diff --git a/pyproject.toml b/pyproject.toml
index 927cf7612c..28ed380266 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -326,14 +326,24 @@ cu130onlytorch291 = [
     "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.33.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
 ]
 cu126onlytorch2100 = [
+    # The explicit torch pin replicates the xformers 0.0.34 wheel's transitive
+    # torch==2.10.0 pin on machines where the x86-64-only wheel is skipped
+    # (Linux aarch64 has real 2.10.0 wheels; Windows ARM64 fails loudly).
+    # Without it an ARM64 install of the cu126-torch2100 wrapper would proceed
+    # unpinned and resolve a newer torch.
+    "torch==2.10.0",
     "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
     "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch2100 = [
+    # Same explicit torch pin as cu126onlytorch2100 above.
+    "torch==2.10.0",
     "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
     "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu130onlytorch2100 = [
+    # Same explicit torch pin as cu126onlytorch2100 above.
+    "torch==2.10.0",
     "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
     "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
diff --git a/tests/test_torch2110_cuda_extras.py b/tests/test_torch2110_cuda_extras.py
index b9524d6e4b..19d40560f4 100644
--- a/tests/test_torch2110_cuda_extras.py
+++ b/tests/test_torch2110_cuda_extras.py
@@ -86,3 +86,19 @@ def test_torch2110_wrapper_references_matching_leaf(cuda: str, variant: str):
         "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
         f"unsloth[{cuda}onlytorch2110]",
     ]
+
+
+@pytest.mark.parametrize("cuda", ["cu126", "cu128", "cu130"])
+def test_cuda12_torch2100_keeps_torch_pinned_off_x86(cuda: str):
+    # The torch2100 leaves used to rely on the xformers 0.0.34 wheel's transitive
+    # torch==2.10.0 pin. Now that the x86-64-only wheels carry platform_machine
+    # markers, the leaf must pin torch explicitly so an ARM64 install stays on
+    # torch 2.10 (Linux aarch64 wheels exist) or fails loudly (Windows ARM64)
+    # instead of resolving an unpinned newer torch.
+    reqs = _reqs(_extra(f"{cuda}onlytorch2100"))
+    (torch_req,) = reqs["torch"]
+    assert str(torch_req.specifier) == "==2.10.0", (
+        f"{cuda}onlytorch2100 must pin torch==2.10.0 for machines where the "
+        f"x86-64-only xformers wheel (and its transitive pin) is skipped"
+    )
+    assert torch_req.marker is None, "the torch pin must apply on every machine"