So I have a script that extracts tables from excel files then makes a call to aws and sends the table to Claude 3.5 through aws bedrock, for classification together with a prompt. I recently moved this script to AWS and when I run the same script, with the same file from AWS I get a different classification for one specific table. * Same script * Same model * Same temperature * Same tokens * Same original file * Same prompt Gets me a different classification for 1 one specific table (there are like 10 tables in this file and all of them get classified correctly except for one 1 table in AWS but locally I get all the classifications correct) Now I understand that a LLMs nature is not deterministic etc etc, but when I run the file on aws 10 times I get the wrong classification all the 10 times, when I run it locally I get the right classification all 10 times. What is worst is that the value for the wrong classification IS THE SAME wrong value all 10 times. I need to understand what could possible be wrong here. Why locally I get the right classification but on AWS it always fails (on a specific table). Are the prompts read different on aws? Can it be the way the table its being read in AWS is differently from the way its being read locally? I am converting the tables to a df and then to a string representation but in order to somehow keep the structure I am doing this: table_str = df_to_process.to_markdown(index=False, tablefmt="pipe")