address PR review

Author: Kotomi-Du

include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h

@@ -386,8 +386,6 @@ static const char* const kOrtSessionOptionsQDQMatMulNBitsAccuracyLevel = "sessio
 // "Efficient": OS treats this workload is efficiency oriented with low scheduling priority and efficient processor performance.
 static const char* const kOrtEpDynamicOptionsWorkloadType = "ep.dynamic.workload_type";
 
-static const char* const kOrtEpDynamicOptionsKVCacheRewind = "ep.llm.kvcache_rewind";
-
 // Disables model compilation during session initialization.
 //
 // If this option is set to "1", inference session creation will fail with error code ORT_MODEL_REQUIRES_COMPILATION

onnxruntime/core/providers/openvino/openvino_execution_provider.cc

@@ -277,7 +277,7 @@ common::Status OpenVINOExecutionProvider::SetEpDynamicOptions(gsl::span<const ch
           }
         }
       }
-    } else if (key == "kvcache_rewind" || key == kOrtEpDynamicOptionsKVCacheRewind) {
+    } else if (key == "kvcache_rewind") {
       // Convert kvcache_rewind value to int64_t
       int64_t index;
       try {

onnxruntime/core/providers/openvino/ov_interface.cc

@@ -457,11 +457,6 @@ void StatefulOVInferRequest::Infer() {
   OVInferRequest::Infer();
 }
 
-void OVInferRequest::RewindKVCache(size_t index) {
-  LOGS_DEFAULT(INFO) << log_tag << "RewindKVCache: Rewinding OpenVINO-internal KVCache state to index=" << index;
-  ovInfReq.reset_state();
-}
-
 void StatefulOVInferRequest::RewindKVCache(size_t index) {
   LOGS_DEFAULT(INFO) << log_tag << "RewindKVCache: Rewinding OpenVINO-internal KVCache state to index=" << index;
 

onnxruntime/core/providers/openvino/ov_interface.h

@@ -145,7 +145,7 @@ class OVInferRequest {
   ov::InferRequest& GetNewObj() {
     return ovInfReq;
   }
-  virtual void RewindKVCache([[maybe_unused]] size_t index);
+  virtual void RewindKVCache([[maybe_unused]] size_t index) {}
 };
 
 class StatefulOVInferRequest : public OVInferRequest {