diff --git a/scripts/margin.sh b/scripts/margin.sh
new file mode 100755
index 0000000..d5c9138
--- /dev/null
+++ b/scripts/margin.sh
@@ -0,0 +1,46 @@
+#!/bin/bash
+
+set -x
+
+HOST="test"
+
+export KMP_AFFINITY="respect,noreset,granularity=fine,balanced"
+export OMP_NUM_THREADS=$(grep ^cpu\\scores /proc/cpuinfo | uniq |  awk '{print $4}')
+echo "Cores configured $OMP_NUM_THREADS"
+export ONEDNN_VERBOSE=0
+
+for i in 1 2 3 4 5 6 7
+do
+	CNNS=(resnet50)
+	for COMPILER in ipex_onednn_graph
+	do
+		for DTYPE in float32 bfloat16
+		do
+		for BS in 0001 0032 
+		do
+			for name in "${CNNS[@]}"
+			do
+				echo "Benchmark $name with BS=$BS and DTYPE=$DTYPE"
+				numactl -m 0 --physcpubind=0-31 benchmark-run -b cnn -p "name='${name}',batch_size='$BS'" --dtype "${DTYPE}" --benchmark_desc "${name}_bs$BS" --host "${HOST}" -c "${COMPILER}" --skip_verification
+			done
+		done
+		done
+	done
+
+
+	LLMS=(gptj llama2-7b)
+	for COMPILER in ipex
+	do
+		for BS in 1 8
+		do
+			for DTYPE in bfloat16
+			do
+				for name in "${LLMS[@]}"
+				do
+					echo "Benchmark $name with DTYPE=$DTYPE"
+					numactl -m 0 --physcpubind=0-31 benchmark-run -b llm -p "name='${name}',batch_size=${BS}" --dtype "${DTYPE}" --benchmark_desc "${name}_bs$BS" --host "${HOST}" -c "${COMPILER}" --skip_verification
+				done
+			done
+		done
+	done
+done
diff --git a/scripts/margin2.sh b/scripts/margin2.sh
new file mode 100755
index 0000000..76c6aee
--- /dev/null
+++ b/scripts/margin2.sh
@@ -0,0 +1,76 @@
+#!/bin/bash
+
+set -x
+
+HOST="test"
+
+export KMP_AFFINITY="respect,noreset,granularity=fine,balanced"
+export OMP_NUM_THREADS=4
+echo "Cores configured $OMP_NUM_THREADS"
+export ONEDNN_VERBOSE=0
+
+for i in 1 2 3 4 5 6 7
+do
+	CNNS=(resnet50)
+	for COMPILER in ipex_onednn_graph
+	do
+		for DTYPE in float32 bfloat16
+		do
+		for BS in 0001
+		do
+			for name in "${CNNS[@]}"
+			do
+				echo "Benchmark $name with BS=$BS and DTYPE=$DTYPE"
+				export BENCH_COMMAND="benchmark-run -b cnn -p name='${name}',batch_size='$BS' --dtype ${DTYPE} --benchmark_desc ${name}_bs${BS}each4 --host ${HOST} -c ${COMPILER} --skip_verification"
+				numactl -m 0 --physcpubind=0-3 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=4-7 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=8-11 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=12-15 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=16-19 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=20-23 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=24-27 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=28-31 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=32-35 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=36-39 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=40-43 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=44-47 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=48-51 $BENCH_COMMAND &
+				numactl -m 0 --physcpubind=52-55 $BENCH_COMMAND &
+				wait $(jobs -p)
+			done
+		done
+		done
+	done
+
+
+	LLMS=(gptj llama2-7b)
+	for COMPILER in ipex
+	do
+		for BS in 1
+		do
+			for DTYPE in bfloat16
+			do
+				for name in "${LLMS[@]}"
+				do
+					echo "Benchmark $name with DTYPE=$DTYPE"
+					export BENCH_COMMAND="benchmark-run -b llm -p name='${name}',batch_size='$BS' --dtype ${DTYPE} --benchmark_desc ${name}_bs${BS}each4 --host ${HOST} -c ${COMPILER} --skip_verification"
+					numactl -m 0 --physcpubind=0-3 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=4-7 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=8-11 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=12-15 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=16-19 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=20-23 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=24-27 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=28-31 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=32-35 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=36-39 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=40-43 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=44-47 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=48-51 $BENCH_COMMAND &
+					numactl -m 0 --physcpubind=52-55 $BENCH_COMMAND &
+					wait $(jobs -p)
+					done
+			done
+		done
+	done
+done
diff --git a/scripts/margin_setup.sh b/scripts/margin_setup.sh
new file mode 100755
index 0000000..3482f57
--- /dev/null
+++ b/scripts/margin_setup.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+
+# We expect to have this repo present and this script run as
+# ./scripts/margin.sh
+
+# install miniconda
+wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
+curl -o Miniconda3-latest-Linux-x86_64.sh -L https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
+    sh Miniconda3-latest-Linux-x86_64.sh -u -b -p ./miniconda && \
+    rm -f Miniconda3-latest-Linux-x86_64.sh
+source ./miniconda/bin/activate
+
+# set up env
+conda create -y -n margin python=3.11
+conda activate margin
+# Install ipex & pytorch
+python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
+python -m pip install intel-extension-for-pytorch
+python -m pip install transformers==4.35.2
+python -m pip install oneccl_bind_pt --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/cpu/us/
+
+# Install benchmarks
+pip install -e .