kernels-bot commited on 6 days ago

Commit

43a023e

verified ·

1 Parent(s): 41ffaab

Uploaded using `kernel-builder`.

Browse files

Files changed (32) hide show

build/torch210-cxx11-cu126-x86_64-linux/_ops.py +3 -3
build/torch210-cxx11-cu126-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} +1 -1
build/torch210-cxx11-cu128-x86_64-linux/_ops.py +3 -3
build/torch210-cxx11-cu128-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} +1 -1
build/torch210-cxx11-cu130-x86_64-linux/_ops.py +3 -3
build/{torch211-cxx11-cu130-x86_64-linux/_paged_attention_cuda_83cf4a3.abi3.so → torch210-cxx11-cu130-x86_64-linux/_paged_attention_cuda_0278991.abi3.so} +2 -2
build/torch210-cxx11-rocm70-x86_64-linux/_ops.py +3 -3
build/torch210-cxx11-rocm70-x86_64-linux/{_paged_attention_rocm_83cf4a3.abi3.so → _paged_attention_rocm_0278991.abi3.so} +1 -1
build/torch210-cxx11-rocm71-x86_64-linux/_ops.py +3 -3
build/torch210-cxx11-rocm71-x86_64-linux/{_paged_attention_rocm_83cf4a3.abi3.so → _paged_attention_rocm_0278991.abi3.so} +1 -1
build/torch211-cxx11-cu126-x86_64-linux/_ops.py +3 -3
build/torch211-cxx11-cu126-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} +1 -1
build/torch211-cxx11-cu128-x86_64-linux/_ops.py +3 -3
build/torch211-cxx11-cu128-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} +1 -1
build/torch211-cxx11-cu130-x86_64-linux/_ops.py +3 -3
build/{torch210-cxx11-cu130-x86_64-linux/_paged_attention_cuda_83cf4a3.abi3.so → torch211-cxx11-cu130-x86_64-linux/_paged_attention_cuda_0278991.abi3.so} +2 -2
build/torch211-cxx11-rocm71-x86_64-linux/__init__.py +21 -0
build/torch211-cxx11-rocm71-x86_64-linux/_custom_ops.py +173 -0
build/torch211-cxx11-rocm71-x86_64-linux/_ops.py +9 -0
build/torch211-cxx11-rocm71-x86_64-linux/_paged_attention_rocm_0278991.abi3.so +3 -0
build/torch211-cxx11-rocm71-x86_64-linux/metadata.json +17 -0
build/torch211-cxx11-rocm71-x86_64-linux/paged_attention/__init__.py +26 -0
build/torch211-cxx11-rocm71-x86_64-linux/platforms.py +92 -0
build/torch211-cxx11-rocm72-x86_64-linux/__init__.py +21 -0
build/torch211-cxx11-rocm72-x86_64-linux/_custom_ops.py +173 -0
build/torch211-cxx11-rocm72-x86_64-linux/_ops.py +9 -0
build/torch211-cxx11-rocm72-x86_64-linux/_paged_attention_rocm_0278991.abi3.so +3 -0
build/torch211-cxx11-rocm72-x86_64-linux/metadata.json +17 -0
build/torch211-cxx11-rocm72-x86_64-linux/paged_attention/__init__.py +26 -0
build/torch211-cxx11-rocm72-x86_64-linux/platforms.py +92 -0
build/torch29-cxx11-cu129-x86_64-linux/_ops.py +3 -3
build/torch29-cxx11-cu129-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} +1 -1

build/torch210-cxx11-cu126-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/torch210-cxx11-cu126-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f84331fb1023844b101c03c8f12818bb3b09c273a9442b631cc2efe87b1eee2f
 size 140162704

 version https://git-lfs.github.com/spec/v1
+oid sha256:df019ba5a2781d6fd2a8edfe24f300f223b6eeb270e7fc2b995ea4488d11605e
 size 140162704

build/torch210-cxx11-cu128-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/torch210-cxx11-cu128-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37c0783a4a3628ffc43d64b65090cb4fa8b2f5cc2fe913a51901378f518d11af
 size 167726096

 version https://git-lfs.github.com/spec/v1
+oid sha256:323329653b95e9a5bdd56e1072b86c646b8391914ad98a79975e92a83e634b0e
 size 167726096

build/torch210-cxx11-cu130-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/{torch211-cxx11-cu130-x86_64-linux/_paged_attention_cuda_83cf4a3.abi3.so → torch210-cxx11-cu130-x86_64-linux/_paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df2b4931f36d5314b23cf3dd3a718665c9c84ad055ca2d844adb8e34be4aac24
-size 86548312

 version https://git-lfs.github.com/spec/v1
+oid sha256:e55cb05e3e5a9459da7e71d31a13c6d51b7d130b1a2f63337067d56dc32556e4
+size 86560024

build/torch210-cxx11-rocm70-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_rocm_83cf4a3
-ops = torch.ops._paged_attention_rocm_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_rocm_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_rocm_0278991
+ops = torch.ops._paged_attention_rocm_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_rocm_0278991::{op_name}"

build/torch210-cxx11-rocm70-x86_64-linux/{_paged_attention_rocm_83cf4a3.abi3.so → _paged_attention_rocm_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c715078de15626c6dc53b2bb321828478a33952ed5bac5e6f5730a984445b321
 size 58992416

 version https://git-lfs.github.com/spec/v1
+oid sha256:661e8cfb7f49a9bb0688311059de91a4ac6eae2c555544ce61ea097e509db3ee
 size 58992416

build/torch210-cxx11-rocm71-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_rocm_83cf4a3
-ops = torch.ops._paged_attention_rocm_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_rocm_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_rocm_0278991
+ops = torch.ops._paged_attention_rocm_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_rocm_0278991::{op_name}"

build/torch210-cxx11-rocm71-x86_64-linux/{_paged_attention_rocm_83cf4a3.abi3.so → _paged_attention_rocm_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e531858b1c7c996812b84a6801d240f9fa650ccc078f74b706d72414bae7965
 size 58971840

 version https://git-lfs.github.com/spec/v1
+oid sha256:01804c0edeec54fb2ca559d143d6a4cdc8f9327688e0cbb5903bd86378d480e4
 size 58971840

build/torch211-cxx11-cu126-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/torch211-cxx11-cu126-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b092432ec63eed6ffba55b648a0d484c3dfe58b6b712cea9dfb527dc38066dc
 size 140147224

 version https://git-lfs.github.com/spec/v1
+oid sha256:7b8716b3bcf741dff62a5b2b98f3e2be5e2ff3c5a0bbe553cd86a62902629ab5
 size 140147224

build/torch211-cxx11-cu128-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/torch211-cxx11-cu128-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:669f48c97e2970d820919e972762bf0cc64bc657a53a6b7452de5dc9fd1e4844
 size 167710616

 version https://git-lfs.github.com/spec/v1
+oid sha256:5fd9534a941066a044ce1512488aacae199f4e2beb8708f438065aaf45440f48
 size 167710616

build/torch211-cxx11-cu130-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/{torch210-cxx11-cu130-x86_64-linux/_paged_attention_cuda_83cf4a3.abi3.so → torch211-cxx11-cu130-x86_64-linux/_paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fc05b440e24ece432bd009e23dbf721d191d03cfa3d020c2d52d3eaface9992
-size 86563792

 version https://git-lfs.github.com/spec/v1
+oid sha256:94bc376712fefef06a7ba4743872bceed08a1451eacc8c27c70afd5fc7132a3b
+size 86548640

build/torch211-cxx11-rocm71-x86_64-linux/__init__.py ADDED Viewed

	@@ -0,0 +1,21 @@

+from ._custom_ops import (
+    convert_fp8,
+    copy_blocks,
+    paged_attention_v1,
+    paged_attention_v2,
+    reshape_and_cache,
+    reshape_and_cache_flash,
+    swap_blocks,
+)
+from ._ops import ops
+__all__ = [
+    "convert_fp8",
+    "copy_blocks",
+    "ops",
+    "paged_attention_v1",
+    "paged_attention_v2",
+    "reshape_and_cache",
+    "reshape_and_cache_flash",
+    "swap_blocks",
+]

build/torch211-cxx11-rocm71-x86_64-linux/_custom_ops.py ADDED Viewed

	@@ -0,0 +1,173 @@

+from typing import List, Optional
+import torch
+from ._ops import ops
+# page attention ops
+def paged_attention_v1(
+    out: torch.Tensor,
+    query: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    num_kv_heads: int,
+    scale: float,
+    block_tables: torch.Tensor,
+    seq_lens: torch.Tensor,
+    block_size: int,
+    max_seq_len: int,
+    alibi_slopes: Optional[torch.Tensor],
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+    tp_rank: int = 0,
+    blocksparse_local_blocks: int = 0,
+    blocksparse_vert_stride: int = 0,
+    blocksparse_block_size: int = 64,
+    blocksparse_head_sliding_step: int = 0,
+) -> None:
+    ops.paged_attention_v1(
+        out,
+        query,
+        key_cache,
+        value_cache,
+        num_kv_heads,
+        scale,
+        block_tables,
+        seq_lens,
+        block_size,
+        max_seq_len,
+        alibi_slopes,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+        tp_rank,
+        blocksparse_local_blocks,
+        blocksparse_vert_stride,
+        blocksparse_block_size,
+        blocksparse_head_sliding_step,
+    )
+def paged_attention_v2(
+    out: torch.Tensor,
+    exp_sum: torch.Tensor,
+    max_logits: torch.Tensor,
+    tmp_out: torch.Tensor,
+    query: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    num_kv_heads: int,
+    scale: float,
+    block_tables: torch.Tensor,
+    seq_lens: torch.Tensor,
+    block_size: int,
+    max_seq_len: int,
+    alibi_slopes: Optional[torch.Tensor],
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+    tp_rank: int = 0,
+    blocksparse_local_blocks: int = 0,
+    blocksparse_vert_stride: int = 0,
+    blocksparse_block_size: int = 64,
+    blocksparse_head_sliding_step: int = 0,
+) -> None:
+    ops.paged_attention_v2(
+        out,
+        exp_sum,
+        max_logits,
+        tmp_out,
+        query,
+        key_cache,
+        value_cache,
+        num_kv_heads,
+        scale,
+        block_tables,
+        seq_lens,
+        block_size,
+        max_seq_len,
+        alibi_slopes,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+        tp_rank,
+        blocksparse_local_blocks,
+        blocksparse_vert_stride,
+        blocksparse_block_size,
+        blocksparse_head_sliding_step,
+    )
+def reshape_and_cache(
+    key: torch.Tensor,
+    value: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    slot_mapping: torch.Tensor,
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+) -> None:
+    ops.reshape_and_cache(
+        key,
+        value,
+        key_cache,
+        value_cache,
+        slot_mapping,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+    )
+def reshape_and_cache_flash(
+    key: torch.Tensor,
+    value: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    slot_mapping: torch.Tensor,
+    kv_cache_dtype: str,
+    k_scale: torch.Tensor,
+    v_scale: torch.Tensor,
+) -> None:
+    ops.reshape_and_cache_flash(
+        key,
+        value,
+        key_cache,
+        value_cache,
+        slot_mapping,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+    )
+def copy_blocks(
+    key_caches: List[torch.Tensor],
+    value_caches: List[torch.Tensor],
+    block_mapping: torch.Tensor,
+) -> None:
+    ops.copy_blocks(key_caches, value_caches, block_mapping)
+def swap_blocks(
+    src: torch.Tensor, dst: torch.Tensor, block_mapping: torch.Tensor
+) -> None:
+    ops.swap_blocks(src, dst, block_mapping)
+def convert_fp8(
+    output: torch.Tensor, input: torch.Tensor, scale: float = 1.0, kv_dtype: str = "fp8"
+) -> None:
+    ops.convert_fp8(output, input, scale, kv_dtype)
+__all__ = [
+    "convert_fp8",
+    "paged_attention_v1",
+    "paged_attention_v2",
+    "reshape_and_cache",
+    "copy_blocks",
+]

build/torch211-cxx11-rocm71-x86_64-linux/_ops.py ADDED Viewed

	@@ -0,0 +1,9 @@

+import torch
+from . import _paged_attention_rocm_0278991
+ops = torch.ops._paged_attention_rocm_0278991
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_paged_attention_rocm_0278991::{op_name}"

build/torch211-cxx11-rocm71-x86_64-linux/_paged_attention_rocm_0278991.abi3.so ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:03d841758b3d039ab8702d4a13bb3c52772c8116f452904f11fdbc9e7c07afa0
+size 58964168

build/torch211-cxx11-rocm71-x86_64-linux/metadata.json ADDED Viewed

	@@ -0,0 +1,17 @@

+{
+  "version": 1,
+  "license": "Apache-2.0",
+  "python-depends": [],
+  "backend": {
+    "type": "rocm",
+    "archs": [
+      "gfx1030",
+      "gfx1100",
+      "gfx1101",
+      "gfx906",
+      "gfx908",
+      "gfx90a",
+      "gfx942"
+    ]
+  }
+}

build/torch211-cxx11-rocm71-x86_64-linux/paged_attention/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+import ctypes
+import importlib.util
+import sys
+from pathlib import Path
+from types import ModuleType
+def _import_from_path(file_path: Path) -> ModuleType:
+    # We cannot use the module name as-is, after adding it to `sys.modules`,
+    # it would also be used for other imports. So, we make a module name that
+    # depends on the path for it to be unique using the hex-encoded hash of
+    # the path.
+    path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
+    module_name = path_hash
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None:
+        raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    if module is None:
+        raise ImportError(f"Cannot load module {module_name} from spec")
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)  # type: ignore
+    return module
+globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))

build/torch211-cxx11-rocm71-x86_64-linux/platforms.py ADDED Viewed

	@@ -0,0 +1,92 @@

+import os
+import random
+from abc import ABC, abstractmethod
+from functools import lru_cache, wraps
+from typing import Callable, ParamSpec, TypeVar
+import numpy as np
+import torch
+IS_ROCM = torch.version.hip is not None
+IS_MPS = torch.backends.mps.is_available()
+class Platform(ABC):
+    @classmethod
+    def seed_everything(cls, seed: int) -> None:
+        """
+        Set the seed of each random module.
+        `torch.manual_seed` will set seed on all devices.
+        Loosely based on: https://github.com/Lightning-AI/pytorch-lightning/blob/2.4.0/src/lightning/fabric/utilities/seed.py#L20
+        """
+        random.seed(seed)
+        np.random.seed(seed)
+        torch.manual_seed(seed)
+    @abstractmethod
+    def get_device_name(self, device_id: int = 0) -> str: ...
+    @abstractmethod
+    def is_cuda(self) -> bool: ...
+    @abstractmethod
+    def is_rocm(self) -> bool: ...
+    @abstractmethod
+    def is_mps(self) -> bool: ...
+class CudaPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(0)
+    def is_cuda(self) -> bool:
+        return True
+    def is_rocm(self) -> bool:
+        return False
+    def is_mps(self) -> bool:
+        return False
+class RocmPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(device_id)
+    def is_cuda(self) -> bool:
+        return False
+    def is_rocm(self) -> bool:
+        return True
+    def is_mps(self) -> bool:
+        return False
+class MpsPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(device_id)
+    def is_cuda(self) -> bool:
+        return False
+    def is_rocm(self) -> bool:
+        return False
+    def is_mps(self) -> bool:
+        return True
+current_platform = (
+    RocmPlatform() if IS_ROCM else
+    MpsPlatform() if IS_MPS else
+    CudaPlatform() if torch.cuda.is_available() else
+    None
+)

build/torch211-cxx11-rocm72-x86_64-linux/__init__.py ADDED Viewed

	@@ -0,0 +1,21 @@

+from ._custom_ops import (
+    convert_fp8,
+    copy_blocks,
+    paged_attention_v1,
+    paged_attention_v2,
+    reshape_and_cache,
+    reshape_and_cache_flash,
+    swap_blocks,
+)
+from ._ops import ops
+__all__ = [
+    "convert_fp8",
+    "copy_blocks",
+    "ops",
+    "paged_attention_v1",
+    "paged_attention_v2",
+    "reshape_and_cache",
+    "reshape_and_cache_flash",
+    "swap_blocks",
+]

build/torch211-cxx11-rocm72-x86_64-linux/_custom_ops.py ADDED Viewed

	@@ -0,0 +1,173 @@

+from typing import List, Optional
+import torch
+from ._ops import ops
+# page attention ops
+def paged_attention_v1(
+    out: torch.Tensor,
+    query: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    num_kv_heads: int,
+    scale: float,
+    block_tables: torch.Tensor,
+    seq_lens: torch.Tensor,
+    block_size: int,
+    max_seq_len: int,
+    alibi_slopes: Optional[torch.Tensor],
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+    tp_rank: int = 0,
+    blocksparse_local_blocks: int = 0,
+    blocksparse_vert_stride: int = 0,
+    blocksparse_block_size: int = 64,
+    blocksparse_head_sliding_step: int = 0,
+) -> None:
+    ops.paged_attention_v1(
+        out,
+        query,
+        key_cache,
+        value_cache,
+        num_kv_heads,
+        scale,
+        block_tables,
+        seq_lens,
+        block_size,
+        max_seq_len,
+        alibi_slopes,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+        tp_rank,
+        blocksparse_local_blocks,
+        blocksparse_vert_stride,
+        blocksparse_block_size,
+        blocksparse_head_sliding_step,
+    )
+def paged_attention_v2(
+    out: torch.Tensor,
+    exp_sum: torch.Tensor,
+    max_logits: torch.Tensor,
+    tmp_out: torch.Tensor,
+    query: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    num_kv_heads: int,
+    scale: float,
+    block_tables: torch.Tensor,
+    seq_lens: torch.Tensor,
+    block_size: int,
+    max_seq_len: int,
+    alibi_slopes: Optional[torch.Tensor],
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+    tp_rank: int = 0,
+    blocksparse_local_blocks: int = 0,
+    blocksparse_vert_stride: int = 0,
+    blocksparse_block_size: int = 64,
+    blocksparse_head_sliding_step: int = 0,
+) -> None:
+    ops.paged_attention_v2(
+        out,
+        exp_sum,
+        max_logits,
+        tmp_out,
+        query,
+        key_cache,
+        value_cache,
+        num_kv_heads,
+        scale,
+        block_tables,
+        seq_lens,
+        block_size,
+        max_seq_len,
+        alibi_slopes,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+        tp_rank,
+        blocksparse_local_blocks,
+        blocksparse_vert_stride,
+        blocksparse_block_size,
+        blocksparse_head_sliding_step,
+    )
+def reshape_and_cache(
+    key: torch.Tensor,
+    value: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    slot_mapping: torch.Tensor,
+    kv_cache_dtype: str,
+    k_scale: float,
+    v_scale: float,
+) -> None:
+    ops.reshape_and_cache(
+        key,
+        value,
+        key_cache,
+        value_cache,
+        slot_mapping,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+    )
+def reshape_and_cache_flash(
+    key: torch.Tensor,
+    value: torch.Tensor,
+    key_cache: torch.Tensor,
+    value_cache: torch.Tensor,
+    slot_mapping: torch.Tensor,
+    kv_cache_dtype: str,
+    k_scale: torch.Tensor,
+    v_scale: torch.Tensor,
+) -> None:
+    ops.reshape_and_cache_flash(
+        key,
+        value,
+        key_cache,
+        value_cache,
+        slot_mapping,
+        kv_cache_dtype,
+        k_scale,
+        v_scale,
+    )
+def copy_blocks(
+    key_caches: List[torch.Tensor],
+    value_caches: List[torch.Tensor],
+    block_mapping: torch.Tensor,
+) -> None:
+    ops.copy_blocks(key_caches, value_caches, block_mapping)
+def swap_blocks(
+    src: torch.Tensor, dst: torch.Tensor, block_mapping: torch.Tensor
+) -> None:
+    ops.swap_blocks(src, dst, block_mapping)
+def convert_fp8(
+    output: torch.Tensor, input: torch.Tensor, scale: float = 1.0, kv_dtype: str = "fp8"
+) -> None:
+    ops.convert_fp8(output, input, scale, kv_dtype)
+__all__ = [
+    "convert_fp8",
+    "paged_attention_v1",
+    "paged_attention_v2",
+    "reshape_and_cache",
+    "copy_blocks",
+]

build/torch211-cxx11-rocm72-x86_64-linux/_ops.py ADDED Viewed

	@@ -0,0 +1,9 @@

+import torch
+from . import _paged_attention_rocm_0278991
+ops = torch.ops._paged_attention_rocm_0278991
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_paged_attention_rocm_0278991::{op_name}"

build/torch211-cxx11-rocm72-x86_64-linux/_paged_attention_rocm_0278991.abi3.so ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:dfa4a719ca8f34b3e46947f2c30b8d8d434d360870317816711f60343f01a5f1
+size 56640560

build/torch211-cxx11-rocm72-x86_64-linux/metadata.json ADDED Viewed

	@@ -0,0 +1,17 @@

+{
+  "version": 1,
+  "license": "Apache-2.0",
+  "python-depends": [],
+  "backend": {
+    "type": "rocm",
+    "archs": [
+      "gfx1030",
+      "gfx1100",
+      "gfx1101",
+      "gfx906",
+      "gfx908",
+      "gfx90a",
+      "gfx942"
+    ]
+  }
+}

build/torch211-cxx11-rocm72-x86_64-linux/paged_attention/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+import ctypes
+import importlib.util
+import sys
+from pathlib import Path
+from types import ModuleType
+def _import_from_path(file_path: Path) -> ModuleType:
+    # We cannot use the module name as-is, after adding it to `sys.modules`,
+    # it would also be used for other imports. So, we make a module name that
+    # depends on the path for it to be unique using the hex-encoded hash of
+    # the path.
+    path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
+    module_name = path_hash
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None:
+        raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    if module is None:
+        raise ImportError(f"Cannot load module {module_name} from spec")
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)  # type: ignore
+    return module
+globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))

build/torch211-cxx11-rocm72-x86_64-linux/platforms.py ADDED Viewed

	@@ -0,0 +1,92 @@

+import os
+import random
+from abc import ABC, abstractmethod
+from functools import lru_cache, wraps
+from typing import Callable, ParamSpec, TypeVar
+import numpy as np
+import torch
+IS_ROCM = torch.version.hip is not None
+IS_MPS = torch.backends.mps.is_available()
+class Platform(ABC):
+    @classmethod
+    def seed_everything(cls, seed: int) -> None:
+        """
+        Set the seed of each random module.
+        `torch.manual_seed` will set seed on all devices.
+        Loosely based on: https://github.com/Lightning-AI/pytorch-lightning/blob/2.4.0/src/lightning/fabric/utilities/seed.py#L20
+        """
+        random.seed(seed)
+        np.random.seed(seed)
+        torch.manual_seed(seed)
+    @abstractmethod
+    def get_device_name(self, device_id: int = 0) -> str: ...
+    @abstractmethod
+    def is_cuda(self) -> bool: ...
+    @abstractmethod
+    def is_rocm(self) -> bool: ...
+    @abstractmethod
+    def is_mps(self) -> bool: ...
+class CudaPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(0)
+    def is_cuda(self) -> bool:
+        return True
+    def is_rocm(self) -> bool:
+        return False
+    def is_mps(self) -> bool:
+        return False
+class RocmPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(device_id)
+    def is_cuda(self) -> bool:
+        return False
+    def is_rocm(self) -> bool:
+        return True
+    def is_mps(self) -> bool:
+        return False
+class MpsPlatform(Platform):
+    @classmethod
+    @lru_cache(maxsize=8)
+    def get_device_name(cls, device_id: int = 0) -> str:
+        return torch.cuda.get_device_name(device_id)
+    def is_cuda(self) -> bool:
+        return False
+    def is_rocm(self) -> bool:
+        return False
+    def is_mps(self) -> bool:
+        return True
+current_platform = (
+    RocmPlatform() if IS_ROCM else
+    MpsPlatform() if IS_MPS else
+    CudaPlatform() if torch.cuda.is_available() else
+    None
+)

build/torch29-cxx11-cu129-x86_64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _paged_attention_cuda_83cf4a3
-ops = torch.ops._paged_attention_cuda_83cf4a3
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_paged_attention_cuda_83cf4a3::{op_name}"

 import torch
+from . import _paged_attention_cuda_0278991
+ops = torch.ops._paged_attention_cuda_0278991
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_paged_attention_cuda_0278991::{op_name}"

build/torch29-cxx11-cu129-x86_64-linux/{_paged_attention_cuda_83cf4a3.abi3.so → _paged_attention_cuda_0278991.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:914b49c90182aad733f2f3060f3738aac9faf4baafe7fb308a3f1d7aacea3cf2
 size 182088696

 version https://git-lfs.github.com/spec/v1
+oid sha256:70b87d9a50353e72cf4573916bfc5482062855d5834cb10a89e9548c3cd129aa
 size 182088696