1 -0
- build/torch210-cxx11-rocm71-x86_64-linux/paged_attention/__init__.py +26 -0
- build/{torch28-cxx11-rocm64-x86_64-linux/paged_attention → torch210-cxx11-rocm71-x86_64-linux}/platforms.py +0 -0
- build/torch28-cxx11-cu126-x86_64-linux/__init__.py +21 -0
- build/{torch29-cxx11-cu126-x86_64-linux/paged_attention → torch28-cxx11-cu126-x86_64-linux}/_custom_ops.py +0 -0
- build/torch28-cxx11-cu126-x86_64-linux/_ops.py +9 -0
- build/torch28-cxx11-cu126-x86_64-linux/_paged_attention_61fdee0.abi3.so +3 -0
- build/torch28-cxx11-cu126-x86_64-linux/metadata.json +1 -0
- build/torch28-cxx11-cu126-x86_64-linux/paged_attention/__init__.py +25 -20
- build/{torch29-cxx11-cu126-x86_64-linux/paged_attention → torch28-cxx11-cu126-x86_64-linux}/platforms.py +0 -0
- build/torch28-cxx11-cu128-x86_64-linux/__init__.py +21 -0
- build/{torch29-cxx11-cu128-x86_64-linux/paged_attention → torch28-cxx11-cu128-x86_64-linux}/_custom_ops.py +0 -0
- build/torch28-cxx11-cu128-x86_64-linux/_ops.py +9 -0
- build/torch28-cxx11-cu128-x86_64-linux/_paged_attention_61fdee0.abi3.so +3 -0
- build/torch28-cxx11-cu128-x86_64-linux/metadata.json +1 -0
- build/torch28-cxx11-cu128-x86_64-linux/paged_attention/__init__.py +25 -20
- build/{torch29-cxx11-cu128-x86_64-linux/paged_attention → torch28-cxx11-cu128-x86_64-linux}/platforms.py +0 -0
- build/torch28-cxx11-cu129-x86_64-linux/__init__.py +21 -0
build/torch210-cxx11-cu126-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/_ops.py
RENAMED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/{torch28-cxx11-cu129-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu126-x86_64-linux/_paged_attention_61fdee0.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:03f696c4c474048cff123b8717faee0360b50e6bd4f409a2bedf6f897f58731b
|
| 3 |
+
size 140162672
|
build/torch210-cxx11-cu126-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch210-cxx11-cu126-x86_64-linux/paged_attention/__init__.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
| 3 |
+
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch28-cxx11-cu126-x86_64-linux/paged_attention → torch210-cxx11-cu126-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch210-cxx11-cu128-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/_ops.py
RENAMED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu128-x86_64-linux/_paged_attention_61fdee0.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0fc9eab1173a3561eb41273b124f0be3410761691653ab2e15fcbbf13dddf1b2
|
| 3 |
+
size 167726072
|
build/torch210-cxx11-cu128-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch210-cxx11-cu128-x86_64-linux/paged_attention/__init__.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
| 3 |
+
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu128-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch210-cxx11-cu130-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/{torch28-cxx11-cu128-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/_ops.py
RENAMED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/{torch29-cxx11-cu126-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-cu130-x86_64-linux/_paged_attention_61fdee0.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:02541ec822b88d3a774823154414c689c258080c719887103f26f82db12a2ca6
|
| 3 |
+
size 86563760
|
build/torch210-cxx11-cu130-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch210-cxx11-cu130-x86_64-linux/paged_attention/__init__.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
| 3 |
+
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch28-cxx11-cu129-x86_64-linux/paged_attention → torch210-cxx11-cu130-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch210-cxx11-rocm70-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/_ops.py
RENAMED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/{torch28-cxx11-cu128-x86_64-linux/paged_attention/_paged_attention_287831d.abi3.so → torch210-cxx11-rocm70-x86_64-linux/_paged_attention_61fdee0.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:393f597797a88bc946476a4309c1a90bb5742e4566bdc311f04bbdd3d80648bf
|
| 3 |
+
size 118183688
|
build/torch210-cxx11-rocm70-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch210-cxx11-rocm70-x86_64-linux/paged_attention/__init__.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
| 3 |
+
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch28-cxx11-rocm63-x86_64-linux/paged_attention → torch210-cxx11-rocm70-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch210-cxx11-rocm71-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch28-cxx11-rocm64-x86_64-linux/paged_attention → torch210-cxx11-rocm71-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/torch210-cxx11-rocm71-x86_64-linux/_ops.py
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
+
|
| 5 |
+
def add_op_namespace_prefix(op_name: str):
|
| 6 |
+
"""
|
| 7 |
+
Prefix op by namespace.
|
| 8 |
+
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/torch210-cxx11-rocm71-x86_64-linux/_paged_attention_61fdee0.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f2b2140dfa254ee167124932eda2da2ad9697c55a51b2c9a069a8293067e1710
|
| 3 |
+
size 118089384
|
build/torch210-cxx11-rocm71-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch210-cxx11-rocm71-x86_64-linux/paged_attention/__init__.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
| 3 |
+
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch28-cxx11-rocm64-x86_64-linux/paged_attention → torch210-cxx11-rocm71-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch28-cxx11-cu126-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch29-cxx11-cu126-x86_64-linux/paged_attention → torch28-cxx11-cu126-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/torch28-cxx11-cu126-x86_64-linux/_ops.py
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
+
|
| 5 |
+
def add_op_namespace_prefix(op_name: str):
|
| 6 |
+
"""
|
| 7 |
+
Prefix op by namespace.
|
| 8 |
+
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/torch28-cxx11-cu126-x86_64-linux/_paged_attention_61fdee0.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:38d093ba15e819a09622c0f4f87b895bebf7cef5c9e0a24a85f8c6fc66e2e797
|
| 3 |
+
size 140143504
|
build/torch28-cxx11-cu126-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch28-cxx11-cu126-x86_64-linux/paged_attention/__init__.py
CHANGED
|
@@ -1,21 +1,26 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
copy_blocks,
|
| 4 |
-
paged_attention_v1,
|
| 5 |
-
paged_attention_v2,
|
| 6 |
-
reshape_and_cache,
|
| 7 |
-
reshape_and_cache_flash,
|
| 8 |
-
swap_blocks,
|
| 9 |
-
)
|
| 10 |
-
from ._ops import ops
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch29-cxx11-cu126-x86_64-linux/paged_attention → torch28-cxx11-cu126-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch28-cxx11-cu128-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|
build/{torch29-cxx11-cu128-x86_64-linux/paged_attention → torch28-cxx11-cu128-x86_64-linux}/_custom_ops.py
RENAMED
|
File without changes
|
build/torch28-cxx11-cu128-x86_64-linux/_ops.py
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from . import _paged_attention_61fdee0
|
| 3 |
+
ops = torch.ops._paged_attention_61fdee0
|
| 4 |
+
|
| 5 |
+
def add_op_namespace_prefix(op_name: str):
|
| 6 |
+
"""
|
| 7 |
+
Prefix op by namespace.
|
| 8 |
+
"""
|
| 9 |
+
return f"_paged_attention_61fdee0::{op_name}"
|
build/torch28-cxx11-cu128-x86_64-linux/_paged_attention_61fdee0.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:98bf417805044e0a2aecc82d925892da63df504f3fdde94017bfe4c2e334ca20
|
| 3 |
+
size 167710768
|
build/torch28-cxx11-cu128-x86_64-linux/metadata.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"python-depends":[]}
|
build/torch28-cxx11-cu128-x86_64-linux/paged_attention/__init__.py
CHANGED
|
@@ -1,21 +1,26 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
copy_blocks,
|
| 4 |
-
paged_attention_v1,
|
| 5 |
-
paged_attention_v2,
|
| 6 |
-
reshape_and_cache,
|
| 7 |
-
reshape_and_cache_flash,
|
| 8 |
-
swap_blocks,
|
| 9 |
-
)
|
| 10 |
-
from ._ops import ops
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import ctypes
|
| 2 |
+
import sys
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
+
import importlib
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from types import ModuleType
|
| 7 |
+
|
| 8 |
+
def _import_from_path(file_path: Path) -> ModuleType:
|
| 9 |
+
# We cannot use the module name as-is, after adding it to `sys.modules`,
|
| 10 |
+
# it would also be used for other imports. So, we make a module name that
|
| 11 |
+
# depends on the path for it to be unique using the hex-encoded hash of
|
| 12 |
+
# the path.
|
| 13 |
+
path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
|
| 14 |
+
module_name = path_hash
|
| 15 |
+
spec = importlib.util.spec_from_file_location(module_name, file_path)
|
| 16 |
+
if spec is None:
|
| 17 |
+
raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
|
| 18 |
+
module = importlib.util.module_from_spec(spec)
|
| 19 |
+
if module is None:
|
| 20 |
+
raise ImportError(f"Cannot load module {module_name} from spec")
|
| 21 |
+
sys.modules[module_name] = module
|
| 22 |
+
spec.loader.exec_module(module) # type: ignore
|
| 23 |
+
return module
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))
|
build/{torch29-cxx11-cu128-x86_64-linux/paged_attention → torch28-cxx11-cu128-x86_64-linux}/platforms.py
RENAMED
|
File without changes
|
build/torch28-cxx11-cu129-x86_64-linux/__init__.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ._custom_ops import (
|
| 2 |
+
convert_fp8,
|
| 3 |
+
copy_blocks,
|
| 4 |
+
paged_attention_v1,
|
| 5 |
+
paged_attention_v2,
|
| 6 |
+
reshape_and_cache,
|
| 7 |
+
reshape_and_cache_flash,
|
| 8 |
+
swap_blocks,
|
| 9 |
+
)
|
| 10 |
+
from ._ops import ops
|
| 11 |
+
|
| 12 |
+
__all__ = [
|
| 13 |
+
"convert_fp8",
|
| 14 |
+
"copy_blocks",
|
| 15 |
+
"ops",
|
| 16 |
+
"paged_attention_v1",
|
| 17 |
+
"paged_attention_v2",
|
| 18 |
+
"reshape_and_cache",
|
| 19 |
+
"reshape_and_cache_flash",
|
| 20 |
+
"swap_blocks",
|
| 21 |
+
]
|