Seems like this thread has died, but for posterity, databricks provides the following code for installing poppler on a cluster. The code is sourced from the dbdemos accelerators, specifically the "LLM Chatbot With Retrieval Augmented Generation (RAG) and Llama 2 70B" (https://notebooks.databricks.com/demos/llm-rag-chatbot/index.html#) demo. In the 01-PDF-Advanced-Data-Preparation notebook there's code to remote execute the 00-init-advanced notebook and in that notebook, you'll find the code below. 

#install poppler on the cluster (should be done by init scripts)
def install_ocr_on_nodes():
    """
    install poppler on the cluster (should be done by init scripts)
    """
    # from pyspark.sql import SparkSession
    import subprocess
    num_workers = max(1,int(spark.conf.get("spark.databricks.clusterUsageTags.clusterWorkers")))
    command = "sudo rm -rf /var/cache/apt/archives/* /var/lib/apt/lists/* && sudo apt-get clean && sudo apt-get update && sudo apt-get install poppler-utils tesseract-ocr -y" 
    def run_subprocess(command):
        try:
            output = subprocess.check_output(command, stderr=subprocess.STDOUT, shell=True)
            return output.decode()
        except subprocess.CalledProcessError as e:
            raise Exception("An error occurred installing OCR libs:"+ e.output.decode())
    #install on the driver
    run_subprocess(command)
    def run_command(iterator):
        for x in iterator:
            yield run_subprocess(command)
    # spark = SparkSession.builder.getOrCreate()
    data = spark.sparkContext.parallelize(range(num_workers), num_workers) 
    # Use mapPartitions to run command in each partition (worker)
    output = data.mapPartitions(run_command)
    try:
        output.collect();
        print("OCR libraries installed")
    except Exception as e:
        print(f"Couldn't install on all node: {e}")
        raise e