Model upload fails when version is omitted because LitLogger explicitly uses "latest"

Overview

Uploading a model with the dict-like API fails when version is omitted.

LitLogger converts version=None to "latest" and sends it as an explicit model version. However, "latest" is reserved by the backend and cannot be used as a version name.

Reproduction

import torch.nn as nn

import litlogger
from litlogger import Model

experiment = litlogger.init(
    name="model-upload-test",
    teamspace="<teamspace>",
)

model = nn.Linear(10, 1)
experiment["model"] = Model(model)

experiment.finalize()

Actual behavior

The upload fails with a 400 Bad Request:

{"code":3,"message":"version cannot be named latest","details":[]}

Expected behavior

When version is omitted, the Lightning SDK should automatically assign a version such as v1 or v2.

Explicitly specified versions should continue to work as before.

Environment

• LitLogger: main (812be06)
• lightning-sdk: 2026.8.14
• Python: 3.12
• OS: macOS

Additional context

Model currently converts an omitted version to "latest":

self.version = version or "latest"

The Lightning SDK supports version=None and automatically generates a vX version during upload.

A possible fix is to preserve None when no version is specified, then add a regression test covering model uploads without an explicit version.

Start at the Model class where an omitted version is converted to "latest", then trace the upload path to confirm how version=None is handled by the Lightning SDK. Add a regression test for uploading a model without an explicit version; done means omitted versions upload successfully while explicit versions still work.