I am trying to tune a chat-bison model on Vertex AI and I running into an error.

Google Cloud Build with AI Custom ML & MLOps
1

message: "[2023-11-29 17:56:05,372] [ERROR]: Failed with error: [Expecting property name enclosed in double quotes: line 2 column 1 (char 2)]. "

This is the error I am getting. I am trying to tune the chat model with a jsonl file. It’s structure is like this: I can’t provide the original dataset but it looks like this and I can assure that even in the original dataset I have taken measures to ensure that the property name are enclosed in double quotes (") and not single quotes. In fact there are no single quotes in the dataset. I am not sure at all why I keep getting this error.

{
“context”: “You are a pirate dog named Captain Barktholomew.”,
“messages”: [
{
“author”: “user”,
“content”: “Hi”
},
{
“author”: “assistant”,
“content”: “Argh! What brings ye to my ship?”
},
{
“author”: “user”,
“content”: “What’s your name?”
},
{
“author”: “assistant”,
“content”: “I be Captain Barktholomew, the most feared pirate dog of the seven seas.”
}
]
}

I even tried to run a tuning job on the example dataset as well but it gave me the same error regardless.

2

The error message you’re encountering suggests there might be an issue with the JSON formatting, even though you’ve ensured that property names are enclosed in double quotes. The error often occurs when there’s a hidden character or a specific formatting issue that isn’t immediately visible.

Here are a few steps you might consider to troubleshoot this:

  1. Check for Hidden Characters: Sometimes, there might be invisible characters causing issues. You can try copying the content into a text editor or IDE that shows hidden characters, enabling you to spot any irregularities.

  2. Inspect JSON Encoding: Ensure that the file encoding is correct. Sometimes, encoding issues can lead to parsing errors.

  3. Try with a Simplified JSON File: Create a simpler version of your dataset with just a couple of records and see if the problem persists. This might help isolate the issue.

  4. Review Error Logs Thoroughly: The error message might sometimes be misleading. Reviewing logs or debugging tools closely might reveal more specific information about the problem.

If the problem persists despite these attempts, there might be a more intricate problem at play. Sometimes, libraries or tools used for processing JSON data might have specific requirements or limitations that could cause issues.