diff --git a/csrc/infinicore/src/ops/mha_kvcache/mha_kvcache_infiniops.cc b/csrc/infinicore/src/ops/mha_kvcache/mha_kvcache_infiniops.cc
index 6fdac33e3..801358b95 100644
--- a/csrc/infinicore/src/ops/mha_kvcache/mha_kvcache_infiniops.cc
+++ b/csrc/infinicore/src/ops/mha_kvcache/mha_kvcache_infiniops.cc
@@ -25,7 +25,7 @@ std::size_t implementation_index_for_device(
         return 0;
     }
     if (device_type == infini::ops::Device::Type::kMoore) {
-        return 8;
+        return 16;
     }
     return 16;
 }
diff --git a/csrc/infinicore/src/ops/multi_head_attention_varlen/mha_varlen_infiniops.cc b/csrc/infinicore/src/ops/multi_head_attention_varlen/mha_varlen_infiniops.cc
index 043ce070a..6c456423d 100644
--- a/csrc/infinicore/src/ops/multi_head_attention_varlen/mha_varlen_infiniops.cc
+++ b/csrc/infinicore/src/ops/multi_head_attention_varlen/mha_varlen_infiniops.cc
@@ -23,7 +23,7 @@ std::size_t implementation_index_for_device(
         return 0;
     }
     if (device_type == infini::ops::Device::Type::kMoore) {
-        return 8;
+        return 16;
     }
     return 16;
 }
diff --git a/csrc/pybind11/engine/engine.hpp b/csrc/pybind11/engine/engine.hpp
index c5e85577c..24419cec6 100644
--- a/csrc/pybind11/engine/engine.hpp
+++ b/csrc/pybind11/engine/engine.hpp
@@ -121,19 +121,15 @@ inline void bind_infer_engine(py::module &m) {
             }
             return state_dict_tp_all;
         })
-        .def("process_weights_after_loading", &InferEngine::process_weights_after_loading, "Process the weights after loading on all workers (e.g., for quantization)")
-        .def(
-            "forward", [](InferEngine &self, const InferEngine::Input &input) -> InferEngine::Output {
+        .def("process_weights_after_loading", &InferEngine::process_weights_after_loading, "Process the weights after loading on all workers (e.g., for quantization)", py::call_guard<py::gil_scoped_release>())
+        .def("forward", [](InferEngine &self, const InferEngine::Input &input) -> InferEngine::Output {
                 // IMPORTANT: Release the GIL before calling forward() to allow other Python threads
                 // to run concurrently during inference (which may block for a long time).
                 // Do NOT remove this — without it, the GIL is held throughout inference and will
                 // deadlock or stall any other Python thread (e.g., request handling, scheduling).
                 py::gil_scoped_release release;
-                return self.forward(input);
-            },
-            "Run inference on all ranks with arbitrary arguments")
-        .def(
-            "reset_cache", [](InferEngine &self, std::shared_ptr<cache::CacheConfig> cfg) { self.reset_cache(cfg ? cfg.get() : nullptr); }, py::arg("cache_config") = py::none())
+                return self.forward(input); }, "Run inference on all ranks with arbitrary arguments")
+        .def("reset_cache", [](InferEngine &self, std::shared_ptr<cache::CacheConfig> cfg) { self.reset_cache(cfg ? cfg.get() : nullptr); }, py::arg("cache_config") = py::none())
         .def("get_kv_cache", &InferEngine::get_kv_cache, "Get per-rank kv cache list")
         .def("get_cache_config", [](const InferEngine &self) -> std::shared_ptr<cache::CacheConfig> {
             auto cfg = self.get_cache_config();