fix

CMakeLists.txt

@@ -4,7 +4,7 @@ option(USE_BANG "Support BANG MLU" OFF)
 option(USE_KUNLUN "Support KUNLUN XPU" OFF)
 option(USE_ASCEND "Support HUAWEI ASCEND" OFF)
 option(USE_INTELCPU "Support INTELCPU" OFF)
-option(USE_BACKTRACE "Print backtrace on exception and segmentation fault" OFF)
+option(USE_BACKTRACE "Print backtrace on exception and segmentation fault" ON)
 option(USE_PROTOBUF "Serialize and deserialize tensors" OFF)
 option(BUILD_NNET "Build nnet" OFF)
 option(BUILD_DIST "Build project for distributed running" OFF)

Makefile

@@ -39,7 +39,7 @@ endif
 
 build:
 	mkdir -p build/$(TYPE)
-	cd build/$(TYPE) && cmake $(CMAKE_OPT) ../.. && make -j
+	cd build/$(TYPE) && cmake $(CMAKE_OPT) ../.. && make -j8
 
 clean:
 	rm -rf build

README.md

@@ -33,13 +33,14 @@ There are several configurable CMake options, see the [CMakeLists.txt](/CMakeLis
 
 ## Roadmap
 
+- [RefactorGraph](https://github.com/InfiniTensor/RefactorGraph) is a newly designed AI framework that is set to replace the current main branch.
 - [EinNet](https://github.com/InfiniTensor/InfiniTensor/tree/NNET_e2e) is going to be merged into the main branch.
 - Integration of [PET](https://github.com/thu-pacman/PET), a tensor program optimizer supporting partially equivalent transformations.
 - Supported hardware
   - ✔ NVIDIA GPU
   - ✔ Cambricon MLU
+  - ✔ Kunlunxin XPU
   - ⬜ Ascend NPU
-  - ⬜ Kunlunxin XPU
 
 ## Contributor Guide
 
@@ -71,4 +72,4 @@ Please cite EinNet or PET in your publications if it helps your research:
   pages={37--54},
   year={2021}
 }
-```
+```

env.sh

@@ -47,7 +47,6 @@ export KUNLUN_HOME=/usr/local/xpu
 # ├── tools
 # ├── version
 # └── XTDK
-#export ASCEND_HOME=/usr/local/Ascend/ascend-toolkit/6.3
 export ASCEND_HOME=/usr/local/Ascend/ascend-toolkit/latest
 source /usr/local/Ascend/ascend-toolkit/set_env.sh
 source /usr/local/Ascend/toolbox/set_env.sh

src/core/graph_handler.cc

@@ -334,7 +334,7 @@ Tensor GraphHandlerObj::attentionKVCache(Tensor input_k_cache,
             std::move(input_k_cache), std::move(input_v_cache),
             std::move(input_q), std::move(input_k), std::move(input_v),
             std::move(position_id), output_matmul);
-        return {output_matmul};
+        return output_matmul;
     } else {
         return g
             ->addOp<AttentionKVCacheObj>(