Skip to content

File Uploads

Upload datasets to Explorer via the browser

Apart from API-based uploads, you can also upload datasets to Explorer using the file upload UI. This is useful when you have a dataset in a ready-to-upload jsonl format and want to upload it to Explorer.

Browser-Based File Upload

To upload a dataset to Explorer, navigate to the home page and click on the New Dataset button. This will open the file upload dialog.

File Upload Dialog

Click on the Choose a File button and select the dataset file from your local machine. Once the file is selected, click on the Create button to start the upload process.

To ensure a successful upload, make sure your dataset is in the correct format as described next.

JSONL Format

The explorer supports two types of trace formats: raw event lists and annotated event lists.

Raw Event Lists

Raw event lists are jsonl files where each line is a JSON array of events. Each event is a dictionary with at least a role and content field according to the trace format described in this chapter.

[{ "role": "user", "content": "Hello, world!" }, { "role": "assistant", "content": "Hi!" }]
[{ "role": "user", "content": "How are you?" }, { "role": "assistant", "content": "I'm good, thanks!" }]
...

To include trace-level metadata in raw event lists, include a metadata JSON object as the first event in a line. Metadata is optional and can be used to store additional information about a specific trace.

[{"metadata": {"key": "value"}}, { "role": "user", "content": "Hello, world!" }, { "role": "assistant", "content": "Hi!" }]

Annotated Event Lists

Annotated event lists support the same format as raw event lists, but also include annotations.

For this, each line is a JSON object with the fields messages, annotations (optional) and metadata (optional, for trace-level metadata).

{"messages": [{ "role": "user", "content": "Hello, world!" }, { "role": "assistant", "content": "Hi!" }], "annotations": [{"content": "example annotation", "address": "messages.0.content:5-10"}], "metadata": {"key": "value"}}
{"messages": [{ "role": "user", "content": "How are you?" }, { "role": "assistant", "content": "I'm good, thanks!" }], "annotations": [], "metadata": {}}

Dataset-Level Metadata

In addition to trace-level metadata, you can also provide metadata at the dataset level.

To provide dataset metadata, include the following metadata JSON object as the first line of your uploaded .jsonl file. Like with trace-level metadata, dataset metadata is optional and can be of any structure to store additional information about the dataset.

{"metadata": {...}}

Dataset-level metadata is support for both raw and annotated event lists.