Merge pull request #27 from MotifTechnologies/fix/cudagraph-offload-conflict

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch210-cxx11-cu126-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7be82307f66be4bb841072ecdb3d105dc73bc9ee9ca21b1ce33bddc24113f4d1
+oid sha256:a45eb1ca39a9122a7ec8171ecc1ed8d46b0bd6ec6fcf66e1f5635f2ba8b946ce
 size 1940944

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

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch210-cxx11-cu128-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fa10afe7c505f69ccf7a98aca116b6d551b9577ecce2dab2559c6c3b433be20
+oid sha256:63fcd4e511d075431c2e7e2fefbb96dcced540e2c0ca1316faf29fdfd78e253a
 size 2004144

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

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch210-cxx11-cu130-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f72217eb59ce93935f593b0fdfc7f3bfc4e05f18ad9d5384c2325b27ad7ff136
+oid sha256:62348f4dd6a65cfc98cf101dff522ad8ae88c56ee99fb7cfed2c33b3222262d8
 size 2004728

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

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch210-cxx11-rocm70-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df63b7dafe62fa5e910de1123729cbe3496015e2c0110785d9bd510bf65c2eaa
+oid sha256:c9d1940e448de8c2290bb3142a1d41433731dba06c6d68738f974af26cb4f240
 size 1866400

build/torch210-cxx11-rocm70-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch210-cxx11-rocm71-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7215ee4575fa44f482a98e563c5af2d60089e36d32fe8a3dcffe3fb5f587300f
+oid sha256:1eb0a09bd1b20303563fe0cb3e8e2234dae40241e3d0d56502fc5a6ae442588f
 size 1866112

build/torch210-cxx11-rocm71-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch28-cxx11-cu126-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37bcda6973440cdeb880e411dbaf12220cef0bab18299b4922b6a504ab109b42
+oid sha256:14140b8568b7a04120bf30a13f95a5dfa44255d7d107c0330b61ce92e79404d0
 size 1936664

build/torch28-cxx11-cu126-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch28-cxx11-cu128-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9cd35014d41034ed35fbad31c19c80e7b3977cea889a865eb12db705678bb29
+oid sha256:b8a8e75d22105a30822529b34963a13a0a60a20a862bc7da971857f54008105c
 size 1999872

build/torch28-cxx11-cu128-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch28-cxx11-cu129-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64072037c62afbfefe92a35b026794f2bd406fddef38bf58d318d3bae7652a29
+oid sha256:69697c5e77ccd05e26a3eca114732d905f9c3b8c68d335b13f129c56a710c7a3
 size 1999872

build/torch28-cxx11-cu129-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch28-cxx11-rocm63-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65135fe756ed97f2bd21fefda883b6a7b90179ebd7c0a882673239daf9d9aa6a
+oid sha256:01a9dda5ac371798b9fd86b77b8cf7b572721a423ee69971910abe69a62d88fd
 size 1865080

build/torch28-cxx11-rocm63-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

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

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch28-cxx11-rocm64-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c562a3b7e6f3032ff56473531e9e08fceb2c86f8804080330c896ab8f0dd32af
+oid sha256:c1fb5220ced504ce27281ceb69fc21ce69de51baad5485a67973bb6be489aeb9
 size 1865168

build/torch28-cxx11-rocm64-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

build/torch29-cxx11-cu126-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch29-cxx11-cu126-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bfd2c3e3de0e507b209e24959bd89d131dfcf82136e07f078796fae5812b7fed
+oid sha256:930e785c6e2e19242fca7a43c1384693552777d4e0d07ba52558d1212bc158fd
 size 1936664

build/torch29-cxx11-cu126-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

build/torch29-cxx11-cu128-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch29-cxx11-cu128-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd4f09efe7fdf7a1cb7a6dce7ee903a6eef7ad8fc38936cd16baf40ec26a444b
+oid sha256:b49a73f9b90d9962a9bf859d85678dfd4f042762a561287098c30af2786779fd
 size 1999872

build/torch29-cxx11-cu128-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

build/torch29-cxx11-cu130-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch29-cxx11-cu130-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58df65218cd174d1e2754b9ce325309042d22f7ed13fc9bd4fbc07037bb9db80
+oid sha256:818efe7b03956810294c938e701163f7c06285e72631a5fb78b4f82eba72d8a6
 size 2000456

build/torch29-cxx11-cu130-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

build/torch29-cxx11-rocm63-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch29-cxx11-rocm63-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:49e1841486948932d0cafd1c53a207c5864bf23d05ed46039e20f9f7b6f81363
+oid sha256:51e9488c2003b603f0c9f09c03e7ddfb576074f2fdc6d94363e7dfd07f5a7111
 size 1865112

build/torch29-cxx11-rocm63-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

build/torch29-cxx11-rocm64-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _optimizer_b68ea5b_dirty
-ops = torch.ops._optimizer_b68ea5b_dirty
+from . import _optimizer_2dce952_dirty
+ops = torch.ops._optimizer_2dce952_dirty
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_optimizer_b68ea5b_dirty::{op_name}"
+    return f"_optimizer_2dce952_dirty::{op_name}"

build/torch29-cxx11-rocm64-x86_64-linux/{_optimizer_b68ea5b_dirty.abi3.so → _optimizer_2dce952_dirty.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b32fcd009415d933331735d3771fc2d56e9b67a5ed1772cd8392d201e68166b5
+oid sha256:887bca38f051d28d35af84921d78943d2eff7a3d86a48c0f29f0e908ae577d0c
 size 1865232

build/torch29-cxx11-rocm64-x86_64-linux/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()

torch-ext/optimizer/newton_schulz.py

@@ -213,10 +213,8 @@ def zeropower_via_newtonschulz5(G, steps=5):
     if key not in _ns_per_shape:
         _ns_per_shape[key] = torch.compile(_zeropower_via_newtonschulz5,
                                            options={
-                                               "triton.cudagraphs": True,
                                                "shape_padding": False
                                            })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()
 
 
@@ -229,8 +227,6 @@ def zeropower_via_newtonschulz5_batched(G, steps=5):
         _ns_per_shape[key] = torch.compile(
             _zeropower_via_newtonschulz5_batched,
             options={
-                "triton.cudagraphs": True,
                 "shape_padding": False
             })
-    torch.compiler.cudagraph_mark_step_begin()
     return _ns_per_shape[key](G, steps).clone()