diff --git a/include/glow/LLVMIRCodeGen/LLVMBackend.h b/include/glow/LLVMIRCodeGen/LLVMBackend.h
index 3ae28d32d8..5dbc84f983 100644
--- a/include/glow/LLVMIRCodeGen/LLVMBackend.h
+++ b/include/glow/LLVMIRCodeGen/LLVMBackend.h
@@ -44,7 +44,7 @@ class LLVMBackendOptions {
   /// ABI to be used by this backend.
   std::string abi_;
   /// Float ABI to be used by this backend.
-  llvm::Optional<llvm::FloatABI::ABIType> floatABI_;
+  std::optional<llvm::FloatABI::ABIType> floatABI_;
   /// Code model used by this backend.
   llvm::CodeModel::Model codeModel_;
   /// Code model used by this backend for bundles.
@@ -75,11 +75,11 @@ class LLVMBackendOptions {
   /// Sets ABI used by this backend.
   void setABIName(llvm::StringRef abi) { abi_ = abi.str(); }
   /// \returns Float ABI used by this backend.
-  llvm::Optional<llvm::FloatABI::ABIType> getFloatABI() const {
+  std::optional<llvm::FloatABI::ABIType> getFloatABI() const {
     return floatABI_;
   }
   /// Sets Float ABI used by this backend.
-  void setFloatABI(llvm::Optional<llvm::FloatABI::ABIType> floatABI) {
+  void setFloatABI(std::optional<llvm::FloatABI::ABIType> floatABI) {
     floatABI_ = floatABI;
   }
   /// \returns code model used by this backend.
diff --git a/include/glow/Optimizer/GraphOptimizer/CompilationContext.h b/include/glow/Optimizer/GraphOptimizer/CompilationContext.h
index 0c272ddbc9..b237c8bdee 100644
--- a/include/glow/Optimizer/GraphOptimizer/CompilationContext.h
+++ b/include/glow/Optimizer/GraphOptimizer/CompilationContext.h
@@ -21,6 +21,8 @@
 #include "glow/Quantization/Base/Base.h"
 #include "glow/Support/Error.h"
 
+#include <optional>
+
 namespace glow {
 namespace runtime {
 struct PartitionConfig;
@@ -273,7 +275,7 @@ struct OptimizationOptions {
   /// If it is true (false), perform (not perform) ASAP op placement in DAG
   /// optimization; If it is not set, use acc perf GFlag APLASAPPlacement to
   /// determine whether to perform ASAP op placement or not
-  llvm::Optional<bool> enableAPLASAPPlacement;
+  std::optional<bool> enableAPLASAPPlacement;
 
   /// If true does int64 to int32 type demotion if backend supports for specific
   /// nodes.
@@ -311,8 +313,8 @@ struct OptimizationOptions {
     PRINT_VALUE(foldElemKindConversionIntoIO, dump_str)
     PRINT_VALUE(foldStaticPlaceholderConversions, dump_str)
     PRINT_VALUE(useSparseNNPartitioningScheme, dump_str)
-    if (enableAPLASAPPlacement.hasValue()) {
-      PRINT_VALUE(enableAPLASAPPlacement.getValue(), dump_str)
+    if (enableAPLASAPPlacement) {
+      PRINT_VALUE(enableAPLASAPPlacement.value(), dump_str)
     }
     PRINT_VALUE(enableTypeDemotion, dump_str)
     PRINT_VALUE(enableQuantParamChanges, dump_str)
diff --git a/include/glow/Support/TensorPool.h b/include/glow/Support/TensorPool.h
index 59c0f6aeeb..1ccf31945b 100644
--- a/include/glow/Support/TensorPool.h
+++ b/include/glow/Support/TensorPool.h
@@ -17,11 +17,11 @@
 #define GLOW_TENSORPOOL_H
 
 #include "glow/Base/Tensor.h"
-#include "llvm/ADT/Optional.h"
 
 #include <atomic>
 #include <iostream>
 #include <mutex>
+#include <optional>
 #include <unordered_map>
 #include <vector>
 
@@ -74,7 +74,7 @@ class TensorPool final {
   /// previously been added by initialize. If the pool is empty this will
   /// allocate a new Tensor unless preventAllocs was set true at construction
   /// time.
-  llvm::Optional<Tensor> get(TypeRef ty);
+  std::optional<Tensor> get(TypeRef ty);
 
   /// Return a Tensor \p t to the pool. This Tensor must have been previously
   /// allocated by this TensorPool.
diff --git a/lib/Backends/Interpreter/InterpreterNodes.cpp b/lib/Backends/Interpreter/InterpreterNodes.cpp
index 4dad851c4f..8c3ed9030c 100644
--- a/lib/Backends/Interpreter/InterpreterNodes.cpp
+++ b/lib/Backends/Interpreter/InterpreterNodes.cpp
@@ -1476,7 +1476,7 @@ static void fwdMaxPool(Tensor *inW, Tensor *outW, Tensor *argmaxW,
   ShapeHW kdim(kernelSizes);
   ShapeHW sdim(strides);
 
-  llvm::Optional<Handle<int64_t>> argmaxH;
+  std::optional<Handle<int64_t>> argmaxH;
   if (argmaxW) {
     argmaxH = argmaxW->getHandle<int64_t>();
   }
@@ -6678,7 +6678,7 @@ void BoundInterpreterFunction::fwdIntNBitSplitEmbeddingWeightedBagsImpl(
   auto weightsTysH = weightsTys->getHandle<uint8_t>();
   auto dimOffsetsH = dimOffsets->getHandle<int32_t>();
   auto weightsOffsetsH = weightsOffsets->getHandle<WeightsOffsetTy>();
-  llvm::Optional<Handle<IndiceWeightTy>> indiceWeightsH;
+  std::optional<Handle<IndiceWeightTy>> indiceWeightsH;
   if (indiceWeights) {
     indiceWeightsH = indiceWeights->getHandle<IndiceWeightTy>();
   }
diff --git a/lib/CodeGen/MemoryAllocator.cpp b/lib/CodeGen/MemoryAllocator.cpp
index 4d707d08c4..ece975dfbd 100644
--- a/lib/CodeGen/MemoryAllocator.cpp
+++ b/lib/CodeGen/MemoryAllocator.cpp
@@ -20,6 +20,7 @@
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 
 #define DEBUG_TYPE "memory-allocator"
diff --git a/lib/Graph/Log.cpp b/lib/Graph/Log.cpp
index 810397d0ac..301b574f8a 100644
--- a/lib/Graph/Log.cpp
+++ b/lib/Graph/Log.cpp
@@ -20,6 +20,7 @@
 #include "glow/Graph/Graph.h"
 #include "glow/Graph/Node.h"
 #include "glow/Graph/NodeValue.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/FormatVariadic.h"
diff --git a/lib/LLVMIRCodeGen/LLVMIRGen.cpp b/lib/LLVMIRCodeGen/LLVMIRGen.cpp
index 20a1831c6b..8da1963558 100644
--- a/lib/LLVMIRCodeGen/LLVMIRGen.cpp
+++ b/lib/LLVMIRCodeGen/LLVMIRGen.cpp
@@ -99,8 +99,8 @@ static std::mutex initTargetMutex;
 
 void LLVMIRGen::initTargetOptions(llvm::TargetOptions &targetOpts,
                                   const LLVMBackendOptions &backendOpts) {
-  if (backendOpts.getFloatABI().hasValue()) {
-    targetOpts.FloatABIType = backendOpts.getFloatABI().getValue();
+  if (backendOpts.getFloatABI().has_value()) {
+    targetOpts.FloatABIType = backendOpts.getFloatABI().value();
   }
   if (!backendOpts.getABIName().empty()) {
     targetOpts.MCOptions.ABIName = backendOpts.getABIName();
diff --git a/lib/Onnxifi/Base.cpp b/lib/Onnxifi/Base.cpp
index 15f8358845..2b10449a22 100644
--- a/lib/Onnxifi/Base.cpp
+++ b/lib/Onnxifi/Base.cpp
@@ -347,15 +347,14 @@ onnxStatus Graph::adjustInputs(uint32_t inputsCount,
       continue;
     }
 
-    llvm::Optional<Tensor> inputTensorOpt = tensorPool_.get(inPhPtr->getType());
-    if (!inputTensorOpt.hasValue()) {
+    std::optional<Tensor> inputTensorOpt = tensorPool_.get(inPhPtr->getType());
+    if (!inputTensorOpt.has_value()) {
       DLOG(FATAL) << "Tensorpool tensor not found for input "
                   << inOnnxTensor.name;
       return ONNXIFI_STATUS_INTERNAL_ERROR;
     }
     // We want fresh DeviceResidencyInfo for this fresh Tensor.
-    externalIOBindings.emplace_back(inPhPtr,
-                                    std::move(inputTensorOpt.getValue()));
+    externalIOBindings.emplace_back(inPhPtr, std::move(inputTensorOpt.value()));
     Tensor &inputTensor = externalIOBindings.back().second;
     inputTensor.resetDeviceInfo();
 
diff --git a/lib/Optimizer/GraphOptimizer/GraphOptimizer.cpp b/lib/Optimizer/GraphOptimizer/GraphOptimizer.cpp
index d514c5d203..ee3548fe92 100644
--- a/lib/Optimizer/GraphOptimizer/GraphOptimizer.cpp
+++ b/lib/Optimizer/GraphOptimizer/GraphOptimizer.cpp
@@ -5474,9 +5474,9 @@ struct ChannelShuffleParams {
 /// as ReshapeNode->TransposeNode->ReshapeNode) for which \p node is the leading
 /// ReshapeNode. \returns The original ChannelShuffle parameters if possible and
 /// empty Optional otherwise.
-static llvm::Optional<ChannelShuffleParams>
+static std::optional<ChannelShuffleParams>
 getChannelShuffleParams(const ReshapeNode &node) {
-  auto resM = llvm::Optional<ChannelShuffleParams>();
+  std::optional<ChannelShuffleParams> resM;
 
   llvm::ArrayRef<dim_t> inputDims = node.getInput().dims();
   llvm::ArrayRef<dim_t> resultDims = node.getDims();
@@ -5539,7 +5539,7 @@ bool FoldChannelShuffle::run(Function *F, const CompilationContext &cctx) {
 
     // Compute the original parameters to ChannelShuffle.
     auto paramsM = getChannelShuffleParams(*RN1);
-    if (!paramsM.hasValue()) {
+    if (!paramsM.has_value()) {
       continue;
     }
 
diff --git a/lib/Runtime/HostManager/HostManager.cpp b/lib/Runtime/HostManager/HostManager.cpp
index f46e75a83e..bb4f30c2f0 100644
--- a/lib/Runtime/HostManager/HostManager.cpp
+++ b/lib/Runtime/HostManager/HostManager.cpp
@@ -901,7 +901,7 @@ Error HostManager::runNetworkBlocking(
 }
 
 void HostManager::dispatchNextRun() {
-  llvm::Optional<InferRequest> pRequest;
+  std::optional<InferRequest> pRequest;
   std::shared_lock<std::shared_timed_mutex> networkLock(networkLock_);
   {
     // hmm this lock is hot but I still have it as a unique lock because
@@ -921,8 +921,8 @@ void HostManager::dispatchNextRun() {
     }
   }
 
-  assert(pRequest.hasValue());
-  InferRequest request = std::move(pRequest.getValue());
+  assert(pRequest.has_value());
+  InferRequest request = std::move(pRequest.value());
   auto startTime = TraceEvent::now();
   auto requestReceived = request.startTime;
   executor_->run(
diff --git a/lib/Support/TensorPool/TensorPool.cpp b/lib/Support/TensorPool/TensorPool.cpp
index 36cd7cafe0..36646c11b2 100644
--- a/lib/Support/TensorPool/TensorPool.cpp
+++ b/lib/Support/TensorPool/TensorPool.cpp
@@ -16,9 +16,11 @@
 
 #include "glow/Support/TensorPool.h"
 
+#include <optional>
+
 namespace glow {
 
-llvm::Optional<Tensor> TensorPool::get(TypeRef ty) {
+std::optional<Tensor> TensorPool::get(TypeRef ty) {
   stats_.totalGets++;
 
   std::unique_lock<std::mutex> l(lock_);
@@ -27,7 +29,7 @@ llvm::Optional<Tensor> TensorPool::get(TypeRef ty) {
 
   if (it == pools_.end()) {
     if (preventInlineAllocs_) {
-      return llvm::Optional<Tensor>();
+      return std::nullopt;
     }
 
     stats_.totalTypes++;
@@ -36,7 +38,7 @@ llvm::Optional<Tensor> TensorPool::get(TypeRef ty) {
 
   if (it->second.empty()) {
     if (preventInlineAllocs_) {
-      return llvm::Optional<Tensor>();
+      return std::nullopt;
     }
 
     // Don't need to alloc under the lock.
diff --git a/tests/unittests/TensorPoolTest.cpp b/tests/unittests/TensorPoolTest.cpp
index 98c0f8c9da..ffd65b3604 100644
--- a/tests/unittests/TensorPoolTest.cpp
+++ b/tests/unittests/TensorPoolTest.cpp
@@ -31,7 +31,7 @@ TEST(TensorPool, BasicTest) {
   Type ty(ElemKind::FloatTy, {1, 2, 3});
   pool.reserve(&ty, 1);
 
-  Tensor T = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
   EXPECT_TRUE(T.getType().isEqual(ty));
   EXPECT_EQ(T.dims(), ty.dims());
 
@@ -52,12 +52,12 @@ TEST(TensorPool, ReclaimAndGet) {
   Type ty(ElemKind::FloatTy, {1, 2, 3});
   pool.reserve(&ty, 1);
 
-  Tensor T = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
   auto *backingPtr = T.getUnsafePtr();
 
   pool.reclaim(std::move(T));
 
-  Tensor T2 = std::move(pool.get(&ty).getValue());
+  Tensor T2 = std::move(pool.get(&ty).value());
   // They are the same buffer.
   EXPECT_EQ(T2.getUnsafePtr(), backingPtr);
 
@@ -78,8 +78,8 @@ TEST(TensorPool, Extends) {
   Type ty(ElemKind::FloatTy, {1, 2, 3});
   pool.reserve(&ty, 1);
 
-  Tensor T = std::move(pool.get(&ty).getValue());
-  Tensor T2 = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
+  Tensor T2 = std::move(pool.get(&ty).value());
   EXPECT_TRUE(T.getType().isEqual(T2.getType()));
   EXPECT_TRUE(T.getType().isEqual(ty));
   EXPECT_TRUE(T2.getType().isEqual(ty));
@@ -105,15 +105,15 @@ TEST(TensorPool, DoesntExtend) {
   Type ty(ElemKind::FloatTy, {1, 2, 3});
   pool.reserve(&ty, 1);
 
-  Tensor T = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
   Type Tt = T.getType();
 
   auto T2opt = pool.get(&ty);
-  EXPECT_FALSE(T2opt.hasValue());
+  EXPECT_FALSE(T2opt.has_value());
 
   pool.reclaim(std::move(T));
 
-  T = std::move(pool.get(&ty).getValue());
+  T = std::move(pool.get(&ty).value());
   EXPECT_EQ(Tt, T.getType());
 
   const auto &stats = pool.getStats();
@@ -132,8 +132,8 @@ TEST(TensorPool, Noreserve) {
   TensorPool pool;
   Type ty(ElemKind::FloatTy, {1, 2, 3});
 
-  Tensor T = std::move(pool.get(&ty).getValue());
-  Tensor T2 = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
+  Tensor T2 = std::move(pool.get(&ty).value());
 
   EXPECT_TRUE(T.getType().isEqual(T2.getType()));
 
@@ -162,8 +162,8 @@ TEST(TensorPool, MultipleTypes) {
   std::vector<Tensor> tensors;
   // Ten total allocs.
   for (int i = 0; i < 5; ++i) {
-    Tensor T = std::move(pool.get(&ty).getValue());
-    Tensor T2 = std::move(pool.get(&ty2).getValue());
+    Tensor T = std::move(pool.get(&ty).value());
+    Tensor T2 = std::move(pool.get(&ty2).value());
     EXPECT_FALSE(T.getType().isEqual(T2.getType()));
     EXPECT_TRUE(T.getType().isEqual(ty));
     EXPECT_TRUE(T2.getType().isEqual(ty2));
@@ -200,14 +200,14 @@ TEST(TensorPool, MultipleTypesReclaim) {
   pool.reserve(&ty, 1);
   pool.reserve(&ty2, 1);
 
-  Tensor T = std::move(pool.get(&ty).getValue());
-  Tensor T2 = std::move(pool.get(&ty2).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
+  Tensor T2 = std::move(pool.get(&ty2).value());
 
   pool.reclaim(std::move(T));
   pool.reclaim(std::move(T2));
 
-  T = std::move(pool.get(&ty).getValue());
-  T2 = std::move(pool.get(&ty2).getValue());
+  T = std::move(pool.get(&ty).value());
+  T2 = std::move(pool.get(&ty2).value());
 
   pool.reclaim(std::move(T));
   pool.reclaim(std::move(T2));
@@ -231,7 +231,7 @@ TEST(TensorPool, PlaceholderBindingsReclaim) {
   Module mod;
 
   auto *PH = mod.createPlaceholder(&ty, "test", false);
-  bindings.insert(PH, std::move(pool.get(&ty).getValue()));
+  bindings.insert(PH, std::move(pool.get(&ty).value()));
 
   /// Insert a non managed tensor.
   auto *PH2 = mod.createPlaceholder(&ty, "test2", false);
@@ -249,7 +249,7 @@ TEST(TensorPool, PlaceholderBindingsReclaim) {
   EXPECT_EQ(stats.totalGets, 1);
   EXPECT_EQ(stats.totalReclaims, 1);
 
-  bindings.insert(PH, std::move(pool.get(&ty).getValue()));
+  bindings.insert(PH, std::move(pool.get(&ty).value()));
 
   bindings.erase(PH);
   const auto &stats2 = pool.getStats();
@@ -263,7 +263,7 @@ TEST(TensorPool, Clear) {
   TensorPool pool;
   Type ty(ElemKind::FloatTy, {1, 2, 3});
 
-  Tensor T = std::move(pool.get(&ty).getValue());
+  Tensor T = std::move(pool.get(&ty).value());
   pool.reclaim(std::move(T));
 
   const auto &stats = pool.getStats();
@@ -277,7 +277,7 @@ TEST(TensorPool, Clear) {
 
   pool.clear();
 
-  T = std::move(pool.get(&ty).getValue());
+  T = std::move(pool.get(&ty).value());
   pool.reclaim(std::move(T));
 
   const auto &stats2 = pool.getStats();
diff --git a/torch_glow/src/CachingGraphRunner.cpp b/torch_glow/src/CachingGraphRunner.cpp
index 1c07876e48..57737801e0 100644
--- a/torch_glow/src/CachingGraphRunner.cpp
+++ b/torch_glow/src/CachingGraphRunner.cpp
@@ -913,13 +913,13 @@ CachingGraphRunner::convertPyTorchInputToGlowInput(
     // For backends that does not support partial tensor, last-element padding
     // based on size
     auto inputTensorOpt = tensorPool_.get(ty);
-    if (!inputTensorOpt.hasValue()) {
+    if (!inputTensorOpt) {
       std::stringstream ss;
       ss << "Tensorpool tensor not found for input " << ptTensor.name();
       return MAKE_ERR(ss.str());
     }
     // We want fresh DeviceResidencyInfo for this fresh Tensor.
-    glow::Tensor inputTensor(std::move(inputTensorOpt.getValue()));
+    glow::Tensor inputTensor(std::move(inputTensorOpt.value()));
     inputTensor.resetDeviceInfo();
     if (ptTensor.data_ptr()) {
       auto *inTensorPtr = inputTensor.getUnsafePtr();