Validate JSONL Training File

[Lars985]

Hi,

I have written a php connector to the api. Training works well, but not with all files.
After creating the finetuned job, I get this when I request the status of the current training: “The job failed due to an invalid training file”
I use gpt-3.5-turbo-0613. Linefeeds are correct, double quotes ecaped, etc.
The jsonl looks valid, according to php storm’s json validation function.

Are there any other tools, or scripts, to validate my jsonl? The message that I get from the api is not very clear, and doesn’t mention the line, that causes the problems.

Thanks,
lars

[Foxalabs]

Might be worth posting at least a sample of the training set, might be a common issue on every line.

[N2U]

Agreed,

If you need a general approach I’ll suggest this example from the cookbook:

Data preparation and analysis for chat model fine-tuning | OpenAI Cookbook

[Lars985]

It’s propably this string: Und so funktioniert\'s, especially that part: \'s.
how to escape that correctly?

[_j]

The file should have all strings in double-quotes, thus no need to escape an apostrophe.

The data is JSON, not python autoquotes.

This code should crash nicely on bad json:

Fine-Tuning of my personal blog

tiktoken, which I mentioned above, is a python token counting library that you can install simply with pip. pip install tiktoken Because I’m a fun guy, I just sat down and coded a per-line token counter for jsonl training files. import json import tiktoken class Tokenizer: def __init__(self, model_name): self.tokenizer = tiktoken.get_encoding(model_name) def count(self, text): encoded_text = self.tokenizer.encode(text) return len(encoded_text) def messag…