linux build fixes

src/models/debugging.cpp

@@ -1,5 +1,6 @@
 #include "../generators.h"
 #include "debugging.h"
+#include <inttypes.h>
 
 namespace Generators {
 static constexpr size_t c_value_count = 10; // Dump this many values from the start of a tensor
@@ -26,7 +27,7 @@ void DumpValues(ONNXTensorElementDataType type, const void* p_values_raw, size_t
     case Ort::TypeToTensorType<int64_t>::type: {
       auto* p_values = reinterpret_cast<const int64_t*>(p_values_raw);
       for (size_t i = 0; i < count; i++) {
-        printf("%lld ", p_values[i]);
+        printf("%" PRId64 " ", p_values[i]);
       }
       break;
     }

src/models/input_ids.cpp

@@ -67,7 +67,7 @@ void InputIDs<T>::Update(RoamingArray<int32_t> next_tokens_unk) {
   }
 }
 
-template InputIDs<int32_t>;
-template InputIDs<int64_t>;
+template struct InputIDs<int32_t>;
+template struct InputIDs<int64_t>;
 
 }  // namespace Generators

src/models/kv_cache.cpp

@@ -64,7 +64,7 @@ void KV_Cache_Combined::PickPastState(std::span<const int32_t> beam_indices, int
   auto element_count = shape_[0] * past_key_size;
 
   const OrtValue& present = *presents_[index];
-  auto past = OrtValue::CreateTensor<ScoreType>(*model_.allocator_device_, shape_);
+  std::unique_ptr<OrtValue> past = OrtValue::CreateTensor<ScoreType>(*model_.allocator_device_, shape_);
   auto past_span = std::span<ScoreType>(past->GetTensorMutableData<ScoreType>(), element_count);
   auto present_span = std::span<const ScoreType>(present.GetTensorData<ScoreType>(), element_count);
 
@@ -179,7 +179,7 @@ void KV_Cache::PickPastState(std::span<const int32_t> beam_indices, int index) {
   auto element_count = shape_[0] * block_size_per_beam;
 
   const OrtValue& present_value = *presents_[index];
-  auto past_value = OrtValue::CreateTensor<ScoreType>(*model_.allocator_device_, shape_);
+  std::unique_ptr<OrtValue> past_value = OrtValue::CreateTensor<ScoreType>(*model_.allocator_device_, shape_);
   auto past_span = std::span<ScoreType>(past_value->GetTensorMutableData<ScoreType>(), element_count);
   auto present_span = std::span<const ScoreType>(present_value.GetTensorData<ScoreType>(), element_count);
 

src/models/position_ids.cpp

@@ -100,7 +100,7 @@ void PositionIDs<T>::Update(int current_length) {
     attention_mask_shape_[1] = current_length;
 
     const auto* old_data = attention_mask_->GetTensorData<T>();
-    auto attention_mask = OrtValue::CreateTensor<T>(*model_.allocator_device_, attention_mask_shape_);
+    std::unique_ptr<OrtValue> attention_mask = OrtValue::CreateTensor<T>(*model_.allocator_device_, attention_mask_shape_);
     auto* data = attention_mask->GetTensorMutableData<T>();
 
     switch (model_.device_type_) {
@@ -125,7 +125,7 @@ void PositionIDs<T>::Update(int current_length) {
   }
 }
 
-template PositionIDs<int32_t>;
-template PositionIDs<int64_t>;
+template struct PositionIDs<int32_t>;
+template struct PositionIDs<int64_t>;
 
 }  // namespace Generators

src/ort_genai_c.h

@@ -1,6 +1,7 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 #include <stdint.h>
+#include <cstddef>
 
 #ifdef __cplusplus
 extern "C" {