Data
ShareGPT format
What is sharegpt format?
ShareGPT is a JSON conversation format storing a list of turns, each with a sender and a value. It became a de facto interchange standard for fine-tuning datasets and is typically converted into a model’s native chat format before training.
Its usefulness is as an interchange layer. Datasets are shared in it, and each pipeline converts it into whatever its target model expects, which keeps the dataset independent of the model choice.
The conversion is where mistakes happen. A conversion that drops the system message, or maps the roles wrongly, produces a dataset that trains without error and yields a model that behaves oddly for reasons nothing in the logs explains.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.