[Question] How to load pre-trained models from HuggingFace? Missing config.json

[qwj980317]

Hi @shiyu-coder,

Thank you for open-sourcing Kronos! I'm trying to load pre-trained models from HuggingFace Hub but encountered an issue.

Problem

When calling:

from model import Kronos, KronosTokenizer

tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
model = Kronos.from_pretrained("NeoQuasar/Kronos-small")
Getting error:

TypeError: KronosTokenizer.__init__() missing 16 required positional arguments
It seems HuggingFace repos are missing
config.json
files.

[saiganesh47]

Hi @shiyu-coder,

Thank you for open-sourcing Kronos! I'm trying to load pre-trained models from HuggingFace Hub but encountered an issue.

Problem

When calling:

from model import Kronos, KronosTokenizer

tokenizer = KronosTokenizer.from_pretrained("NeoQuasar/Kronos-Tokenizer-base")
model = Kronos.from_pretrained("NeoQuasar/Kronos-small")
Getting error:

TypeError: KronosTokenizer.init() missing 16 required positional arguments
It seems HuggingFace repos are missing
config.json
files.

Thanks for the report !
The error occurs because the Kronos models on HuggingFace are distributed with custom model/tokenizer code, rather than the standard Transformers config layout. Since the repos do not include a config.json, calling from_pretrained() without enabling remote code causes Transformers to fall back to the default loader, which leads to the init missing positional arguments error.
This ensures the custom loader inside the repository is used and the model initializes correctly.

To load Kronos correctly, please enable remote code execution and pin the revisions:
from model import Kronos, KronosTokenizer
tokenizer = KronosTokenizer.from_pretrained(
"NeoQuasar/Kronos-Tokenizer-base",
revision="0e0117387f39004a9016484a186a908917e22426",
trust_remote_code=True
)
model = Kronos.from_pretrained(
"NeoQuasar/Kronos-small",
revision="901c26c1332695a2a8f243eb2f37243a37bea320",
trust_remote_code=True
)