Phoenix supports uploading datasets not only as CSV files but also as JSON objects. You can upload a dataset using a pandas DataFrame, a CSV file, or a series of dictionary objects, which can be JSON-like structures. This flexibility allows you to choose the format that best suits your needs. For instance, you can upload a dataset using dictionary objects by specifying inputs, outputs, and metadata as lists of dictionaries, which is similar to JSON format (<https://docs.arize.com/phoenix/api/client#upload_dataset|source>).
If you are interested in JSONL format, there is a feature to download datasets in JSONL format, which suggests that JSON support is considered in the platform's design (<https://github.com/Arize-ai/phoenix/pull/3495|source>). Would you like more details on how to upload datasets using JSON-like structures?
Sources:
- <https://github.com/Arize-ai/phoenix/pull/3495|GitHub Pull Request: JSONL Download>
- <https://docs.arize.com/phoenix/api/client#upload_dataset|https://docs.arize.com/phoenix/api/client#upload_dataset>
