Files

Upload and manage datasets for use in fine-tuning, evals, and batch inference.

Upload

To upload a new dataset file:

tg files upload [FILENAME]

Here’s a sample output:

$ tg files upload ./example.jsonl
Uploading file example.jsonl: 100%|██████████████████████████████| 5.18M/5.18M [00:01<00:00, 4.20MB/s]
Success!
file-d931200a-6b7f-476b-9ae2-8fddd5112308

The printed file-… identifier is the assigned file-id for this file object. Pass --json to get the full response body instead.

Parameters

Flag	Description
--no-check	Skip local file validation before uploading.
--purpose [string]	The purpose of the file. One of the following: fine-tune eval eval-sample eval-output eval-summary batch-generated batch-api

Check

You can precheck that a file is in the correct format.

tg files check [PATH]

Here’s a sample output:

$ tg files check ./local-file.jsonl
Validating file: 1 lines [00:00, 7476.48 lines/s]
✅ Checks passed

List

To list previously uploaded files:

tg files list

Retrieve

To retrieve the metadata of a previously uploaded file:

tg files retrieve [FILE_ID]

Here’s a sample output:

$ tg files retrieve file-d931200a-6b7f-476b-9ae2-8fddd5112308
Retrieved file details
Id:      file-d931200a-6b7f-476b-9ae2-8fddd5112308 
Name:    info_provided_validation_tokenized.parquet
Size:    303.4 KB
Type:    parquet
Purpose: fine-tune
Created: 03/16/2026, 03:58 PM

Retrieve content

To download a previously uploaded file:

Download to file

tg files retrieve-content <FILE-ID> \
  --output ./

Here’s a sample output:

$ tg files retrieve-content file-d931200a-6b7f-476b-9ae2-8fddd5112308 --output ./
File saved to ./example.jsonl

By default, the file is written to the current directory under its original filename. Pass --output PATH (or -o PATH) to write to a different directory or filename, or --stdout to stream the contents to stdout.

Delete

To delete a previously uploaded file:

tg files delete [FILE_ID]

Here’s a sample output:

$ tg files delete file-d931200a-6b7f-476b-9ae2-8fddd5112308
√ File deleted