export TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
export FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE
vllm serve openvla/openvla-7b \
    --trust-remote-code --dtype bfloat16 --max-model-len 512 \
    --enforce-eager --mm-processor-cache-gb 0

  # 2. In another terminal, send a chat completion request with an image:
  curl -s http://localhost:8000/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
      "model": "openvla/openvla-7b",
      "messages": [{
        "role": "user",
        "content": [
          {"type": "image_url", "image_url": {"url": "data:image/png;base64,'"$(python3 -c "
  import base64, io; from PIL import Image
  img = Image.open('/scratch/mgehre/openvla/sample_data/bridge_sample_01.jpg').convert('RGB').resize((224,224), Image.BILINEAR)
  buf = io.BytesIO(); img.save(buf, format='PNG'); print(base64.b64encode(buf.getvalue()).decode())
  ")"'"}},
          {"type": "text", "text": "In: What action should the robot take to put the blue cube on the right side of the table on top of the rectangular block?\nOut: "}
        ]
      }],
      "max_tokens": 7,
      "temperature": 0.0
    }' | python3 -m json.tool