startup.sh

#!/bin/bash

export HF_HOME="/data/.huggingface"

echo "PWD: $(pwd)"
echo $HF_TOKEN > .hf_token
echo "LS: $(ls -als)"

while true; do nvidia-smi; sleep 600; done &

python3 -c "import torch; \
    print(f\"is availeble = {torch.cuda.is_available()}\"); \
    print(f\"device count = {torch.cuda.device_count()}\"); \
    print(f\"current device = {torch.cuda.current_device()}\")"


COMPONENT="serve.controller" 
echo "Starting $COMPONENT"
nohup python3 -m serve.controller --host 0.0.0.0 --port 10000 2>&1 | tee "$COMPONENT.log" &
while [ `grep -c "Uvicorn running on" "$COMPONENT.log"` -eq '0' ];do
        sleep 1s;
        echo "wait $COMPONENT to be running"
done
echo "$COMPONENT is running"


# sleep 30

# echo "Starting prism-dinosiglip+13b"
# python3 -m interactive_demo --port 40000 --model_id prism-dinosiglip+13b &
# P4=$!



# echo "Starting prism-dinosiglip+7b"
# python3 -m interactive_demo --port 40000 --model_id prism-dinosiglip+7b &
COMPONENT="interactive_demo"
echo "Starting $COMPONENT"
nohup python3 -m interactive_demo --port 40000 --model_id prism-dinosiglip+7 | tee "$COMPONENT.log" &
while [ `grep -c "loaded prismatic prism-dinosiglip+7b" "$COMPONENT.log"` -eq '0' ];do
        sleep 1s;
        echo "wait $COMPONENT to be running"
done
echo "$COMPONENT is running"




# python3 -m interactive_demo --port 40002 --model_id prism-dinosiglip-controlled+7b &
# P6=$!

# python3 -m interactive_demo --port 40003 --model_id llava-v1.5-7b &
# P7=$!

sleep 30

echo "Starting serve.gradio_web_server"
python3 -m serve.gradio_web_server --controller http://127.0.0.1:10000 --model-list-mode reload & # --share
P_WEB=$!

sleep 30

ls -als $HF_HOME
tree --du -h $HF_HOME

wait $P_CON $P_WEB $P_P7B # $P5 $P6 $P7