diff --git a/imperative/python/megengine/dtr/dtr.py b/imperative/python/megengine/dtr/dtr.py
index f19c0d2a054d4dd58399c856da08669440b27498..382bbec38c9ea23716f3174ea562290c2c43aec3 100644
--- a/imperative/python/megengine/dtr/dtr.py
+++ b/imperative/python/megengine/dtr/dtr.py
@@ -9,6 +9,7 @@
 import re
 from typing import Union
 
+from ..core._imperative_rt.core2 import clear_candidates as _clear_candidates
 from ..core._imperative_rt.core2 import set_option as _set_option
 
 _eviction_threshold = 0
@@ -128,3 +129,4 @@ def disable():
     _set_option("enable_dtr_auto_drop", 0)
     _set_option("enable_drop", 0)
     _set_option("record_computing_path", 0)
+    _clear_candidates()
diff --git a/imperative/python/src/tensor.cpp b/imperative/python/src/tensor.cpp
index 6a90f3c829e917404d28b9c9ce1bcf755d7e11d7..87f2459d5b7f3b3a683573996429302c191e47e5 100644
--- a/imperative/python/src/tensor.cpp
+++ b/imperative/python/src/tensor.cpp
@@ -1020,6 +1020,7 @@ void init_tensor(py::module m) {
     m.def("set_option", [](std::string name, size_t value) {
         interpreter_for_py->set_option(name, value);
     });
+    m.def("clear_candidates", []() { interpreter_for_py->clear_candidates(); });
     m.def("get_option",
           [](std::string name) { return interpreter_for_py->get_option(name); });
     m.def("_set_drop_flag",
diff --git a/imperative/src/impl/interpreter/interpreter_impl.cpp b/imperative/src/impl/interpreter/interpreter_impl.cpp
index f3f98dc3bea52adb1aa93ff382096597089279b3..d80f23e7bcd9f6188ad2cc72c2176a40c630b963 100644
--- a/imperative/src/impl/interpreter/interpreter_impl.cpp
+++ b/imperative/src/impl/interpreter/interpreter_impl.cpp
@@ -494,6 +494,12 @@ void ChannelImpl::set_option(std::string name, size_t value) {
     m_buffer.enqueue(SetOption{name, value});
 }
 
+void ChannelImpl::clear_candidates() {
+    MGB_LOCK_GUARD(m_spin);
+    mgb_assert(check_available(), "Channel already closed");
+    m_dtr.candidates.clear();
+}
+
 TensorInfo* ChannelImpl::alloc() {
     auto& state = get_channel_state();
     auto info = [this] {
@@ -798,7 +804,7 @@ void ChannelImpl::do_apply_op(const ApplyOp& cmd, std::string reason) {
                 i->compute_time = estimate_compute_time;
             }
         }
-        m_dtr.unpin(cmd.inputs);
+        m_dtr.unpin(cmd.inputs, state);
     }
     MGB_RECORD_EVENT(OpExecuteFinishEvent, apply_id, {}, reason);
     // End profiling operator
@@ -1430,12 +1436,19 @@ void ChannelImpl::sample_on_device(CompNode device, bool force) {
 void ChannelImpl::DynamicSublinear::pin(const SmallVector<TensorInfo*>& vec) {
     for (auto i : vec) {
         i->pin();
+        erase_candidate(i);
     }
 }
 
-void ChannelImpl::DynamicSublinear::unpin(const SmallVector<TensorInfo*>& vec) {
+void ChannelImpl::DynamicSublinear::unpin(
+        const SmallVector<TensorInfo*>& vec, WorkerState& state) {
     for (auto i : vec) {
         i->unpin();
+        if (i->pinned == 0 &&
+            i->size_exceeds_thd(state.options.dtr_evictee_minimum_size) &&
+            i->cand_index == UINT_MAX) {
+            insert_candidate(i);
+        }
     }
 }
 
@@ -1504,7 +1517,7 @@ TensorInfo* ChannelImpl::DynamicSublinear::find_best_tensor(
             ti = vi;
         }
         auto i = candidates[ti];
-        if (i->producer && i->ptr && !i->pinned && i->evict_type == EvictType::NONE) {
+        if (i->producer && i->ptr && i->evict_type == EvictType::NONE) {
             double neighbor_cost = estimate_neighbor_cost(i);
             size_t begin_ptr =
                     reinterpret_cast<size_t>(i->ptr->blob()->storage().get());
@@ -1561,7 +1574,12 @@ void ChannelImpl::DynamicSublinear::insert_candidate(TensorInfo* ptr) {
 }
 
 void ChannelImpl::DynamicSublinear::erase_candidate(TensorInfo* ptr) {
-    // some tensors may be erased already, so just skip them
+    // close dtr will just clear candidates, so nothing to erase
+    if (candidates.empty()) {
+        ptr->cand_index = UINT_MAX;
+        return;
+    }
+    // some tensors may be erased already, just skip them
     if (ptr->cand_index != UINT_MAX) {
         std::swap(candidates[ptr->cand_index], candidates.back());
         candidates[ptr->cand_index]->cand_index = ptr->cand_index;
diff --git a/imperative/src/impl/interpreter/interpreter_impl.h b/imperative/src/impl/interpreter/interpreter_impl.h
index a786130b18028e2e0d9ff8abf1ca49d0c272001d..0b480205d50db3a196a0135000dfbd5b75a41f8d 100644
--- a/imperative/src/impl/interpreter/interpreter_impl.h
+++ b/imperative/src/impl/interpreter/interpreter_impl.h
@@ -64,6 +64,7 @@ struct ChannelImpl : Interpreter::Channel {
 
     size_t get_option(std::string name) override;
     void set_option(std::string name, size_t value) override;
+    void clear_candidates() override;
 
     void start_profile() override;
     void stop_profile() override;
@@ -308,7 +309,7 @@ private:
         /*!
          * \brief unpin the tensors in vec
          */
-        void unpin(const SmallVector<TensorInfo*>& vec);
+        void unpin(const SmallVector<TensorInfo*>& vec, WorkerState& state);
 
         /*!
          * \brief add the tensor to the candidate set
diff --git a/imperative/src/include/megbrain/imperative/interpreter.h b/imperative/src/include/megbrain/imperative/interpreter.h
index 88a28bf782e9e60d1b1bbb4c72fe8c3f5e33c26a..86fe691a454281b8a82af0625726f5af9120225a 100644
--- a/imperative/src/include/megbrain/imperative/interpreter.h
+++ b/imperative/src/include/megbrain/imperative/interpreter.h
@@ -57,6 +57,7 @@ struct Interpreter {
 
         virtual size_t get_option(std::string name) = 0;
         virtual void set_option(std::string name, size_t value) = 0;
+        virtual void clear_candidates() = 0;
 
         virtual void start_profile() = 0;
         virtual void stop_profile() = 0;