Revert "fix typo in readme (#7)" (#8)

This reverts commit 2d926c3d4773b82c5a5f03904780a40d25a1afee.

benchmarks/README.md

@@ -15,7 +15,7 @@ Results can be saved as **CSV files** or **plots**.
 ## Usage
 
 ```bash
-python run_cases.py --case <CASE> [--plot] [--save-path <DIR>]
+python main.py --case <CASE> [--plot] [--save-path <DIR>]
 ```
 
 - `--case` (required): one of `rms`, `add_rms`, `poly`, `mul_poly`
@@ -25,8 +25,8 @@ python run_cases.py --case <CASE> [--plot] [--save-path <DIR>]
 ## Examples
 
 ```bash
-python run_cases.py --case add_rms --save-path ./results/
-python run_cases.py --case poly --plot --save-path ./plots/
+python main.py --case add_rms --save-path ./results/
+python main.py --case poly --plot --save-path ./plots/
 ```
 
 ## Output

build/{torch27-cxx11-rocm63-x86_64-linux/activation/_activation_2f66548_dirty.abi3.so → torch27-cxx11-cu118-x86_64-linux/activation/_activation_20250907180255.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:525a7d6eebafd2c18e2a15c17162b8328c5ade11f87fe9032a136e92c182e888
-size 2764584
+oid sha256:d21a85bf21aa74f1281541e658acfd4f4326d902efe3578b059eccf054443284
+size 8089696

build/torch27-cxx11-cu118-x86_64-linux/activation/{_activation_2f66548_dirty.abi3.so → _activation_e5e2eeb_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eee7aed4f37c13c6accf42db53acad998b46dfc3c6fd70e976b552482a08118e
+oid sha256:ec9ea7edc8b27f7983e20d615ab470cef6b82975afc214becfddfd05a867a839
 size 8600336

build/torch27-cxx11-cu118-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd84c828d4c15e96d65d6c8f0eb7a945ee8167d92e978b2ebce03eeaf41e7fce
+size 4405112

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

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

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

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch27-cxx11-cu126-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74d4955271509451b946495da75f69a0f978e7258b8303fe3c077e585c0d3e6a
+size 8272456

build/torch27-cxx11-cu126-x86_64-linux/activation/{_activation_2f66548_dirty.abi3.so → _activation_e5e2eeb_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:06901740688c1d1eed4d91275d28864866ac4a768ea6b97d4894f15869f5cedd
+oid sha256:5d3511410cdc288d2fafc500223ed2e625e360f50fa341809cf892fb2c822924
 size 8779000

build/torch27-cxx11-cu126-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caffcadbb99fbaa27e8a81d5ef508f2e1a798e7626d618c3cf5b0d387d2c8686
+size 4618624

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

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

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

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch27-cxx11-cu128-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bf0d2ab5ff5520704e0b0c959b61d0043d360cfd4335950e69677873a87e436
+size 12792112

build/torch27-cxx11-cu128-x86_64-linux/activation/{_activation_2f66548_dirty.abi3.so → _activation_e5e2eeb_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b85e3e1f218392802dc8642b037f06ad7ac388d597ca9d1fb2b7314e9af5f30
+oid sha256:25efc9c32e4bd6609a8326025aad861cbf79b544893755fe44519c9df7224c40
 size 13818872

build/torch27-cxx11-cu128-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b7c6ece8e8d316c4cc5fe46b1cec4422b2f61e9bb7240af71a2b4a35975d8e6
+size 6676528

build/torch27-cxx11-cu128-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch27-cxx11-cu128-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch27-cxx11-rocm63-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:640322a8fac8fd9d8e9f195a3034c4ee0f81ee1acf897fd7c482a84ce47a1bec
+size 4160688

build/torch27-cxx11-rocm63-x86_64-linux/activation/_activation_e5e2eeb_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c80d05690547f2842d416ebb85c9f830370373bc7e6c54ba08eec61b3690280f
+size 4386744

build/torch27-cxx11-rocm63-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4be173820e2a4bf4b6b8de6b63faf6544b599d9b0583f650a940adaef4a048b3
+size 2899184

build/torch27-cxx11-rocm63-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch27-cxx11-rocm63-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch28-cxx11-cu126-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1768d8d5072ac06d937cb5332988c6b3bfaa191f72d1369a22d2c577e9a3bca2
+size 8215280

build/torch28-cxx11-cu126-x86_64-linux/activation/{_activation_2f66548_dirty.abi3.so → _activation_e5e2eeb_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0d8ee61b8e62210df21e40e31560b84c4e985ba448917a8e95bf44167cb9505
+oid sha256:440f5c17a7ddaf73c506bbc84fd1405e2e188b8ceaf4977910608be6b91e89bf
 size 8730200

build/torch28-cxx11-cu126-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb222449350310f90f7271f34fcf9052c9eec28021fee0348130a8f239a97bf4
+size 4571976

build/torch28-cxx11-cu126-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch28-cxx11-cu126-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch28-cxx11-cu128-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37a572bd877980ab8c0331ca5682191cb5a2b1f05bc69ea493a9e24f7728ba3f
+size 12730840

build/torch28-cxx11-cu128-x86_64-linux/activation/{_activation_2f66548_dirty.abi3.so → _activation_e5e2eeb_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cad1446a247f173104e280c0e38fef98cdb68aa005f76a555192b73397fdda7a
+oid sha256:1dfb6d468f9cef0239d4ea47f0a247fa721befc5b8db86e1cddfc25f1814b67a
 size 13770064

build/torch28-cxx11-cu128-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79be6527f579de1133e50a66310d7d0690649dcac63009a54b5e68809408f12a
+size 6634208

build/torch28-cxx11-cu128-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch28-cxx11-cu128-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch28-cxx11-cu129-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f15919c4cac697cde550af16256e338472400e50df751e93622350c7f626bc8
+size 12726208

build/torch28-cxx11-cu129-x86_64-linux/activation/_activation_2f66548_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9fddc0b356a5bae74d48638c35d3d0d566812a152d249534371c5e733c3d0dce
-size 13753152

build/torch28-cxx11-cu129-x86_64-linux/activation/_activation_e5e2eeb_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0815a50e61497b357b2b90fc28602b3f53a25da1161edd2cb0b0fbebc7c62bf6
+size 13757248

build/torch28-cxx11-cu129-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d95e4491d35cb022a6eaa2febbc555f203893f989a4fb1cc483b2632f141869
+size 6687456

build/torch28-cxx11-cu129-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch28-cxx11-cu129-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch28-cxx11-rocm63-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e72d4bb4459a5da96ca5eda1d305237a361140f0e25360e3d20326a22f1b6d47
+size 4165584

build/torch28-cxx11-rocm63-x86_64-linux/activation/_activation_2f66548_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3bebd1aa6a3414be5d77c1306c50310d05df8d75a7ea6c3dbd2bb6bb41a98685
-size 2765376

build/torch28-cxx11-rocm63-x86_64-linux/activation/_activation_e5e2eeb_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d404c88b72f1b6da551a64b3373395e80403a52ccff14fc401be3e8ee184d83
+size 4387536

build/torch28-cxx11-rocm63-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58116124bb2b5d11de2753dd0c30a1e4c84759f18599da7016c791bad37528e9
+size 2899984

build/torch28-cxx11-rocm63-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch28-cxx11-rocm63-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch28-cxx11-rocm64-x86_64-linux/activation/_activation_20250907180255.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3325c2748cf7a070383068995078f93f440cc95fbed491d00bd414cdd851376
+size 4171472

build/torch28-cxx11-rocm64-x86_64-linux/activation/_activation_2f66548_dirty.abi3.so

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:49b998e99c3b26775a59f8bdfe2033fd7871ce4743f8234a4d8360c9867afdaf
-size 2771080

build/torch28-cxx11-rocm64-x86_64-linux/activation/_activation_e5e2eeb_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8d52dee20ba3c4619f7c614984f656f34f32dd74ba6cf866cf80f32245117cf
+size 4393240

build/torch28-cxx11-rocm64-x86_64-linux/activation/_activation_f517c97_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65319d3d93ac3bf0f2939fa4e53ddfc8cd633b9e396cde3a97d63b9041ba03a7
+size 2885344

build/torch28-cxx11-rocm64-x86_64-linux/activation/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _activation_2f66548_dirty
-ops = torch.ops._activation_2f66548_dirty
+from . import _activation_e5e2eeb_dirty
+ops = torch.ops._activation_e5e2eeb_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_activation_2f66548_dirty::{op_name}"
+    return f"_activation_e5e2eeb_dirty::{op_name}"

build/torch28-cxx11-rocm64-x86_64-linux/activation/rms_norm.py

@@ -70,9 +70,8 @@ class FusedAddRMSNormFunction(torch.autograd.Function):
         weight_grad = torch.empty_like(
             weight) if ctx.needs_input_grad[2] else None
 
-        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad,
-                                        add_output_grad, add_output, weight,
-                                        eps)
+        ops.fused_add_rms_norm_backward(grad, weight_grad, output_grad, add_output_grad, add_output,
+                              weight, eps)
         input_grad = grad if need_in else None
         residual_grad = grad if need_res else None
 

build/torch29-cxx11-cu126-x86_64-linux/activation/__init__.py

@@ -1,52 +0,0 @@
-import torch
-
-from . import layers
-from ._ops import ops
-from .poly_norm import FusedMulPolyNormFunction, PolyNormFunction
-from .rms_norm import FusedAddRMSNormFunction, RMSNormFunction
-
-
-def poly_norm(
-    x: torch.Tensor,
-    weight: torch.Tensor,
-    bias: torch.Tensor,
-    eps: float = 1e-6,
-) -> None:
-    return PolyNormFunction.apply(x, weight, bias, eps)
-
-
-def fused_mul_poly_norm(
-    x: torch.Tensor,
-    mul: torch.Tensor,
-    weight: torch.Tensor,
-    bias: torch.Tensor,
-    eps: float = 1e-6,
-) -> None:
-    return FusedMulPolyNormFunction.apply(x, mul, weight, bias, eps)
-
-
-def rms_norm(
-    x: torch.Tensor,
-    weight: torch.Tensor,
-    eps: float = 1e-6,
-) -> None:
-    return RMSNormFunction.apply(x, weight, eps)
-
-
-def fused_add_rms_norm(
-    x: torch.Tensor,
-    residual: torch.Tensor,
-    weight: torch.Tensor,
-    eps: float = 1e-6,
-) -> None:
-    return FusedAddRMSNormFunction.apply(x, residual, weight, eps)
-
-
-__all__ = [
-    "poly_norm",
-    "fused_mul_poly_norm",
-    "rms_norm",
-    "fused_add_rms_norm",
-    "layers",
-    "ops",
-]