[Auto-Parallel] Add ci for tensor_fusion and overlap in auto-dy (#10598)

* Update ci_case_auto.sh

* Update ci_case_auto.sh

* Update ci_case_auto.sh

* Update ci_case_auto.sh

* Update ci_case_auto.sh

* Update ci_case_auto.sh

* Update ci_case_auto.sh

Author: Xing-lil

scripts/distribute/ci_case_auto.sh

@@ -221,15 +221,17 @@ function llama_dygraph_auto_bs4_bf16_SD2() {
 
         export CUDA_DEVICE_MAX_CONNECTIONS=1
 
-        flags=("" "FLAGS_fuse_allreduce_in_opt" "FLAGS_fuse_reducescatter_in_opt")
+        flags=("" "FLAGS_fuse_allreduce_in_opt" "FLAGS_fuse_reducescatter_in_opt" "FLAGS_enable_tensor_fusion FLAGS_enable_sharding_overlap")
         for i in "${!flags[@]}"; do
             flag="${flags[$i]}"
 
             if [ -n "$flag" ]; then
-                export "$flag=true"
+                for f in $flag; do
+                    export "$f=true"
+                done
             fi
 
-            task_name="llama_dygraph_auto_bs4_bf16_SD2_$flag"
+            task_name="llama_dygraph_auto_bs4_bf16_SD2_$f"
             case_out_dir="output/$task_name"
             case_log_dir="output/$task_name""_log"
             rm -rf $case_out_dir
@@ -299,6 +301,8 @@ function llama_dygraph_auto_bs4_bf16_SD2() {
                 loss_base=9.23502579
             elif [ "$flag" = "FLAGS_fuse_reducescatter_in_opt" ]; then
                 loss_base=9.23504105
+            elif [ "$flag" = "FLAGS_enable_tensor_fusion FLAGS_enable_sharding_overlap" ]; then
+                loss_base=9.23504868
             else
                 loss_base=-1
             fi
@@ -308,7 +312,9 @@ function llama_dygraph_auto_bs4_bf16_SD2() {
             check_result $FUNCNAME ${loss_base} ${loss} ${ips_base} ${ips} ${mem_base} ${mem}
 
             if [ -n "$flag" ]; then
-                export "$flag=false"
+                for f in $flag; do
+                    export "$f=false"
+                done
             fi
         done
         echo "=========== $FUNCNAME run  end ==========="