Uploaded using `kernel-builder`.

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch210-cxx11-cu126-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2dc06be886cf33ed300ddf18a4b47a0c6bf86ecaf3fcf1270f7c76e2155475ff
+oid sha256:b2ca4629138cc61e16f0e0440f65c0f012cce964a49394a3c011e61bbeed789f
 size 8200568

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch210-cxx11-cu128-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29271f8a63c361c613578dd95dcf90de69567dc6711dc0bbd3741626d54ae07a
+oid sha256:f661db8bac0b471dbe2a53a72ac0a3060294002c0158fd1c2dfc1fb635bf3b1a
 size 11905904

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch210-cxx11-cu130-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92f83279a8fe700b52af5bcb004ece45dd678061d90d790f6576961a040b3a3e
-size 10310752
+oid sha256:a5de9e68f719d1e9f5feb466c2ab6844e64191923103f83bce9636992f8c79e2
+size 10307160

build/torch210-cxx11-xpu20253-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_xpu_17de4fe
-ops = torch.ops._rotary_xpu_17de4fe
+from . import _rotary_xpu_1ff2c4a
+ops = torch.ops._rotary_xpu_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_xpu_17de4fe::{op_name}"
+    return f"_rotary_xpu_1ff2c4a::{op_name}"

build/torch210-cxx11-xpu20253-x86_64-linux/{_rotary_xpu_17de4fe.abi3.so → _rotary_xpu_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53435ac1dc220c7bacfff30fe0f9119269d8f08928780ef8d53f664cc8d462fa
+oid sha256:071f6c80b72223f9640eeb229ded44d36f13826714ed1ec90dc3317b0400b539
 size 2301504

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch211-cxx11-cu126-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e784ee2ad7a9c68601fc0e40734f940e15e83b738e514b275c164f7bb5740cf
+oid sha256:26abf96731f0a53085c242187f0955d70cc20e8c9f6047a2d4d1569efc07d87e
 size 8193600

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch211-cxx11-cu128-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5b56c5504c4f8fccd032bb25a6dcfd168ba35b62edb22b68a31cb48e64113aa
+oid sha256:9c5c9ff2340b8a94fbc7920c238556b18498c903d8d931da93fda84c1d9ae675
 size 11894840

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch211-cxx11-cu130-x86_64-linux/_rotary_cuda_17de4fe.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:807bdc834886287cbf730c475412278109a6f43131e95b15b66fe75686f54634
-size 10303784

build/torch211-cxx11-cu130-x86_64-linux/_rotary_cuda_1ff2c4a.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ef1f14ac10914f04945d5512be813414ebe128499b81267a1e2d41bf15b0393
+size 10296096

build/torch211-cxx11-xpu20253-x86_64-linux/__init__.py

@@ -0,0 +1,52 @@
+from typing import Optional, Tuple
+import torch
+
+from ._ops import ops
+
+
+def apply_rotary(
+    x1: torch.Tensor,
+    x2: torch.Tensor,
+    cos: torch.Tensor,
+    sin: torch.Tensor,
+    out1: torch.Tensor,
+    out2: torch.Tensor,
+    conj: bool,
+) -> None:
+    ops.apply_rotary(x1, x2, cos, sin, out1, out2, conj)
+
+
+def apply_rotary_transformers(
+    q: torch.Tensor,
+    k: torch.Tensor,
+    cos: torch.Tensor,
+    sin: torch.Tensor,
+    unsqueeze_dim: int = 1,
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    """
+    Rotary kernel implementation wrapper
+    Adapts rotary kernel implementation to match transformers apply_rotary_pos_emb signature
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    q_rotated = q.clone()
+    k_rotated = k.clone()
+
+    # Get half dimension for rotation
+    half_dim = q.shape[-1] // 2
+    q1 = q_rotated[..., :half_dim]
+    q2 = q_rotated[..., half_dim:]
+    k1 = k_rotated[..., :half_dim]
+    k2 = k_rotated[..., half_dim:]
+    if cos.shape[-1] != half_dim:
+        # Trim cos/sin to match half_dim
+        cos = cos[..., :half_dim]
+        sin = sin[..., :half_dim]
+
+    apply_rotary(q1, q2, cos, sin, q1, q2, False)
+    apply_rotary(k1, k2, cos, sin, k1, k2, False)
+    return q_rotated, k_rotated
+
+
+__all__ = ["apply_rotary", "apply_rotary_transformers"]

build/torch211-cxx11-xpu20253-x86_64-linux/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rotary_xpu_1ff2c4a
+ops = torch.ops._rotary_xpu_1ff2c4a
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rotary_xpu_1ff2c4a::{op_name}"

build/torch211-cxx11-xpu20253-x86_64-linux/_rotary_xpu_1ff2c4a.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:866dec73565cc4bc480f07f9b176b617a445e79a23c030820cf36f4c667f1040
+size 2301504

build/torch211-cxx11-xpu20253-x86_64-linux/metadata.json

@@ -0,0 +1,8 @@
+{
+  "version": 1,
+  "license": "BSD-3-Clause",
+  "python-depends": [],
+  "backend": {
+    "type": "xpu"
+  }
+}

build/torch211-cxx11-xpu20253-x86_64-linux/rotary/__init__.py

@@ -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/torch29-cxx11-cu129-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _rotary_cuda_17de4fe
-ops = torch.ops._rotary_cuda_17de4fe
+from . import _rotary_cuda_1ff2c4a
+ops = torch.ops._rotary_cuda_1ff2c4a
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rotary_cuda_17de4fe::{op_name}"
+    return f"_rotary_cuda_1ff2c4a::{op_name}"

build/torch29-cxx11-cu129-x86_64-linux/{_rotary_cuda_17de4fe.abi3.so → _rotary_cuda_1ff2c4a.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcd3ab1517eabfaf2fcca055deaf85cc8031f097d7982df460a891bea5206add
+oid sha256:73512553b334acb47e085a7975ead67470641381044836837d819adffd22e1b5
 size 11964176