diff --git a/open_diloco/run_training.sh b/open_diloco/run_training.sh
index 7198cfa..bc84335 100755
--- a/open_diloco/run_training.sh
+++ b/open_diloco/run_training.sh
@@ -6,6 +6,11 @@
 #
 ## the command above will use a total of 8 gpu and create 4 diloco workers each of them with two gpu training ddp/fsdp wise
 
+
+# you can either pass a fixed initial peer or set it to auto and the script will start a dht server for you
+## # ./run_training.sh 2 1 auto --per-device-train-batch-size 16 --batch-size 512 --local-steps 10  --total-steps 88000 --c4-tiny
+
+
 # Function to get CUDA devices based on the number of GPUs and index
 function get_cuda_devices() {
     local num_gpu=$1
@@ -31,6 +36,25 @@ NUM_GPU=$2
 INITIAL_PEER=$3  # Set INITIAL_PEER from the second argument
 shift 3      # Remove the first three arguments so $@ contains only additional Python arguments
 
+mkdir -p logs
+echo "Initial peer: $INITIAL_PEER"
+
+# Check if INITIAL_PEER is set to 'auto' and adjust accordingly
+if [ "$INITIAL_PEER" = "auto" ]; then
+    # start the dht server
+    echo "Starting DHT server"
+    hivemind-dht --host_maddr /ip4/0.0.0.0/tcp/12345 --identity_path fixed_key.pem  > logs/log_dht 2>&1 &
+    
+    INITIAL_PEER=""
+    # get the initial peer from the logs, loop until the peer is found
+    while [ -z "$INITIAL_PEER" ]; do
+        sleep 1
+        INITIAL_PEER=$(awk '/Running a DHT instance/ {print $NF}' logs/log_dht)
+
+    done
+fi
+echo "Initial peer: $INITIAL_PEER"
+
 # Ensure the logs directory exists
 mkdir -p logs