NVIDIA · naoyam · Jan 30, 2026 · Jan 30, 2026 · Jan 30, 2026
diff --git a/csrc/device_lower/id_model_options.h b/csrc/device_lower/id_model_options.h
@@ -16,9 +16,6 @@ namespace nvfuser {
 
 class IdModelOptions {
  public:
-  IdModelOptions()
-      : tensor_indexer_enabled_(isOptionEnabled(EnableOption::IdModel)) {}
-
   void setTensorIndexer(bool b) {
     tensor_indexer_enabled_ = b;
   }
@@ -36,8 +33,8 @@ class IdModelOptions {
   }
 
  private:
-  // Enable TensorIndexer
-  bool tensor_indexer_enabled_ = false;
+  // Enable TensorIndexer by default
+  bool tensor_indexer_enabled_ = true;
 };
 
 } // namespace nvfuser
diff --git a/csrc/device_lower/lower2device.cpp b/csrc/device_lower/lower2device.cpp
@@ -295,7 +295,6 @@ namespace {
 // given Fusion
 IdModelOptions getIdModelOptions(Fusion* fusion) {
   IdModelOptions options;
-  options.setTensorIndexer(true);
 
   // If not supported, disable use of TensorIndexer by default. It is
   // still used if explicitly opted-in (see, for example,

diff --git a/csrc/options.cpp b/csrc/options.cpp
@@ -162,7 +162,6 @@ const std::unordered_map<std::string, EnableOption>& getEnableOptions() {
           {"cutlass_scheduler", EnableOption::CutlassScheduler},
           {"fuse_matmul", EnableOption::FuseMatmul},
           {"fuse_multiple_matmuls", EnableOption::FuseMultipleMatmuls},
-          {"id_model", EnableOption::IdModel},
           {"id_model_extra_validation", EnableOption::IdModelExtraValidation},
           {"io_to_lower_precision", EnableOption::IoToLowerPrecision},
           {"kernel_db", EnableOption::KernelDb},

diff --git a/csrc/options.h b/csrc/options.h
@@ -103,7 +103,6 @@ enum class EnableOption {
   CutlassScheduler, //! Enable the CUTLASS scheduler and executor
   FuseMatmul, //! Enable automatic fusion of matmul and linear ops
   FuseMultipleMatmuls, //! Allow fusing more than one matmul in a single kernel
-  IdModel, //! Enable IdModel
   IdModelExtraValidation, //! Enable extra error checking when building IdModel
   IoToLowerPrecision, //! Enable castInputOutputToLowerPrecision. #1889 explains
                       //! why we disabled it by default.

diff --git a/python/nvfuser_direct/__init__.py b/python/nvfuser_direct/__init__.py
@@ -367,14 +367,11 @@ def execute(
             # Delete the _fusion and reference the fusion inside FusionExecutorCache
             del self._fusion
 
-        # Add "id_model" as a default enable option
-        default_enable_options = ["id_model"]
-        merged_enable_options = default_enable_options + _enable_options
-
+        # Note: id_model is now enabled by default in C++, so no need to add it here
         return self.fec.execute(
             inputs,
             device=self._get_device_index(device),
-            _enable_options=merged_enable_options,
+            _enable_options=_enable_options,
             _disable_options=_disable_options,
         )
 

diff --git a/tests/cpp/test_abstract_tensor.cpp b/tests/cpp/test_abstract_tensor.cpp
@@ -20,7 +20,6 @@ class AbstractTensorTest : public NVFuserTest {
 
   void SetUp() override {
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     fusion_ptr_ = std::make_unique<Fusion>();
     fusion_guard_ptr_ = std::make_unique<FusionGuard>(fusion_ptr_.get());
     auto size = IrBuilder::create<Val>(16, DataType::Index);

diff --git a/tests/cpp/test_allocation_domain.cpp b/tests/cpp/test_allocation_domain.cpp
@@ -21,13 +21,7 @@
 
 namespace nvfuser {
 
-class AllocationDomainTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using AllocationDomainTest = NVFuserTest;
 
 using ::testing::ElementsAre;
 

diff --git a/tests/cpp/test_allocation_order_inference.cpp b/tests/cpp/test_allocation_order_inference.cpp
@@ -23,13 +23,7 @@ namespace nvfuser {
 
 using testing::ElementsAre;
 
-class AllocationOrderInferenceTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using AllocationOrderInferenceTest = NVFuserTest;
 
 std::vector<int64_t> getAllocationOrder(TensorView* tv) {
   std::optional<std::vector<int64_t>> permutation =

diff --git a/tests/cpp/test_argsort.cpp b/tests/cpp/test_argsort.cpp
@@ -22,13 +22,7 @@
 
 namespace nvfuser {
 
-class ArgsortTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using ArgsortTest = NVFuserTest;
 
 // Parameterized test fixture for BasicExecution with different data types
 class ArgsortTestBasicExecution

diff --git a/tests/cpp/test_bfs.cpp b/tests/cpp/test_bfs.cpp
@@ -865,7 +865,6 @@ TEST_F(FindAllExprsTest, Rotation) {
   std::vector<int64_t> shape({16, 100});
 
   EnableOptionsGuard enable_options_guard;
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
 
   auto tv0 = makeConcreteTensor(shape);
   fusion.addInput(tv0);

diff --git a/tests/cpp/test_circular_buffering.cpp b/tests/cpp/test_circular_buffering.cpp
@@ -22,8 +22,6 @@ TEST_F(NVFuserTest, BarSyncWarpSpecializedPointwise) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   int64_t number_of_stages = 4;
   int64_t prefetch_distance = 1;
   int64_t tensor_outer_dim = 128;
@@ -99,8 +97,6 @@ TEST_F(NVFuserTest, RegisterSharingCircularBufferingPointwiseCustom) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   int64_t number_of_stages = 4;
   int64_t prefetch_distance = 1;
   int64_t tensor_outer_dim = 128;
@@ -183,8 +179,6 @@ TEST_F(NVFuserTest, RegisterSharingCircularBufferingPointwiseNested) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   int64_t number_of_stages = 4;
   int64_t prefetch_distance = 1;
   int64_t tensor_outer_dim = 128;
@@ -267,7 +261,6 @@ class CircularBufferingTest : public NVFuserFixtureParamTest<StageAndPrefetch> {
     number_of_stages = std::get<0>(GetParam());
     prefetch_distance = std::get<1>(GetParam());
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 };
 
@@ -1101,7 +1094,6 @@ class TmaCircularBufferingTest
     // NOTE: Multiple of 16 required for inner dimension
     NVF_ERROR(tensor_inner_dim % 16 == 0);
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 
   bool testEnablesWarpSpecialization() {
@@ -1220,8 +1212,6 @@ TEST_F(NVFuserTest, ElectSyncCompatibility) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   TensorView* input = makeContigTensor(3);
   fusion->addInput(input);
   TensorView* output = set(input);
@@ -2414,8 +2404,6 @@ TEST_F(NVFuserTest, TmaRegisterSharingDynamicShapesExpectFail) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   auto tv0 = makeContigTensor(2);
   fusion->addInput(tv0);
 
@@ -2478,8 +2466,6 @@ TEST_P(TmaRegisterSharing, CtaShapeShmoo) {
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();
   FusionGuard fg(fusion.get());
 
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   auto tv0 = makeContigTensor(2);
   fusion->addInput(tv0);
 

diff --git a/tests/cpp/test_circular_buffering_ping_pong.cpp b/tests/cpp/test_circular_buffering_ping_pong.cpp
@@ -331,7 +331,6 @@ TEST_P(SiblingPingPongCircularBuffering, TwoTmaLoads) {
 
   auto [use_id_model, stage_slice_position] = GetParam();
   if (use_id_model) {
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 
   std::unique_ptr<Fusion> fusion = std::make_unique<Fusion>();

diff --git a/tests/cpp/test_cluster.cpp b/tests/cpp/test_cluster.cpp
@@ -31,7 +31,6 @@ class ClusterReductionTest : public NVFuserTest,
  protected:
   void SetUp() override {
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     NVFUSER_TEST_CUDA_ARCH_GUARD(9, 0);
   }
 };

diff --git a/tests/cpp/test_combined_inner_outer_reduction.cpp b/tests/cpp/test_combined_inner_outer_reduction.cpp
@@ -45,7 +45,6 @@ class CombinedSchedulerTest
  protected:
   void SetUp() override {
     NVFuserFixtureParamTest<CombinedSchedulerParams>::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 };
 

diff --git a/tests/cpp/test_compute_with.cpp b/tests/cpp/test_compute_with.cpp
@@ -53,13 +53,7 @@ namespace nvfuser {
 
 using namespace at::indexing;
 
-class ComputeWithTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-    NVFuserTest::SetUp();
-  }
-};
+using ComputeWithTest = NVFuserTest;
 
 namespace {
 

diff --git a/tests/cpp/test_contiguity_id_model.cpp b/tests/cpp/test_contiguity_id_model.cpp
@@ -26,7 +26,6 @@ class ContigIDGroupsTest : public NVFuserTest {
  protected:
   void SetUp() override {
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 
   // Helper to construct a simple backward path for testing

diff --git a/tests/cpp/test_evaluator.cpp b/tests/cpp/test_evaluator.cpp
@@ -17,7 +17,7 @@
 
 namespace nvfuser {
 
-class ExprEvalTest : public NVFuserTest {};
+using ExprEvalTest = NVFuserTest;
 
 using ::testing::ElementsAre;
 using ::testing::HasSubstr;

diff --git a/tests/cpp/test_external_src.cpp b/tests/cpp/test_external_src.cpp
@@ -19,7 +19,7 @@
 
 namespace nvfuser {
 
-class ExternalSrcExample : public NVFuserTest {};
+using ExternalSrcExample = NVFuserTest;
 
 // This is for internal testing only and is intended to be used as a template to
 // compile and run an external source file. By default, it should just

diff --git a/tests/cpp/test_gather.cpp b/tests/cpp/test_gather.cpp
@@ -30,7 +30,6 @@ class GatherTest : public NVFuserTest {
   void SetUp() override {
     // To make the tests using std::rand deterministic
     std::srand(0);
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
   }
 };
 

diff --git a/tests/cpp/test_gpu1.cpp b/tests/cpp/test_gpu1.cpp
@@ -55,13 +55,7 @@ namespace nvfuser {
 
 using namespace at::indexing;
 
-class Gpu1Test : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using Gpu1Test = NVFuserTest;
 
 // A few smoke tests for IrGraphGenerator
 // (These tests exercise IrGraphGenerator through a non-trivial IR,
@@ -2803,7 +2797,6 @@ class Fp4CastTest : public NVFuserTest,
  public:
   void SetUp() override {
     NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     std::tie(dtype_highp, vectorization_factor) = GetParam();
     NVFUSER_TEST_CUDA_ARCH_GUARD(10, 0);
   }
@@ -3008,7 +3001,6 @@ class AdvancedDtypeTest : public NVFuserFixtureParamTest<bool> {
   bool use_dynamic_shape;
   void SetUp() override {
     NVFuserFixtureParamTest<bool>::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     use_dynamic_shape = GetParam();
   }
 };
@@ -3086,7 +3078,6 @@ class Float4E2m1ManualScheduleTestAllArch
   bool dynamic_shape;
   void SetUp() override {
     NVFuserFixtureParamTest<Float4E2m1ManualScheduleTestParams>::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     std::tie(vectorize_factor, dynamic_shape) = GetParam();
   }
 };

diff --git a/tests/cpp/test_gpu2.cpp b/tests/cpp/test_gpu2.cpp
@@ -56,13 +56,7 @@ namespace nvfuser {
 
 using namespace at::indexing;
 
-class Gpu2Test : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using Gpu2Test = NVFuserTest;
 
 TEST_F(Gpu2Test, FusionGlobalIntermediate_CUDA) {
   Fusion fusion;

diff --git a/tests/cpp/test_gpu3.cpp b/tests/cpp/test_gpu3.cpp
@@ -61,13 +61,7 @@ namespace nvfuser {
 
 using namespace at::indexing;
 
-class Gpu3Test : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using Gpu3Test = NVFuserTest;
 
 TEST_F(Gpu3Test, FusionNonDivisibleSplit1_CUDA) {
   Fusion fusion;
@@ -3400,8 +3394,6 @@ TEST_F(Gpu3Test, FusionIssueRepro1844_CUDA) {
 }
 
 TEST_F(Gpu3Test, FusionInsertMagicZero1_CUDA) {
-  EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-
   Fusion fusion;
   FusionGuard fg(&fusion);
 

diff --git a/tests/cpp/test_greedy.cpp b/tests/cpp/test_greedy.cpp
@@ -23,13 +23,7 @@
 
 namespace nvfuser {
 
-class GreedySchedulerTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    NVFuserTest::SetUp();
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-  }
-};
+using GreedySchedulerTest = NVFuserTest;
 
 class GreedySchedulerTestConstraintSize
     : public GreedySchedulerTest,

diff --git a/tests/cpp/test_index_put.cpp b/tests/cpp/test_index_put.cpp
@@ -42,7 +42,6 @@ std::vector<SizeParams> generateSizeOneParams() {
 class IndexPut : public NVFuserFixtureParamTest<SizeParams> {
  protected:
   void SetUp() override {
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
     NVFuserTest::SetUp();
   }
 };

diff --git a/tests/cpp/test_index_select.cpp b/tests/cpp/test_index_select.cpp
@@ -71,13 +71,7 @@ void checkIndexSelectVectorization(
 
 } // namespace
 
-class IndexSelectTest : public NVFuserTest {
- protected:
-  void SetUp() override {
-    EnableOptionsGuard::getCurOptions().set(EnableOption::IdModel);
-    NVFuserTest::SetUp();
-  }
-};
+using IndexSelectTest = NVFuserTest;
 
 TEST_F(IndexSelectTest, Simple1) {
   for (int i = 0; i < 5; ++i) {