add the graphcapture

include/flexflow/ops/fused.h

@@ -101,6 +101,7 @@ class FusedOp : public Op {
   Op *operators[MAX_NUM_FUSED_OPERATORS];
   FusedOpMeta fused_meta[MAX_NUM_WORKERS];
   int numOperators;
+  bool graphCaptured;
 };
 
 }; // namespace FlexFlow

profile.sh

@@ -0,0 +1,2 @@
+nsys profile -w true -t cuda,nvtx,osrt,cudnn,cublas -s cpu -o TP4B4newLegion-lambda -f true -x true \
+    ./inference/incr_decoding/incr_decoding --ll:util 8 -ll:cpu 4 -ll:fsize 20000 -ll:zsize 30000 -llm-model meta-llama/Llama-2-7b-hf -cache-folder /home/xiaoxias/weights --fusion -ll:gpu 4 -prompt ../inference/prompt/chatgpt.json -tensor-parallelism-degree 4

run.sh

@@ -0,0 +1,3 @@
+
+
+python3 ./inference/utils/download_hf_model.py meta-llama/Llama-2-7b-hf --cache-folder  /home/xiaoxias/weights

src/ops/fused.cc

@@ -53,6 +53,7 @@ FusedOp::FusedOp(FFModel &model, Op *op)
          0 /*weights*/,
          0 /*outputs*/) {
   numInputs = 0;
+  graphCaptured = false;
   for (int i = 0; i < op->numInputs; i++) {
     bool found = false;
     // we also need to check region duplicate for the first op in a fused op