chore(muon): update comment

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx11-cu118-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9119d3a6d99c07a17d110d2ccf6042f199d00c839f5efa74008c1642d21e48b0
+oid sha256:98bd4b647ad0ecbae82a5e78f618475b47595c5bb68b3356c09ee8b1f1a57060
 size 1787272

build/torch26-cxx11-cu118-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx11-cu124-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91b76cd5be429f99840e26e8ba55b61f9fdcae19301bd7c082b2e9746a276501
+oid sha256:796ac374cd2eec4260591c5a771c6b324f7dc6c8f34fc5dc211ab8afca546ffe
 size 1824224

build/torch26-cxx11-cu124-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx11-cu126-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:807d59aca5b0403206395a1f4c770b8d644294c17f6af866207c36ac617f0a7d
+oid sha256:254706f111eb794b1409ba48d25649ace5438e2c66027727e84490011ee4c5e6
 size 1824224

build/torch26-cxx11-cu126-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

build/torch26-cxx11-rocm62-x86_64-linux/optimizer/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx11-rocm62-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d2bdd755079fa06a27401b8a26ac425d35514d196f9df4ce1be5c52ebcc9a64
+oid sha256:027a26212a3dd705876ca83015a53b69d17d80fe7c1559fb01d7aacf614edb57
 size 1749744

build/torch26-cxx11-rocm62-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx98-cu118-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4ca177074d4c04630ffaa2e49e41e1451bf198c44c4cc544a664be88475a3b9
+oid sha256:62c4408eaf54197941241ae6150afe1401a8bcf5854488a8b957d1f1546b388a
 size 1787192

build/torch26-cxx98-cu118-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx98-cu124-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3511c3a46297462166d7b773dc2bd8b16f43b7004eee1e4b31d468113051fb55
+oid sha256:166d253c91459e1aa1328a1550b0e3ec4bb7c6057870b1d7472a93cc987cf85a
 size 1824184

build/torch26-cxx98-cu124-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch26-cxx98-cu126-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fa1dad3b3c1d94e7613a35e42afb8c7974d7bf6ce25cd2766590ba65b129f07
+oid sha256:8bb7315b326f9af7a77e023c2b78511190235a8dcc9682abd5b49db1dc2b90f2
 size 1824184

build/torch26-cxx98-cu126-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch27-cxx11-cu118-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe5761d07ed965bf94d00d8a8e6753a7fb571271e73773de9021511e0e6ae2c7
+oid sha256:0a74351ee471271eaf1c8292ed01b7e71e6b1b683704144d68d90b67032ba386
 size 1787368

build/torch27-cxx11-cu118-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch27-cxx11-cu126-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a35c1c4d46f677f0fe35fec9023a866b9bd0f4245624b4e71a9812a1864c01e6
+oid sha256:ffb7e3a786405106908da16e74506fe381b09e5e04a27b1062396e378f63f7f7
 size 1824256

build/torch27-cxx11-cu126-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch27-cxx11-cu128-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b27a334f5b3c1dd922468fc93662f90cc95b4213f3f96a212e34ea8e4f3bf03
+oid sha256:45ee6c653f216af96705a25993d85751648ccd4714a8d6c8c36bdbc8dc19edc5
 size 1883352

build/torch27-cxx11-cu128-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_bdd2678_dirty
-ops = torch.ops._optimizer_bdd2678_dirty
+from . import _optimizer_febdf5b_dirty
+ops = torch.ops._optimizer_febdf5b_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_bdd2678_dirty::{op_name}"
+    return f"_optimizer_febdf5b_dirty::{op_name}"

build/torch27-cxx11-rocm63-x86_64-linux/optimizer/{_optimizer_bdd2678_dirty.abi3.so → _optimizer_febdf5b_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ae60aac17486a756b1926d38e1c20933f57444688e15ba849da3153adcf434e
+oid sha256:8427dae3274100063f3b003a7cebf9565318fcaa2fa340482b2ec9408e9dcea0
 size 1749648

build/torch27-cxx11-rocm63-x86_64-linux/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)

torch-ext/optimizer/muon.py

@@ -5,11 +5,9 @@ import torch
 import torch.distributed as dist
 from torch.distributed._tensor import DTensor
 
-# TODO leave original url and consider LICENSE
-# This code snippet is a modified version adapted from the following GitHub repository:
-# https://github.com/KellerJordan/Muon/blob/master/muon.py
-
 
+# This code snippet is a modified version adapted from the following GitHub repositories:
+# https://github.com/KellerJordan/Muon/blob/master/muon.py
 @torch.no_grad()
 def _zeropower_via_newtonschulz5(G, steps):
     """
@@ -34,7 +32,7 @@ def _zeropower_via_newtonschulz5(G, steps):
         A = X @ X.T
         # B = (
         #    b * A + c * A @ A
-        # )  # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
+        # )
         B = torch.addmm(A, A, A, alpha=c, beta=b)
         # X = a * X + B @ X
         X = torch.addmm(X, B, X, alpha=1.0, beta=a)