File Transfer, Files API limits, and batch statistics
File Transfer in the dashboard, rate limits on the Files API, error counts and score distributions in batch statistics, model status, and clearer prediction errors.
File Transfer in the dashboard
The dashboard has a File Transfer section, for organizations where Avra enables it. Under Uploads, upload files, browse them by folder, and delete the ones you no longer need. Exports holds the files Avra prepares for you. Download from either through links that work for 15 minutes. Separate permissions decide who can download, upload, and delete. See Roles & permissions.
Limits on the Files API
The files endpoints, /v1/api/files/*, now allow 1,500 requests per minute per organization and 300 per minute per caller, a user or an API key. Over the limit, they return 429 Too Many Requests with a Retry-After header that says how many seconds to wait. One upload request now takes at most 1,000 files, so request URLs for more in several calls. See Limits and quotas.
Error counts and score distributions in batch statistics
A batch’s statistics now list the errors behind its failed records, each message with how many records it affected. For a model that returns a score, they also describe the scores’ distribution: the minimum, maximum, mean, standard deviation, median, and a histogram. Read them with GET /v2/api/batches/{id}. See Batch inference.
Model and version status
Models and their versions now report a status in API responses, such as available. Creating a batch for an archived model or version is refused.
Clearer prediction errors
When a prediction fails on its inputs, the error now names the problem and the fields involved, such as Input validation failed: missing_inputs=[cnpj]. See Live predictions.