Update README.md

README.md

@@ -2,9 +2,6 @@
 tags:
 - kernel
 ---
-
-![Status](https://hubwebhook.dholtz.com/shield?repo=kernels-community/activation)
-
 ## Activation
 
 Activation kernels from [vLLM](https://github.com/vllm-project/vllm/blob/main/csrc/activation_kernels.cu).

build.toml

@@ -1,18 +1,17 @@
 [general]
 name = "activation"
-universal = false
 
 [torch]
 src = [
-    "torch-ext/torch_binding.cpp",
-    "torch-ext/torch_binding.h",
+  "torch-ext/torch_binding.cpp",
+  "torch-ext/torch_binding.h"
 ]
 
 [kernel.activation]
-backend = "cuda"
-depends = ["torch"]
+cuda-capabilities = [ "7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0" ]
 src = [
-    "activation/activation_kernels.cu",
-    "activation/cuda_compat.h",
-    "activation/dispatch_utils.h",
+  "activation/activation_kernels.cu",
+  "activation/cuda_compat.h",
+  "activation/dispatch_utils.h",
 ]
+depends = [ "torch" ]

build/{torch27-cxx11-cu118-x86_64-linux → torch25-cxx11-cu118-x86_64-linux}/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/{torch26-cxx11-cu118-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so → torch25-cxx11-cu118-x86_64-linux/activation/_activation_o63kkyjirmkf4.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b925dc27b6a9afd5b6d11e454275222c531a92f7ca27958ac81a78c580665e4d
-size 2448088
+oid sha256:d50cdabfbed1df74e921ac34ff00bca0555977b14ef8082ddae7b1f30985a494
+size 2370160

build/{torch27-cxx11-cu118-x86_64-linux → torch25-cxx11-cu118-x86_64-linux}/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_o63kkyjirmkf4
+ops = torch.ops._activation_o63kkyjirmkf4
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_o63kkyjirmkf4::{op_name}"

build/{torch27-cxx11-cu126-x86_64-linux → torch25-cxx11-cu121-x86_64-linux}/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/{torch26-cxx11-cu126-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so → torch25-cxx11-cu121-x86_64-linux/activation/_activation_vrl36m2ejer54.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70e544ad6448a5576d26147f48403f3e9e593f4a2e24167dc8acb81ce3b7932e
-size 2518600
+oid sha256:2bd0709ef09c8f0c18d1dc4a36c8096c59459bece61f5f5dbea95d1e73f54d44
+size 2393264

build/{torch27-cxx11-cu126-x86_64-linux → torch25-cxx11-cu121-x86_64-linux}/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_vrl36m2ejer54
+ops = torch.ops._activation_vrl36m2ejer54
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_vrl36m2ejer54::{op_name}"

build/{torch27-cxx11-cu128-x86_64-linux → torch25-cxx11-cu124-x86_64-linux}/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/{torch26-cxx11-cu124-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so → torch25-cxx11-cu124-x86_64-linux/activation/_activation_va3moa75vw7c2.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cfdbe510752b57a8dc4671f744bb0a2da5b1646e0b9a19fec02f1505ba044c8c
-size 2509960
+oid sha256:8353447f64e7d2df1a6a341d9c53bced53abef267f079923ae774170d0d57c53
+size 2427936

build/{torch27-cxx11-cu128-x86_64-linux → torch25-cxx11-cu124-x86_64-linux}/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_va3moa75vw7c2
+ops = torch.ops._activation_va3moa75vw7c2
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_va3moa75vw7c2::{op_name}"

build/torch25-cxx98-cu118-x86_64-linux/activation/__init__.py

@@ -0,0 +1,47 @@
+import torch
+
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
+
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
+
+
+def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.silu_and_mul(out, x)
+    return out
+
+
+def gelu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_and_mul(out, x)
+    return out
+
+
+def gelu_tanh_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_tanh_and_mul(out, x)
+    return out
+
+
+def fatrelu_and_mul(out: torch.Tensor, x: torch.Tensor, threshold: float = 0.0) -> None:
+    ops.fatrelu_and_mul(out, x, threshold)
+    return out
+
+
+def gelu_fast(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_fast(out, x)
+    return out
+
+
+def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_new(out, x)
+    return out
+
+
+def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_quick(out, x)
+    return out

build/{torch26-cxx98-cu118-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so → torch25-cxx98-cu118-x86_64-linux/activation/_activation_qr3gs3eckeig4.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:60fd224c33657558f03be5be57cc8d35ade23225b1abd71557b170c8a7010cd1
-size 2440576
+oid sha256:df184a6315118d787a1bd6b435cb45f1ca7828445a1f1c0e55c57645cfbba43a
+size 2362600

build/torch25-cxx98-cu118-x86_64-linux/activation/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _activation_qr3gs3eckeig4
+ops = torch.ops._activation_qr3gs3eckeig4
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_activation_qr3gs3eckeig4::{op_name}"

build/torch25-cxx98-cu121-x86_64-linux/activation/__init__.py

@@ -0,0 +1,47 @@
+import torch
+
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
+
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
+
+
+def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.silu_and_mul(out, x)
+    return out
+
+
+def gelu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_and_mul(out, x)
+    return out
+
+
+def gelu_tanh_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_tanh_and_mul(out, x)
+    return out
+
+
+def fatrelu_and_mul(out: torch.Tensor, x: torch.Tensor, threshold: float = 0.0) -> None:
+    ops.fatrelu_and_mul(out, x, threshold)
+    return out
+
+
+def gelu_fast(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_fast(out, x)
+    return out
+
+
+def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_new(out, x)
+    return out
+
+
+def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_quick(out, x)
+    return out

build/torch25-cxx98-cu121-x86_64-linux/activation/_activation_p7gbzt25w3zg2.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccb13cfc2e45cf483e8b9f77f1760f28b48bcf185508d51b32d45bc759c4e8bb
+size 2385440

build/torch25-cxx98-cu121-x86_64-linux/activation/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _activation_p7gbzt25w3zg2
+ops = torch.ops._activation_p7gbzt25w3zg2
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_activation_p7gbzt25w3zg2::{op_name}"

build/torch25-cxx98-cu124-x86_64-linux/activation/__init__.py

@@ -0,0 +1,47 @@
+import torch
+
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
+
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
+
+
+def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.silu_and_mul(out, x)
+    return out
+
+
+def gelu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_and_mul(out, x)
+    return out
+
+
+def gelu_tanh_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_tanh_and_mul(out, x)
+    return out
+
+
+def fatrelu_and_mul(out: torch.Tensor, x: torch.Tensor, threshold: float = 0.0) -> None:
+    ops.fatrelu_and_mul(out, x, threshold)
+    return out
+
+
+def gelu_fast(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_fast(out, x)
+    return out
+
+
+def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_new(out, x)
+    return out
+
+
+def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
+    ops.gelu_quick(out, x)
+    return out

build/torch25-cxx98-cu124-x86_64-linux/activation/_activation_jg7yaigtn7wco.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f8048853e8cb06e8574a9a9497800d2be438f7989d79f44dcf2e0ced38a75a9
+size 2420192

build/torch25-cxx98-cu124-x86_64-linux/activation/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _activation_jg7yaigtn7wco
+ops = torch.ops._activation_jg7yaigtn7wco
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_activation_jg7yaigtn7wco::{op_name}"

build/torch26-cxx11-cu118-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx11-cu118-x86_64-linux/activation/_activation_ncisyrun7guwk.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cde5439e78ba0e1aaa1937d798b214b46d38cbab8e4384b93a22239fed1a4dd4
+size 2370264

build/torch26-cxx11-cu118-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_ncisyrun7guwk
+ops = torch.ops._activation_ncisyrun7guwk
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_ncisyrun7guwk::{op_name}"

build/torch26-cxx11-cu118-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch26-cxx11-cu124-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx11-cu124-x86_64-linux/activation/_activation_ochhfvlnc3vyc.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6bd20d411c51fc8729b15cab6a60c5c9185222474aa035489e1bff299d76682
+size 2428040

build/torch26-cxx11-cu124-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_ochhfvlnc3vyc
+ops = torch.ops._activation_ochhfvlnc3vyc
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_ochhfvlnc3vyc::{op_name}"

build/torch26-cxx11-cu124-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch26-cxx11-cu126-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx11-cu126-x86_64-linux/activation/_activation_u6vnqubnicksq.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41c18b20c2bf8c49d2d3088a9bc1aad4293df0b57eafc9b141a9e8e595fe551a
+size 2436672

build/torch26-cxx11-cu126-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_u6vnqubnicksq
+ops = torch.ops._activation_u6vnqubnicksq
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_u6vnqubnicksq::{op_name}"

build/torch26-cxx11-cu126-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch26-cxx98-cu118-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx98-cu118-x86_64-linux/activation/_activation_2vn6ty3gfqfb6.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfbcd5da358cd5cb7982d19c8880cf4db6f08b46622a7a953f755ad59e4e1492
+size 2362752

build/torch26-cxx98-cu118-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_2vn6ty3gfqfb6
+ops = torch.ops._activation_2vn6ty3gfqfb6
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_2vn6ty3gfqfb6::{op_name}"

build/torch26-cxx98-cu118-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch26-cxx98-cu124-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx98-cu124-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e364773259dc1b91f3c0d3b076da83c5a9c6ee18ffdace30315c602dffd1dabe
-size 2502264

build/torch26-cxx98-cu124-x86_64-linux/activation/_activation_myvteedxdpqc6.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1bc928823117c800904bcd3492bf1a0c65a32f6d8a842dc039f55e29831ab49
+size 2420344

build/torch26-cxx98-cu124-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_myvteedxdpqc6
+ops = torch.ops._activation_myvteedxdpqc6
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_myvteedxdpqc6::{op_name}"

build/torch26-cxx98-cu124-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch26-cxx98-cu126-x86_64-linux/activation/__init__.py

@@ -1,8 +1,15 @@
 import torch
 
-from ._ops import ops
+try:
+    from ._ops import ops
+except ImportError as e:
+    # Fallback for local development.
+    try:
+        import _activation
 
-from . import layers
+        ops = torch.ops._activition
+    except ImportError:
+        raise e
 
 
 def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
@@ -38,15 +45,3 @@ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
 def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
     ops.gelu_quick(out, x)
     return out
-
-
-__all__ = [
-    "silu_and_mul",
-    "gelu_and_mul",
-    "gelu_tanh_and_mul",
-    "fatrelu_and_mul",
-    "gelu_fast",
-    "gelu_new",
-    "gelu_quick",
-    "layers",
-]

build/torch26-cxx98-cu126-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:7ac88cc0d3c65ab283d20608f3a097be29ee572e7856f10f8d7919536efd95b4
-size 2506808

build/torch26-cxx98-cu126-x86_64-linux/activation/_activation_rbswus6emrhm2.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:474727e434a9cd4ec984a6da7124992ead4ca0fefce9581d0fd503e36c065aed
+size 2424888

build/torch26-cxx98-cu126-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_e99cc09_dirty
-ops = torch.ops._activation_e99cc09_dirty
+from . import _activation_rbswus6emrhm2
+ops = torch.ops._activation_rbswus6emrhm2
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_e99cc09_dirty::{op_name}"
+    return f"_activation_rbswus6emrhm2::{op_name}"

build/torch26-cxx98-cu126-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch27-cxx11-cu118-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e4f9e647eea40d3d3801d5ee57d4917e4c2e8dbfd87cdfebdc40b1b0a1c571fe
-size 2448184

build/torch27-cxx11-cu118-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out

build/torch27-cxx11-cu126-x86_64-linux/activation/_activation_e99cc09_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a2b72ff2a0f2253e4dfe028842b5f15cabf2647d7812bf4662a2de510ca0c489
-size 2518632

build/torch27-cxx11-cu126-x86_64-linux/activation/layers.py

@@ -1,79 +0,0 @@
-import torch
-import torch.nn as nn
-
-from ._ops import ops
-
-
-class SiluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.silu_and_mul(out, x)
-        return out
-
-
-class GeluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_and_mul(out, x)
-        return out
-
-
-class GeluTanhAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.gelu_tanh_and_mul(out, x)
-        return out
-
-
-class FatreluAndMul(nn.Module):
-    can_torch_compile: bool = True
-
-    def __init__(self, threshold: float = 0.0):
-        super().__init__()
-        self.threshold = threshold
-
-    def forward(self, x: torch.Tensor):
-        d = x.shape[-1] // 2
-        output_shape = x.shape[:-1] + (d,)
-        out = torch.empty(output_shape, dtype=x.dtype, device=x.device)
-        ops.fatrelu_and_mul(out, x, self.threshold)
-        return out
-
-
-class FastGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_fast(out, x)
-        return out
-
-
-class NewGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_new(out, x)
-        return out
-
-
-class QuickGELU(nn.Module):
-    can_torch_compile: bool = True
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        out = torch.empty_like(x)
-        ops.gelu_quick(out, x)
-        return out